13th International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions, Proceedings

ISHPMIE 2020 Braunschw eig, Germany 13 th International Symposium on Hazards, Pre v ention and Mitigation of Industrial Explosions Proceedings Physikalisch -Technische Bundesanstalt Otto von Guericke Universität Magdeburg 13 th Symposium Internatio nal Symposium on Hazar ds, Prevention and Miti gation o f Industri al Explosions Proceedings DOI: 10.7795/81 0.20200724 Citation Templ ate: Author 1 , Aut hor 2 , ... : " T itle of article ", pp. n1 - n2 . In: "Proceedings of the 13th Internatio nal Symposium on Hazards, Prev ention and Mitigation of Industrial Explos ions (ISHPMIE 2020)", Braunschweig, G ermany, 2020. DOI : 1 0.7795/810.2020072 4 Herausgeb er: Physikalisch- Technisch e Bundesa nstalt Bundesallee 10 0 38116 Braunsc hweig, G ermany Dr. Michael Bey er Dr. Arnas Luca ssen 3.7 | Fundamen tals of Exp losion Protection phone: +49 531 592- 3700 e-mail: m ichael.beye [email protected] ishpmie2020. ptb.de ishpmie2020 @ptb.de Published unde r CC- BY - ND 4.0 DOI : 10.7795/810.2020072 4 Contents Message from the President of the Host Organization (Joachim Ulrich) 8 Message from the International Organizing Committee (Try gv e Skjold) 9 Message from the Local Organizing Committee (Michael Beyer) 10 Message from the Program Committee (Holger Großhans) 11 List S ymposiu m Committees 12 Session 01: Advances in explosion protection: Strategies, meas ures, and protective equipm ent (1) (Trygve Skjold) Laboratory Development and Pilot-s cale Deploy ment of a Two-part Foamed Rock Dust Connor B. Brown, Inoka E. Perera, Marcia L. Harris, Linda L. Chasko, James D. Addis & Sc ott Klima (#921) 13 Assessment of Release Mitigation of Water -Reactive Ch emicals by Absorbents Ting-Jia Kao, Thanh-Trung Nguyen, Yu -Jhen Lin, Zh i-Xuan Lin, Hsiao -Yu n Tsai, Jenq-Renn Chen (#963) 28 Modeling of Explosion Dynamics in Vessel-Pipe Systems to Evaluate Performance L imitations of Explosion Isolation Systems Lorenz R. Boeck, C. Regis Bauwens & Sergey B. Dorofeev ( #969) 37 Session 02: Advances in explosion protection: Strategies, meas ures, and protective equipm ent (2) (Detlev Markus) CEQAT-DGHS interlaboratory tests for chemical safety: Validation of laboratory test methods by determining the measurement uncertainty and probabilit y of incorrect classification including so - called “Shark profiles” Peter Lüth, Steffen Uhlig, Kirstin Frost, Marcus Malow, Heike M ichael -Schulz, Martin Schmidt & Sabine Zakel (#994) 50 CEQAT-DGHS Interlaboratory tests for chemical safety – A new gravimetric procedure for the gas flow measuremen t for flammable and toxic gases Peter Lüth, Marcus Malow (#1002) 62 On the strength of knowledge in risk assessments for hydrogen systems Trygve Skjold (#1018) 72 Review of the HySEA project Trygve Skjold, Melodia Lucas, Helene Hisken, Gordon Atan ga, Sunil Lak shmipathy, Laurence Bernard, Matthijs van Wingerden, Kees van Wingerden, Jenn ifer X. Wen, Vendra Chandra Madhav Rao, Anubhav Sinha, Marco Carcassi, Martino Schiavetti, Tommaso Pini, Jef Snoeys, Arve Grøns und Hanssen, Changjian Wang, Simon Jallais, El ena V yazmina, Derek Miller & Carl Regis Bauwens (#10 22) 85 Session 03: Detonations and DDT (Lore nz Boeck) Influence of Geometry on Flame Acceleration and DDT in H 2 - CO -Air Mix tures in a Pa rtially Obstructed Channel Daniel Heilbronn, Christoph Barfuss & Thomas Sattelmayer (#918) 107 Impact of Local Flame Quenching on the Flame Accelera tion in H 2 - CO -Air Mixtures in Obstructed Channels Christoph Barfuss, Daniel Heilbronn & Thoma s Sattelmayer (#919) 119 Observations of DDT in narrow channels Yves Ballossier, Josué Melguizo-Gavilanes & Florent Virot (#1025) 129 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 1 Session 04: Electro-chemical energy carriers (Frank Lienesc h) Explosibility Properties of Gases from Lithium-Ion Energy Storage Battery Thermal Runaways Adam Barowy, Pravin Gandhi, Robert Zalosh & Alexandra Klieger (#937) 139 Lithium- I on Energy Storage Battery Explosion Incidents Robert Zalosh, Pr avinray Gandhi & Adam Barowy (#938) 154 Li -ion batteries: characterization of the thermal runaway reactions using a DSC Paola Russo, Maria Luisa Mele (#1031) 166 Session 05: Explosion modelli ng and simulation (1) (Ulrich Krause) Vortex Dynamics and Fractal Structures in Reactive Richtmyer -Meshkov Instabil ity Maximilian Bambauer, Josef Hasslberger, Nilanjan Chakrabo rty & Markus Klein (#898) 175 Physical and Mathematical Modelling of Intera ction of Detonation Waves with Inert Gas Plugs Dmitry Tropin & Igor Bedarev (#908) 189 Boundary Conditions and Grid Dependency in CFD Simulation of Atmospheric Flow Henry Plischka, Johann Turnow & Nikolai Kornev (#1007) 201 Session 06: Explosion modelli ng and simulation (2) (Josué Melguizo-Gavilanes) Numerical modelling of the effects of vessel length- to -diameter ratio (L=D) on pres sure piling Damilare Ogungbemide, Martin P. Clouthier, Chris Clone y, Robert G. Zalosh, Robert C. Ripley, & Paul R. Amyotte (#943) 213 A computational framework for electrification of turbulent liquid flows Mathieu Calero, Holger Grosshans & Miltiadis V. Papalexan dris (#1027) 226 CFD-based simulation of flammable gas dispersion in a complex geometry Miroslav Mynarz, Aleš Tulach, Petr Lepík & Milada Koubkov ( #1030) 238 Session 07: Explosion modelli ng and simulation (3) (Holger Großhans) A Eulerian model for dust deflagrations, including inner particle effects Christoph Spijker, & Harald Raupenstrauch (#902) 253 CFD Simulation of an Unconfined Vapor Cloud Explosion through obstacles using OpenFoam® Cléante Langrée, Erwin Franquet, Julien Reveillon, Guillaume Lecocq & François-Xavier Demoulin (#939) 262 Understanding the role of thermal radiation in dust flame propagation Christophe Proust & Rim Ben M ouss a (#959) 270 Session 08: Explosion modelli ng and simulation (4) (Paola Russo) Impact of photovoltaic power plants on far -field effects of UVCEs Guillaume Lecocq, Antoine Dutertre & Emmanuel Leprette (#912) 284 Influence of Thermal Radiation on Layered Dust Explosi ons Swagnik Guhathakurta & Ryan W. Houim (#1012) 295 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 2 Determination of reaction mechanisms for the gasification and explosion of organic powders Matteo Pietraccini, Eloise Delon, Audrey Santandrea, Stép hanie Pacault, Pierre -Alexandre Glaude, Anthony Dufour & Olivier Dufaud (#1019) 306 Session 09: Explosion prevention (1) (Han nes Kern) Effect of multiple vent characteristic parameters on ext ernal explosion induced by in door premixed methane-air explosion Lei Pang, Qianran Hu, Yang Hu, Pengfei Lv & Kai Yang ( #933) 317 Keeping the Overview on Surface Resistivity Measurements Aaron D. Ratschow, Sigrun Stein & Hans-Jürgen Gross (#945) 337 A new technique to produce well controlled electrical sparks. Application to MIE mea surements Christophe Proust (#960) 348 25 years of ATEX directive: the real role of each Stakehol der Michał Górny & Xavier Lefebvre (#996) 365 Session 10: Explosion prevention (2) (Pa ul Am yotte) Inves ti gation on limiting oxygen concentration of combu s tible gas for safety control during air injection process Pengliang Li, Zhenyi Liu, Mingzhi Li, Yao Zhao, Xuan Li (#93 2) 377 Measurement of the deposit formation during pneumatic transport of polydispers e PMMA powder Nuki Susanti, Holger Grosshans (#896) 389 Suppression effect of inert gas on aluminum dust explosion Shulin Zhang, Mingshu Bi, Haipeng Jiang, Wei Gao (#983) 398 Session 11: Explosion properties of substances and mixtures (1) (Jan Berghmans) Homogeneity of methane-a ir mixture in devices for the determination of explosion limits Eliška Fišerová, Jan Karl (#910) 411 Enhanced Friction and Shock Sensitivities of Hexachloro disilane Hydrolyzed Deposit Mixed with KOH Thanh - Trung Nguyen, Yu-Jhen Lin, Zhi-Xuan Lin, Hui-Chu Tai, Chun-Yi Lee, Hsiao-Yun Tsai, Jenq -Renn Chen & Eugene Y. Ngai (#916) 418 Preliminary Study of Lignocellulosic biomass ignition pr operties estimation from Thermogravimetric Analysis Blanca Castells, Isabel Ame z, Ljiljana Medic, Nieves Fernandez -Anez & Javier Garcia-Torrent (#941) 434 Burning and explosion behaviour of ethanol/water -sucrose mixtures Maria Mitu, Thomas Stolz, Elisabeth Brandes & Sabine Zakel (#1005) 446 Session 12: Explosion properties of substances and mixtures (2) (Jenq-Renn Chen) Flammability characteristics of methane enriched with H 2 using CO 2 in the Spark Test Apparatus Isabel Amez, Blanca Castells, Ljiljana Medic & Javier Garcia -Torrent (#942) 461 Do nanostructured materials influence the igni t ion behavior of combustible dust? Susanne Hacke (#955) 478 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 3 Nex-Hys – Minimum Ignition Temperature of Hybrid Mixt ures Dieter Gabel, Paul Geoerg, Fabian Franken & Ulrich Krause (#966) 484 A Thermal Model for the Minimum Ignition Energy of Dus ts Tengfei Chen, Jan Berghmans, Jan Degrève, Filip Verplaets en, Jo Van Caneghem & Maarten Vanierschot (#988) 494 Session 13: Explosion testing (Robert Zalosh) Safety Related Properties of Tetrafluoroethylene Researc h on the Explosive Decomposition on an Industrial Scale Christian Liebner, Volkmar Schröder & Martyn J. Shenton (# 950) 506 Influence of thermal shock of piezoelectric pressure sensors on the measure ment of explosion pressures Tim Krause, Mirko Meier & Jens Brunzendorf (#953) 514 Modifications of the Standard 1m3 Vessel in S earch of Adequate Values of the Maxi mum Rate of Pressure Rise Wojciech Adamus, Adrian Toman (#979) 529 Turbulence in real flammable gas releases Didier Jamois, Christophe Proust, Jérôme Hébrard, Emm anuel Lepre tte, Helene Hisken, L orenzo Mauri, Gordon Atanga, Melodia Lucas, Kees van Wingerden, Trygve Skjold, Pierre Quillatre, Antoi ne Dutertre, Thibault Marteau, Andrzej Pekalski, Lorraine Jenney, Dan A llason & Mike Johnson (#1010) 535 Session 14: Explosions of sprays and vapors (Olivier Dufau d) Charge-separating processes by spraying water under high pressure Florian Baumann, Matthias Himstedt, Dieter Möckel & Martin Thedens (#905) 550 Presentation of the experimental JIP SPARCLING: Inside and beyond a pressurised LNG release Lauris Joubert, Guillaume Leroy, Steven Betteridge, Elena Vyazmina, Laurence Bernard, R omain Jambut & Jérôme Frindel (#934) 562 Hydrocarbon aerosol explosion: towards hazardous area classification Stephanie El-Zahlanieh, Augustin C harvet, Alexis Vignes, Benoit Tribouilloy & Olivier Dufaud (#1021) 572 Session 15: Flame propagati on and ac celeration (1) (Wookyung Kim) On an Assessment of Dust Explosion Dynamics in th e Standard 20 -l Sphere Zdzisław Dyduch (#981) 584 Ignition Temperatures and Flame Velocities of Metallic Nanomaterials Arne Krietsch, Monica Reyes Rodriguez, Andor Kristen, Dan iel Kadoke, Zaheer Abbas & Ulrich Krause (#949) 591 A flame propagation model for nanopowders Audrey Santandrea, David Torrado, Matteo Pietra ccini, Alexis Vignes, Laurent Perrin & Oli vier Dufaud (#957) 606 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 4 Session 16: Flame propagation and acceleration (2) (Jennifer Wen) Construction of a 4 m long test rig for experimental inve stigations on flame propagation in combustible dust/air mixtures Katja Hüttenbrenner, Hannes Kern, Florian Toth, Julian Glechner & Harald Raupenstrauch (#956) 618 Self-similar propagation of expanding spherical flames in lean hydrogen -air mixtures Wookyung Kim, Takumi Namba, Tomoyuki Johzaki & Takuma Endo (#993) 628 Assessing the influence of real releases on explosions: selected results from large -scale experiments Helene Hisken, Lorenzo Mauri, Gordon Atanga, Melodia Lucas, Kees van Wingerden, Trygve Skjold, Pierre Quillatre, Antoine Dutertre, Thibault Marteau, Andrzej Pek alski, Lorraine Jenney, Dan Allason, Mike Johnson, Emmanuel Leprette, Didier Jamois, Jérôme Hébrar d & Chris to phe Proust (#10 15) 637 Session 17: Gas, dust, and hybrid mixture explosions (1) (Dieter Gabel) Investigations on the effect of particle size on dust disp ersion in MIKE 3 apparatus Yangyue Pan, Christoph Spijker, Hannes Kern & Harald Raupenstrauch (#906) 654 Numerical investigation of overdriving in th e 20 -L Siwek chamber Martin P. Clouthier, Damilare Ogungbemide, Chris Cloney, Robert G. Zalosh, Robert C. Ripley & Paul R. Amyotte (#909) 663 Prenormative Study on the Safety Characteristics of Ex plosion Protection for Hybrid Mixtures of Dusts and Vapours Vanessa Heilmann, Adeyemi Taiwo, Werner Hirsch, Holger Grosshans, Sabine Zakel & Ulrich Krause (#920) 677 The Explosion of Non-nano Iron Dust Suspension in the 20-l Spherical Bomb Enrico Danzi, Gianmaria Pio, Luca Marmo & Ernesto Salzano (#947) 688 Session 18: Gas, dust, and hybrid mixture explosions (2) (Jérôme Taveau) An investigation of micron and nano scale of aluminium dust explosion Po -Jul Chang, Toshio Mogi & Ritsu Dobashi (#927) 697 Minimum Ignition Energy of Coal Dust Clouds in air an d O 2 /CO 2 Atmos pheres with S mall Amount of CH 4 /H 2 Dejian Wu, Arne Krietsch, Frederik Norman & Martin Schmid t (#930) 705 Session 19: Gas, dust, and hybrid mixture explosions (3) (W ei Gao) Modelling of detonation flows in inhomogeneous gas par ticle suspensions within the framework of the reduced kinetics Tatyana Khmel & Sergei Lavruk (#971) 716 Numerical investigation of propagation of heterogeneou s detonation in nanodispersed aluminum particle suspensions in expanding channels Sergey Lavruk, Tatyana Khmel (#974) 728 A Simple Dust Combustion Model for Characterizing Reactivity in Large -Scale Exper iments C. Regis L. Bauwens, Lorenz R. Boeck & Sergey B. Dorofee v (#1016) 739 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 5 Quasi-static dispersion of dusts for the determination of lower explosion limits of h ybrid mixtures Zaheer Abbas, Olivier Dufaud, Dieter Gabel, Arne Krietsch & Ulrich Krause (#1023) 750 Session 20: Gas, dust, and hybrid mixtu re explosions (4) (Ch ristophe Proust) Testing of dust clouds for the electrostatic -spark ignition hazard in industry. Need for a modified approach? Rolf K. Eckhoff (#998) 765 Self-ignition tendency of solid fuels: a gas emissions ap proach for early d etection) Nieves Fernandez -A nez, Blanca Castells-Somoza, Isabel Amez-Arenillas & Javier Garcia - Torrent (#1000) 779 Behaviour of smouldering fires during periodic refilling of wood pellets into silos Nieves Fernandez -A nez, Anita Katharina Meyer, Javier Elio M edina, Gisle Kleppe, Bjarne Christian Hagen & Vidar Frette (#1001) 791 Effect of particle size distribution, drying and milling t echnique on explosibility behaviour of olive pomace waste Matteo Pietraccini, Enrico Danzi, Luca Marmo, Albert Addo & Paul Amyotte (#1024) 800 Session 21: Hydrogen safety ( Daniel Banuti) Pressure peaking phenomena: Large -scale experiments of ignited a nd unignited hydrogen releases Agnieszka Lach, André Vagner Gaathaug (#892) 813 Investigation of the correlation bet ween the electrical power and flame front of a disc harge caused by a slow contact opening in a H 2 /air mixture Carsten Uber, Bogdan Barbu, Michael Hilbert, Frank Berger, Frank Lienesch (#913) 827 Spectroscopic investigations of the correlation between the discharge caused by a sl ow contact opening and the flame front in a H 2 /air mixture Carsten Uber, Steffen Franke, Jens Brunzendorf, Mi chael Hilbert, Dirk Uhrlandt, Frank Lienesch (#914) 836 Session 22: Ignition phenomena (1) (Stefan Essmann) New aspects of the ignition of burnable gas mixtures by low -energy electrical discharges Stefan Essmann, Johann-Robert Kummer, Holger Grosshans, Detlev Markus & Ulrich Maas (#900) 844 Determination of the ignition temperature of flammable li quids under different initial conditions Libor Ševčík & Ondřej Suchý (#911) 858 Comparative study on standardized ignition sources used for explosion testing Stefan Spitzer, Enis Askar, Arne Krietsch & Volkmar S chröder (#925) 864 Spontaneous combustion behaviour of solids: Validation of extrapolation of l aboratory tests by means of semi technical tests up to 1 m 3 and advanced thermoanalytical methods Martin Schmidt, René Erdt, Markus Gödde & Steffen Salg (# 936) 876 Low Temperature Autoignition of Jet A and Surrogate Jet Fuels Conor Martin & Joseph Shepherd (#973) 891 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 6 13th International Symposium on Hazards, Prevention, and Mitigat ion of Industrial Explosions Braunschweig, GERMANY – July 27-31, 2020 Laboratory D evelopment and P ilot -sc ale Deploymen t of a T wo -part F oamed Rock Dust Connor B. Brown a , Inoka E. Pere ra a , Marcia L. Harris a , Linda L. Chasko a , James D. Addis a & Scott Klima a a National Ins titute for Oc cupat ional Safe ty and Health (NI OSH), Pittsburgh, PA , U SA E-mail: CBBrow [email protected] ov Abstract U.S. Code of Federa l Regulations 30 CFR 75.402 and 75.403 require 80% total incombustibl e conten t to be maintained within 40 fe et of the coal mine face via the liberal application of rock dust . Unfortunately, this application of rock dust limits miners’ visi bility down wind and can increase the miners’ exposures to a re spirable nuisance dust. Wet rock dust applied as a slurry is, at times, used to negate these negative effects. Although this aids in meeting the tot al incombustible li m its, the slurry forms a hard cake when dried and no long e r effec tively disperses as needed to suppre ss a coal dust explosion. As a result, a dry rock dust must be re applied to maintain a dis persible layer. Therefore, researchers fr om the National I nstitute for Occupational Safety and Health (NI OSH) h ave been working towa rds finding and testing a foa med rock dust for mulation th at can be applied wet on mine surfaces and re main dispersible once dried which minimizes the likelihood of mine disasters, including mine explosions. The initial tests were aimed at discerni ng dis persion characteristics of three differe nt foamed rock dusts via the NIOSH-developed dispersion chamber and led to identification of a two-part foam with adequate dispersion characteristics. The cu rre nt study was conducted to assess the robust ness of the two-part foamed rock dust. Throu gh a series of laboratory- scale exp eriments using the dispersibility cha mber, the effects o f testing c onditions and produ ct formulations on the foam’s dispersibili ty was determine d . Some of the tested variables include: exposing the foam to high humidity, varying the c omponent levels of the foamed rock dust, altering the roc k dust size distribution, and varying the rock dust types. F urther pilot-scale tests exa mined the atmospheric concentrati ons of dust via persona l dust monitors downwind of foamed rock dust production and application. Additionally, product consistency was recorded during pil ot-scale testing at ke y points in the formulation and a pplication. The results of these experiments will be discussed in this paper. Keywords: dust dispersibility, foamed rock dust, coal mining, explosion preve ntion 1. Introduction The impetus for developi ng a dispe rsible rock dust slurry ca me from the two requirements placed on underground coal mines. The first r equirement, from 30 C FR § 75.2 (M andatory Safety Standards – Underground Coa l Mines) a nd 30 CF R § 75.403 (Maintenanc e of incombustible content of rock dus t ), is the application of dispersible roc k dust in sufficient quantities. The second requirement, 30 CF R § 70.100 (Respirable dust standards), is the time- weighted avera ge li mit of mine workers’ exposure to airborne dust. Rock dust acts as a physical heat sink during an explosion (Cybuls ki, 1975 ; Richmond, et al., 1975; Nagy, 1981) a nd is the pr imary technique used to s upress the re action. Therefore , a pplication of rock dust in sufficient quantiti es is essential to inert coal dust a nd to pre vent continued flame propa gation. Current rock dusts wick moisture, forming a nondispersible wet paste tha t sti ffe ns to a cake when dried, there by significantly reducing or inhibiting the rock dust’s a bility to eff ectively disperse. Coal dust, due to its hydropho bicity, c an remain dispersibl e under these conditions, thereby permitting its Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 13 involvement in a propagating dust explosion. Th e pre sence of moisture in the mine can lead to the condition in which the c oal dust can be dispersed and participate in the expl osion while the rock dust becomes ineffec tual. Additionally, there have been continuing concerns of respirable diseases am ong coal miners resulting in another requirement limiting the total respirable dust to which coal miners are exposed. The respirable siz e fraction of rock dust used to inert a coa l dust explosion ca n contribute to the tot al dust being measured. As a re sult, some mines have tu rned to applying a wet r ock dust to minim ize the miner’s exposure to respirable dust . The wet applica tion of the rock dust all ows the m ine to place dust in key locations, such as the roof and ribs, and cover the exposed coal without exposing the workers to r espirable dus t. However, a side effect of the wet applic ation of the rock dust is that the rock dust no longer is dispersible due to caking and is subsequently ineffective. Currently, MSHA requires an application o f dry rock dust on top of the wet -applied ro ck du st once the wet rock dust dries (MSHA, 2015) and becomes cake d. A possible solution to thi s problem is a foamed rock dust (FRD). The hope is that the FRD h as the benefits of a normal rock dust slurry while remaining dispersible when dry. Currently, the abilit y of the FRD to fracture from a passing pre ssure wave and disperse into individu al partic les to inert a dust explosion is not known. Prior labor atory testing has, however, led to pr omising results showing favoura ble dispersion characteristics with a few specific formulations. Hierarchically, the resulting engineering control of respirable dust through the application of foam dust would foster better healt h for coal miners, while ke eping them safe from coal dust explosions. The F RD also has the potentially additional benefit of better adherence to elevated surfaces in the mi ne entry leading to increased effectivene ss of the inert material. Past studies have shown that rock dust located on the roof, ribs and other elevated surface s is more readily dispersed than dust on the floor (Hartmann, 1957). To this end, laboratory experiments were con ducted on a previously tested roc k dust which incorporated a two-part foam (TPF). Past studies showed promising results when c ompared to other foam products currently on the market (Brown et al., 2018). Howeve r, the unique formulation and application properties of the TPF with rock d ust are still not understood. Therefore, NIOSH researchers conducted a series of l aboratory-scale experiments usin g the N IOSH-developed dispersibility chamber to examine the effects of testing conditions and product formulations on foam dispersibili ty . The insights gained were then applied during the pilot-scale testing in the Bruceton Experimental Mine (BEM) at the P ittsburgh Mining Research Center . Again, a se ries of tests were conducted to investigate the im pact of v arious f actors on the dispersion characteristics of the foam. Additionally, the airborne dust levels were measured during the production and application of th e foam dust to confirm the benefits of the wet-applied method. 2. Experiments Th e test series had two main environments wh ere the s amples were pre pared — in the laboratory at the benchtop scal e and i n the BEM at the pilot scale. The bulk of the s ample preparations were conducted under laboratory conditions. The main method of discerning the performance of the rock dust was it s dispersion performance in the NIOSH dispersion chamber (Perera et al., 2016). The samples wer e compared on a relative basis to the reference rock dust used by NIOSH at the L ake Lynn Experiment al Mine (LLEM) in large-scale explosions t esting. 2.1 Experimental materials This section describes the two experimental materials used in the testing. These materials include th e various rock dusts and a foam. 2.1.1 Rock dust Several ro ck dusts were used in these s erie s o f experiments, including the refere nce limestone rock dust, a set of classified rock dusts that used t he refer enc e rock dust as input, a n d five othe r Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 14 commercially available rock dusts. The reference rock dust has been extensively used in previous experiments at the LLEM, most notably in e xperiments which lead to the implementation of the 80% incombustible rule in the United States ( Cashdollar et al., 2010). The classified rock dust used the same reference rock dust but was sieved through either a 200-mesh or a 400 -mesh sieve. The resulting oversized and undersized material wa s later used in the experiments. 2.1.2 Foam Previous experiments were conducted with v arious commercially available fo ams that could incorporate rock dust. These tests had the similar aim of cre ating a wet-applied material which whe n dried was dispe rsible via a “light blast of air” (30 CFR § 75.2). These exp eriments sho wed that the use of a TP F in this application provide d a promising solution. The TPF used in this series of tests consisted of water, a foaming agent, and a stabilizing a gent (Brown et al., 2018). 2.2 Experimental apparatus es This section describes thre e e xperimental apparatuses used d uring in the testing. These apparatuses allowed re searchers to measure the dispersion characteristics, size distribution and airbor ne dust concentrations. 2.2.1 Dispersion chamber The FRD samples p roduced in the lab or in th e BEM we re tested in N I OSH’s dispersion chamber. The chamber s ends a pulse of air across a sample bed of rock dust , dispersi ng a qu antity of the dust and sending it across a dust probe a s seen in Figure 1. Fig. 1. Dispersion chamber schematic (Perera et al., 2016) Passage of the dust across the probe causes the obscuration of a li ght beam whic h is measured and recorded (Cashdollar, et al., 1981) . The obscuration caused by these entrai ned particles is used as a means of a ssessing the dispersibility of the sa mple. The pulse of air used to initi ate the test was ba sed on the dynamic pressure measure ments o f a n ear-limit propagating coal dus t explosion conducted at LLEM , as shown in Figure 1 and, as such, provide s the rep roducible “light blast of air” descr ibed in 30 CFR § 75.2. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 15 Fig. 2. Horizontal and vertical components of dust explosion pressures measured during LLEM shot #517 The test procedure, notably, does not simulate a shockwave but, rather, creates a similar pressure history per the rapid release of the air pulse. Use of thi s method is described in greater detail by Perera et al. (2016). As shown from the aforementioned large-scale tes ting at LLEM in Figure 2, vertical or static forces are created due to the passing of a shoc k wav e. The static pre ssure in Figure 2 is about three times greater than the dynamic pressure. I t is reasonable to a ssume that the overpressure may destroy the cohesive foam matrix, freeing individual particles or particle agglomerations for dispersion. Computational models h ave observe d these vertical for ces in a ddition to re flected compaction waves in the dust layer just behind the shockwave ( L ai et al., 2018). As a result o f these findings, samples were teste d using the additional 45° nozzle or ienta tion (Figure 3). The change in orientation allowed the researc hers to add a vertical force component. Fig. 3. 45° nozzle orientation 2.2.2 Particle size The particle size dist ributions of rock dust s were determined using a Beckman-Coulter LS 13 320 laser scattering size analyser. The Beckman-Coulter instrument measure s the scatte ring of a 780-n m 0 2 4 6 8 10 12 14 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 Pressure, p si Time, sec Static Wall Pressure Dynamic W ind Pressure Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 16 laser beam by the entra ined dust at various a ngles with respect to the bea m direction. The inst rument returns the particle size dist ribution in terms of the spheric al equivalent via Mie scattering. 2.2.3 Dust concentration monitor Dust conce ntrations were monitored and mea sured during the produ ction and a pplication of the foam rock dust in the BEM. The gravimetric samples were colle cted on pre -weighed filters using the standard Zefon Escort ELF personal air sampl ing pumps after the respirable size fraction was separa ted from the oversized dust by a 10- mm Dorr- Oliver cyclone. Ad ditionally, a continuous personal dust monitor from Thermo Scientific , the PDM 3600, was used to monitor dust concentrations during the test via a tapered eleme nt oscillating microbalance (Page et a l. , 2008). 2.3 Experimental sample preparation This section describe s the proce ss of sample preparation in the lab and in the pilot -scale test facility. In addition this section details the changes made to the sample variables. 2.3.1 Laboratory sample prep The dry rock dust samples, used for th e rel ative c omparison, we re prepared in accordance with the standard operation procedures of the NI OSH-developed dispersion chamber (Perera et al., 2016). To prepare the FRD samples, the TPF was prepared first by combining 300 g of water and 15 g o f the foaming agent. This solution was whipped with a mixer until a consistent foam matrix was established. A stabilizing agent was then slowl y add ed to th e foam matrix. Once blend ed, the stabilized TP F was combined with the rock dust. The samples were then placed under two drying conditions. The f ast-drying condition was the am bient lab environment. The slow-drying condition placed the samples in se aled chamber s with high humidity a nd no airflow. After a week in th e chamber, the samples were taken out and allowed to dry under ambient lab conditions. All samples were dispersed using a 45-degree nozzle. Initial rounds of testing examined the effects o f changing the amount of stabilizer and the amount of rock dust used in the formul ation. The a mounts of stabilizer tested w ere 10 g and 21 g. Th e a mounts of rock dust tested were 600 g, 900 g, and 1,500 g. Other FRD samples w ere made by adding water to the rock dus t prior to being mixed with a different blend of foam. The TP F in these ca ses k ept the ratio of st abilizer and foaming agent at 1:1. The stabilizer levels were 6.5 g, 5 g, and 3.5 g per batch while keeping the rock dust mass at 1,500 g and the a dditional water a t 400 g. The a mount of pre-wet rock dust also varied, but the ra tios of roc k dust to water and the amounts of stabilizer a nd foaming agent (5 g) r emain the same. Five total sa mple sets were run with diff erent commercially available rock dust s using the sa me FRD formulation. The formulation consisted of 300 g of water , 5 g of foaming agent, 5 g of stabilizer, 1,500 g of rock dust, and 400 g of water to pre-wet the rock dust. The final sample set produced in the lab used the same previous formu lation except with four classified batches o f the reference ro ck dust. The r efe rence ro ck dust was si eved through either a 200 - mesh or a 400-mesh sieve. In both ca ses, 1,500 g of the ove rsized or unde rsized material was used in the FRD, ke eping all other ingredients the same. Each combination of drying condition and formulation tested a tot al of five samples. A tabulation of the various sample sets is shown in Table 1 below. Table 1: Laboratory tests examining the impact of various components on the dispersibility of foamed rock dust Sample set FOAM ROCK DUS T Drying Conditions Water (g) Foaming Agent (g) Stabilizer (g) Base rock du st Amount (g) Water (g) Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 17 1 300 15 10 Reference 600 - Fast & Slow 2 300 15 21 Reference 600 - Fast & Slow 3 300 15 10 Reference 900 - Fast & Slow 4 300 15 21 Reference 900 - Fast & Slow 5 300 15 10 Reference 1500 - Fast & Slow 6 300 15 21 Reference 1500 - Fast & Slow 7 300 6.5 6.5 Reference 1500 400 Fast & Slow 8 300 5 5 Reference 1500 400 Fast & Slow 9 300 3.5 3.5 Reference 1500 400 Fast & Slow 10 300 5 5 Reference 1275 340 Fast & Slow 11 300 5 5 Reference 1725 460 Fast & Slow 12 300 5 5 Commercial R D 1 1500 400 Fast 13 300 5 5 Commercial R D 2 1500 400 Fast 14 300 5 5 Commercial R D 3 1500 400 Fast 15 300 5 5 Commercial R D 4 1500 400 Fast 16 300 5 5 Commercial R D 5 1500 400 Fast 17 300 5 5 - 40 0 mesh Reference 1500 400 Fast 18 300 5 5 - 200 mesh Reference 1500 400 Fast 19 300 5 5 + 400 mesh R eference 1500 400 Fast 20 300 5 5 + 2 00 m esh Referenc e 1500 400 Fast 2.3.2 BEM sample preparation Anemometers we re plac ed upwind and downwind of the test area to m onitor airflow velocities. Additionally, the dust sampling equipment was pla ced 100 ft downwind of the test are a. The mortar mixer used to blend and apply the FRD was turned on and allowed to r un to collect a baseline level of diesel partic ulate matter (DPM). The test matrix is t abulate d in Table 2. Table 2: Te st matrix for pilot -scale testing two-part foamed rock dust in the BEM Test Condition s Test Number Dust, kg Water, l Amount of fo am Triplicate 1 45.4 11.4 40 second of foam Triplicate 2 45.4 11.4 40 second of foam Triplicate 3 45.4 11.4 40 second of foam +15% rock du st 4 52.2 11.4 40 second of foam -15% rock dust 5 38.6 11.4 40 second of foam -15% water 6 45.4 9.7 40 second of foam +15% water 7 45.4 13.1 40 second of foam All the tests used the reference rock dust as the base inert material. The firs t three tests were a s et of triplicates that emulated the laboratory formulations. The following four tests looked at variations of th e formulation to see ho w the foam would be impacted if a mine worker would improperly mi x the inputs to the system. The first two tests changed th e a mount of rock dust by ± 15%, and the other two tests varie d the w ater use d to pre-we t the roc k dust by ± 15%. The parameters for the fo am generator would be initially set and not altered by the machine operator. On ce th e inst rumentation was in pla ce, a premeasured amount of dust was placed into the mixing hopper of the mortar mi xer. The premeasured amount of wat er was then added to the rock dust and allo we d to mix. At this point, the Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 18 calibrated fo am generator was turned on, and a f oam was added to the sl urry for a predetermined amount of time. This was mixed until the mixture was homogenous. When the mixer was st opped, a sample was taken with a 1 -qt container and weighed to get a density measurement. Additi onal material was taken from the mixer a nd use d to fill trays for late r testing in the dispersion c hamber. The initial weight of the trays with the f resh sample s was taken at this point for use in later calculations to de termine the FRD density and measured water content . Once the sa mples we re take n out of the mixer, th e rest of the material was dumped into the hopper and the auger pump was turned on. A 1-qt sample was the n taken as the material left the application nozzle. This was then weighed for a density measurement. Additional material was also taken from the end of the nozzle for the dispersion trays. These trays also h ad their mass recorded. Once the l ast sam ple of mat erial was taken, the FRD was a pplied to the roof and ribs of the m ine. This continued until there was no more FRD left for the pump to move. The are a wa s then marked with spray paint to de note the test number, a nd the application equipment and dust monitoring equipment were then moved to the next testing location. The tray samples were kept in the mi ne under the same drying condit ions until the sample weights stabilized, just over one week. The dispersion testing in the lab bega n about 2 we eks a fter the application. 3. Results and discussion 3.1 Lab results 3.1.1 Dispersibility of the reference rock dust (dry powder) The re lative dispersibility of the dry refere nce rock dust wa s tested using the 45° nozzle instead of the horizontal nozzle. The resulting average (Fi gure 4) was used as a benchmark to compa re th e relative dispersibilities of the foamed dusts. The avera ge integral optical densities (IOD) of the FRD is compared to that of the dry rock dust powder. After all data was normalized, the relative IOD for the refere nce rock dust was 1 ± 0.11. Fig. 4. Relative average optical density trace of dry reference rock dust 3.1.2 Environmental/testing conditions of the foam The effec ts of high humidity exposure did show a marke d negative difference in the disp ersibility of the foams wh en comparing the combined averages from sample sets 1 through 11 (Figure 5). Although a number of s amples did disperse after be ing plac ed in the humidity ca binets, it was to a lesser extent than when the samples were dried quickly in the laboratory environment. 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0 0.5 1 1.5 2 2.5 3 3.5 4 Relative o ptical density Time, s R elativ e a ver ag e IO D for re f er enc e r ock du st disper sed at 45° - 1.00 ± 0.11 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 19 Fig. 5. Relative average IOD of fast and slow-drying FRDs compared to the reference rock dust In some cases, the bubbles in the slow -drying foam s began to agglomerate, which sti ffene d its lattice. These resulting agglomerations remained cohesive enough that when the bulk materia l broke , the agglomerations fell out of suspension very quickly and did not reach the dust probe , resulting in poorer performa nce. 3.1.3 Variation in formulation Using sample sets 1 through 6, researchers found that the increased amount of stabilizer, while aiding in the foam’s rigidity during longer periods of high humidit y, negatively affec ted the dispersion character istics by reinforcing the c ohesive characteristics of the dust. Converse ly, the lowe r levels of stabilizer a llowed for a rigid yet friable structure which more readily disperse d. Another important note was th at few samples collapsed due to insufficient amounts o f mater ial being present as the samples dried. Conversely, better dispersion characteristics were seen with samples containing 900 g or 1,500 g of rock dust per batch. Th ese results may be due to the changing ratio of water to ro ck dust, which, cou ld be aff ecting the ra te of drying per unit volume. The add itional material also coul d be simply translating to more material being entrained and, therefore, more obscuration at the dust probe. Lastly, the formu lations which used higher amounts of ro ck dust had a de creased ratio o f stabilizer in the formulation. All these factors may have led to the increases in performance. The results of these variations to the formulation under the fast- and slow-dryin g conditions can be seen in Figure 6 and Figure 7, respec tively. 0 0.2 0.4 0.6 0.8 1 1.2 Fast Slow Reference R elativ e a ver age IOD f or 45° air je t orient ation of slow and f ast drying foame d r ock dust Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 20 Fig. 6. Relative average IO D for fast-drying foam formulations fr om sampl es 1-6 Fig. 7. Relative average IO D for slow-drying foam formulations from samples 1 -6 The perf ormance of the f oam see med to be mos t sensitive to levels of stabili zer in the formulation when considering variations to the components of the FRD. Additional efforts we re made to keep the amount of ro ck dust hig h while de creasing the l eve ls of stabilizer further without causing advers e effects to the foam. To achieve these aims, water was added to the rock dust before being blend ed into the foam. This practice p revented the rock dust from abso rbing to o much water out of the foams ’ lattice, which could have produce d nega tive e ffects. The results fr om sample sets 7 through 11 ca n be seen in Figure 8 for fast-drying samples and F igure 9 for slow -drying samples. 21 10 0 0.2 0.4 0.6 0.8 1 600 900 15 00 Amount of stabilizer p er batch, g Relative average IOD Amount of rock dust per batch, g 21 10 0 0.2 0.4 0.6 0.8 1 600 900 15 00 Amount of stabilizer p er batch, g Relative average IOD Amount of rock d ust per batch, g Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 21 Fig. 8. Relative average I OD for fast-drying foam formulations from samples 7-11 Fig. 9. Relative average I OD for slow-drying foam formulations from samples 7-11 The most consistent formulation that performed well in both the fast - a nd slow-drying conditions used 1,500 g of rock dust with 400 g of water, while the stabilizer and foaming agent we re held at 5 g per b atch. This formulati on was then compared to the origin al dry reference rock dust. Th e obscuration traces a re shown in Figure 10. 3.5 5 6.5 0 0.5 1 1.5 2 1275 g rock dust & 340 g water 1500 g rock dust & 400 g water 1725 g rock dust & 460 g water Amount of foaming and stabilizin g agent, g Relative average IOD 3.5 5 6.5 0 0.5 1 1.5 2 1275 g rock dust & 340 g water 1500 g rock dust & 400 g water 1725 g rock dust & 460 g water Amount of foaming and stabilizin g agent, g Relative average IOD Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 22 The ASTM F716-09 an d ASTM F726-12 standards have b een developed for absorbe nt/adsorbent performa nce and upt ake capacity testing. However, the ASTM standa rds have some shortcomings that make them un attractive for researchers to u se (Bazargan et al., 2015). In particular, th ey all focused on the uptake capacity testing without reference to vaporization or fuming from the liquid . It is c lear that there is a lack of studies conducted t o examine the absorbents in suppressing the fumes genera ted fr om the spill of water-reac ting/fuming liquids which may affect human health. Tests we re conducted in this study by means of a small -scale evaporating cha mber and real-time FTIR analysis with the goal of evaluati ng the diffe rent forms of ab sorbents in terms of their capabilit y to suppress the fuming. 2. Experiments 2.1 Materials The phosphorous Oxychloride used in this work was obtained from J. T. Baker Chemical Company, Phillips burg, New Jers ey with a stated purity of greater than 99.97 %. Sorbents with different properties were used including PetroGua rd ® c hemical spill absorbe nt and T rivore x ® neutra lizing absorbent. PetroGuard ® is a sponge, soft polymeric materials as shown in Fig. 6 that solidifying chemical and oil spill. It is also claim ed that with P etroGua rd ® , most common water reactive chemicals, hazardous hydrocarbons and sil icone-based ch emica ls are absorbed, encapsulated, solidified, immobilized, and stabili ze d quickly and effectively absorbent (Guardian Environmental Technologies, 2020). Trivorex ® is a neutralizing absorbent which may ne utralize acids, bases, oxidizing agents, reducing agents, and solven ts (Prevor, 2020). It is also claimed that the neutraliza tion process will help to suppress the emission of hazardous gases from corrosive chemicals. There is however no report with quantifying efficacy on both absorbents. 2.2 Equipment The experime ntal equipment consisted essentially of e vaporating cha mber and a FTI R spectroscopy. Fig. 1 shows the configuration of the small-scale evaporating cha mber used in this work to facilitate the measurement of vap orization and fuming fr om a liquid pan with controlle d ventilation. This chamber is made of stainless steel and equipped with 3 removable sight glasses to facilitate visual check. The v entilation rat e was controlled by a mass-flow-controlled air stream to a fixed 20 slm with a known dew point temperature of 10.2±0.4 °C. A hood was mounted to the exit of the chamber having an area of 10 0 × 1 00 mm to capture the ve nted gases for subsequent gas treatment. The stainless-steel evapo rating pan ha d an area of 200 × 200 mm and was 20 mm de ep. During the test, the weight of the pan containing POCl 3 liquid and sorbent was recorded by using a precision scale; the temperature was moni tored by means of a type K thermocouple immer se d in the test substance. A FT -IR instrument with a 10 cm li ght path gas c ell was employed to measure the concentration of gasses from th e exit of th e cha mber which was used as the indicator for efficacy of the absorbent to suppress the fuming. The wave number range considered was from 600 to 4000 cm⁻¹ . S pectra were obtained every 2 .14 seconds with a resolution of 1 cm⁻¹ . The chemical functional groups of th e sorbents were determined by a Smiths Ide ntifyIR Attenuated Total Reflection (ATR) Fourier transform infr ared spectrometer with a scan ra nge of 650 – 4000 cm⁻¹ . Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 29 Fig. 1. Detail drawing of evaporating chamber 2.3 Experime ntal procedure After the evaporating pan was filled with around 170 g of liquid POCl 3 , the system was then allowed to come to equilibrium. The time required to come to equilibrium in the first stage was usu ally 10 to 20 minutes. The system was considered to be in equilibrium when the co ncentra tion of the vented gases stays constant. Once the equilibrium of the first stage had been established, the top sight glass was dismantled and a given amount of ab sorbent was then applied and spread evenly over the surface of the liquid POCl 3 . The sight glass was quickly reassembled. The experiment e nded when the system was in e quilibrium which took 45 to 230 minutes depending on the testing sorbent. All tests were performed at room temperature. The temperature and weight of the substance w er e recorded, along with the concentrations of vented g ases. The identification and quantification of gases were confirmed by I R spectra of ga seous POCl 3 , HCl and CO 2 . 3. Results and discussion 3.1 The characterization of test sorbents According to Fingas and Fieldhouse (2011), styrene-butadiene and related polymers have been the most common materials of polymer sorbents. T he IR spe ctrum of PetroGu ard ® (Fig. 2) i s similar to that of butadiene styr ene polymer obse rved in several stud ies (Liu et al., 2019; Luna et al., 2020; Zieba-Palus, 2017) that the band at 697 and 755 cm⁻¹ are attributed to CH deformation vibrations in styrene aromatic rings ; the weak bands at 1491 and 1599 cm⁻¹ can be assigned to ring vibrations of styrene; the weak b ands at 745, 963, and 907 cm⁻¹ can be assigned to 1,4 cis-butadiene, 1,4 trans- butadiene and 1,2-butadiene segment, respectively; the bands at 1377 and 1452 cm⁻¹ can b e attributable to C H 3 deformation vibrations; the bands at 2848, 2917, and 2955 cm⁻¹ can be assigned to C-H stretching vibrati ons; and the ba nds a t 291 7 a nd 2848 cm⁻¹ can also be assigned to elongatio n of C H groups in styrene aromatic rings. The integration of these peaks indicate s that P etroGua rd ® contains mainly butadiene styre ne polymer. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 30 Fig. 2 . IR spectrum of PetroGuard ® absorbent Fig. 3 compares the IR spectrum of Trivorex ® a nd sodium bicarbona te (NaHCO 3 ). As a r esult, Trivorex ® contains signif icant amount of sodium bicarbonate due to the obvious similarities at 687, 829, 1393, 1609, and 1905 cm⁻¹ . In addition, t he ba nd at 1085 cm⁻¹ may be due to the Si -O-Si asymmetric vibrational mode of stoichiometric silica (SiO 2 ) (Tolstoy et al., 2003). These observations are in agreement with Mathilde et al. (2019). These peaks confirm that Trivorex ® contains mainly NaHCO 3 which is a c ommon neutralizing agent. Fig. 3. IR spectra of Trivorex ® neutralizing absorbent and sodium bicarbonate (NaHCO 3 ) 3.2 The efficacy of PetroGuard ® The effectiveness of PetroGuard ® applied to POCl 3 was quantitatively judged by perf orming the test with diffe rent amount of absorbent. In the first test, 129.59 g of PetroGuard ® was applied to 164.92 g POCl 3 . Fig. 4 shows the concentration of POCl 3 and HCl of the gas es measured at the exit of the chamber. It shows that after 6 mi nutes P OCl 3 liquid wa s poured into the pan, th e syst em was in equilibrium with the ave rage POCl 3 and HCl concentra tion of 798.5 and 803. 3 ppm, re spectively an d the tempera ture of the liquid increase slightly from 30 to 31 .4 °C. During this period, the average weight loss is 0.311 g/min. Imme diately after applying the PetroGuard ® , fuming was stopped with no detectable POCl 3 or HCl. However, HCl appe ared again after 2 min and POCl 3 a lso reappear fo r a nother 2 min utes. After 50 minutes of applying P etroGua rd ® to the pan containing POCl 3 , the concentration of POC l 3 and HCl were found to be st ea dy at 447.9 ppm and 967.0 ppm, respectively; the temperature slightly increa sed Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 31 from 30.5 to 33.2 ° C. During this period, the average weight loss is 0.166 g/min. According to F ingas and Fieldhouse (2011), the disadvantage of this type of so li difier, styr ene-butadiene-containing absorbent, is that the absorbate c an be r elea sed, es pecially under some pr essure. Therefore, right after coming into contact, POCl 3 was quickly absorbe d by PetroGuard ® that can explain for the period o f 2 minutes without fuming and detectable POCl 3 or HCl. POCl 3 wa s then r eleased slowly and reac ted with moisture to form HCl leading to a signi ficant decrease of 44% of POCl 3 concentration (from 798.5 to 447.9 ppm) whi le 20% in crease in HCl conce ntration was obs erved (from 803.3 to 967. 0 ppm). Based on the experimental observation, we presumed that the re was no chemical r eaction between POCl 3 and PetroGuard ® . Therefore, the h ydrolysis reaction of POCl 3 and moisture yielding HCl is responsible for the temperature rise in this test. Abatement efficiencies based on measured drops in POCl 3 conce ntration and weight loss rate are 43.3% and 46.6%, respec tively. The same reduce d weight loss rate and reduced P OCl 3 concentra tion suggests that the PetroGuard ® helps to suppress the v aporization of POCl 3 liquid through absorbing the liquid. N evertheless, the surface of the absorbed materials remains continued to emit vapor as well undergo hydrolysis with moi sture. Fig. 4. Results for suppressing 164.92 g of POCl 3 with 129.59 g of PetroGuard ® absorbent In the s econd test, 251.5 g of P etroGua rd ® was appli ed to 168.1 g of POCl 3 . Fig. 5 shows that after 35 minutes applying PetroGuard ® to P OCl 3 , the ave rage steady concentrations of POCl 3 and HCl decreased significantly from 813. 1 ppm to 196. 9 ppm and from 1072. 4 ppm to 558.5 ppm, respec tively with a higher the aba tement efficiency of 75.3 % c ompared to 43.3 % in the fir st test. The steady weight loss rate was decreased from 0.322 g/m in to 0.062 g/min, givi ng an abatement efficie ncy of 80.7% wh ich is comparable to the abatement efficiency based on r educed P OCl 3 concentration. Therefore , with increase quantity the PetroGuard ® may covered and isolated the POCl 3 liquid from evaporation and reacting with moisture. Fig. 8 shows th e POCl 3 liquid, POCl 3 liquid covere d with PetroGuard ® , and a bsorbed PetroGuard ® . The absorbed PetroGuard ® wa s soft, wet and still emitting vapor. Thus, P etroGua rd ® may provide a rapid vapor suppres sion of the fuming liquid right after the spills. The rapid suppression is useful for eva cua ting the nearby communit y but the complete removal or abatement still requires other response actions a s the absorbed materia ls re main fuming. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 32 Fig. 5. Results for suppressing 168.10 g of POCl 3 with 251.1 g of PetroGuard ® absorbent. 3.3 The efficacy of Trivorex ® In this test, 128.24 g of Trivorex ® was applied into 172.92 g of POCl 3 . In the first stage, the system was in equilibrium after 15 mi nutes with a steady weight loss ra te of 0.304 g /min . The reaction behavior between POCl 3 and T rivore x ® , how ever, was much different from that of PetroGuard ® . In particular, an immediate ly violent reaction with vigorous boiling took pla ce when Trivorex ® and POCl 3 came into contact. The temperature soared quickly from 32.3 to 108.5 °C in 3 minutes acc ompanied by a lar ge amount of gaseous POCl 3 , HCl and CO 2 as shown in Fig. 6. After a round 210 minutes applying Trivorex ® , the temperature then gradually dropped to 34.1°C and the POCl 3 concentration wa s found to be less than the limit of de tection of it to F T -I R (50~60 ppm). Sodium bicarbonate containing in Trivore x ® as a neutralizing agent which may trigger highly exothermic reac tions and the evolution of CO 2 according to the following schemes: 𝑁𝑎𝐻𝐶𝑂 3 + 𝐻𝐶𝑙 → 𝑁𝑎𝐶𝑙 + 𝐻 2 𝑂 + 𝐶 𝑂 2 (1) 3𝑁𝑎𝐻𝐶𝑂 3 + 𝑃𝑂𝐶 𝑙 3 → 𝐻 3 𝑃 𝑂 4 + 3𝑁𝑎𝐶𝑙 + 3𝐶 𝑂 2 (2) Fig. 7 shows the real weight of POCl 3 and absorbe nt monitored by the precision scale during the experiment. It is clear that the real weight loss was divided into two appare nt parts. In the first part, after 40 seconds applying Trivorex ® , the we ight loss was 71.12 g (decreased fr om 302.62 to 231.5 g). In the second part, the weight loss continued gra dually from 231.5 g to 208.45 g before sli ghtly increased aga in. The tota l weight loss after a pplying Trivorex ® wa s 94.16 g. The rapid weight loss is accompanied by spikes in CO 2 and POCl 3 concentration to more th an 60,000 and 30,000 ppm, respectively. The spikes were however dropped quickly in 30 s to less than half of these peak values. In add ition, CO 2 was completely diminished by 8 min after the application. The presence of C O 2 is a direct evidence of n eutraliz ation acc ording to scheme (2). The spikes in CO 2 and POCl 3 concentrations and weight loss confirm that the weight loss is a direct consequence of neutraliza tion and vaporization of POCl 3 . How ever, the measured c oncentration of CO 2 and POCl 3 are far less than the corresponding weight loss which indi cates that the FT I R concentration measurement is inc apable to cope with such high concentration. The ratio of measu red peak concentration of CO 2 and POCl 3 is about 2 which may provide an indication of relative cont ribution of neutralization and vap orization. The weight fra ction of CO 2 in the total weight loss can thus b e calculated to around 0.36 based on mol ar ratio of CO 2 to POCl 3 . Thus, approximately 25. 6 g of the weight loss can be attributed to neutralization reactions while the remaining 45.52 g is a ttributed to Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 33 the vaporization of POCl 3 . The second part of the weight loss was purely va porization which is 23.05 g. Total vaporization of POCl 3 is 45.54+23.05 = 68.59 g which accounts for 40% of the total liquid spilled in the pan. Thu s, an estimation of th e gross efficacy of abatement by Trivorex ® is approximately 60%. Fig. 6. Results for suppressing 172.92 g of POCl 3 with 128.24 g of Trivorex ® neutralizing absorbent Fig. 7. The weight of liquid POCl 3 and absorbent during the experiment 3.4 Discussions Fig. 8 shows the photo s of POCl 3 liquid, POCl 3 liquid covered with PetroGuard ® , and absorbed PetroGuard ® . The abso rbed PetroGua rd ® w as soft , we t a nd sti ll e mitting va por. Thus, PetroGuard ® may provide a r apid v apor suppression of the fuming li quid right aft er the spil ls. The r apid suppression is useful for e vacuating the n earby co mmunity but the c omplete re moval or a batement still requires other re sponse actions as the absorbed materials remain fuming. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 34 Fig. 8. Photos from left to right: POCl 3 liquid, POCl 3 liquid cov ered with PetroGuard ® , and absorbed PetroGuard ® . Fig. 9 shows the photos of POCl 3 liquid before and after neutralized with Trivorex ® . Although the neutralize d Trivorex ® is no longer fuming with very limited HCl detected, care must be taken due to the violent exothe rmic reaction with vigo rous boiling leading to a significantly higher evaporation rate of POCl 3 and e volution of HCl . Without doubt, the vaporization will crea te a large impac t on the nearby c ommunities. In f act, it will be a challenging task even for e mergency responders in applying Trivorex ® owing the e xc essive he at. Fig. 9. Photos from left to right: POCl 3 liquid before and after ne utralized with Trivorex ® . In comparison of the two absorbents, one be ing temporarily suppressing v aporization while the othe r one provides ultim ate abatement through neutr alization, both suffer diff erent drawbacks and great care s are required for applyi ng both a bsorbents. More work is certainly neede d to develop a better material and/or method f or abating the spill of wa ter-reactive/fuming chemicals. Finally, the current design of chamber provides a consistent and insightful way to assess the risks associated with the abatement of chemica l spills. 4. Conclusions We have d eveloped a chamber method to investigate the eff icacies of different commercial absorbents including PetroGuard ® and Trivorex ® for suppressing the fuming from P OCl 3 spills. Excessive amount of PetroG uard ® will absorb and suppress the fuming effectively. In contrast, ne utralization agent Trivorex ® will react viol ently with POCl 3 relea sing excessive he at leading to the highe r evapora tion ra te of P OCl 3 a nd the evolution o f H Cl gas. The results a re crucial for dev eloping best practice s for mitigating the release of water-reacti ve che micals. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 35 Reference s Barsan, M.E., 2007. NI OSH pocket guide to chemical hazards. Bazargan, A., Tan, J., M cKa y, G., 2015. Standardization of oil sorbent performance testing. Journal of Testing and Evaluation, 43, 1271-1278. Fingas, M., Fieldhouse, B ., 2011. R eview of soli difiers. in: Oil spil l science and technology . Elsevier, 2011, pp. 713-733. Kapias, T., Griffiths, R., Stefanidis, C ., 2001. REA CTPOOL: a code implementing a new multi - compound pool model that accounts for chemical reactions and changing c omposition for spil ls of water reactive chemicals. Journal of hazardous materials, 81, 1-18. Liu, J., Min, X., Zhang, X., Zhu, X., Wang, Z., Wang, T., Fan, X., 2019. A nov el synthetic str ategy for styrene – butadiene – st yrene tri-block copolymer with high c is-1, 4 un its via c hanging catalytic active centres. Roy al Soc iety open sc ience, 6, 190 536. Luna, C.B.B., Araújo, E.M., Siqueira, D.D., Morais, D.D.d.S., Filho, E.A.d .S., Fook, M.V.L., 2020. Incorporation of a recycled rubbe r compound from the shoe industry in p olystyrene: effect of SBS compatibilizer c ontent. Journal of Elastomers & Plastics, 52, 3-28. Mathilde, N., Mathieu, L., Blomet, J., Meyer, M. -C., 2019. Pollution re moval composit ion and use thereof. in: Google Pate nts. Muller, T.L., 2000. Sulfuric acid and sulfur trioxi de. Kirk‐Othmer Encyclopedia of Chemical Technology . Pr evor T rivorex®. https://www.prevor.com/us/trivorex. Tolstoy, V.P., Chernyshova, I ., Skryshevsky, V. A., 2003. Handbook of inf rared spectroscopy of ultrathin films . John Wiley & Sons. Ullmann, F., Gerhartz , W ., Yamamoto, Y.S., Campbell, F.T., Pfefferkorn, R., Rounsaville, J.F., 1985. Ullmann's encyclopedia of industrial chemistry . VCH publishers. Zieba-Palus, J., 2017. The usefulness of infrared s pectroscopy in examinations of adhesive tapes fo r forensic pur poses. Foren sic Sci. Criminol., 2, 1-9. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 36 13th International Symposium on Hazards, Prevention, and Mitigat ion of Industrial Explosions Braunschweig, GERMANY – July 27-31, 2020 Modeling of E xplosion Dynam ics in Vessel-Pipe Sys tems to Evaluate P erformance Limi tations of Explosion Isol at i on Systems Lorenz R. Boec k, C. Regis B auwens & Serge y B. Dorofeev FM Global, Re search D ivision, Norw ood, MA, U SA E-mail: lorenz.bo ec k@ fmglobal.c om Abstract Explosion isolation syste ms provide critical protection for interconnected vessels and work areas, preventing the spre ad of explosions through interconne cting pipes and ducts. The se syste ms not only prevent pro pagating events, but also miti gate the elevated e xplosion hazards of interconnected vessels, caused by pr essure piling and enhanced turbulence. Explosi on isolation systems can , however, fail ca tastrophically if they are not prop erly designed f or a use case. Evaluating the performance limi tations of explosion isolation syst ems includes assessing their pressure resistance, fla me-barrier eff icacy, and appropriate installation distance s, which typically requires extensive testin g. To predict th e performance of a system for use cases outside the tested conditions, models are needed that reliably predict both the explosion dynamics and isolation syste m response. In this study, a previou sly developed physics -based model for explosions in vented v essel-pipe systems is further extended and validated. Large-scale validation experiments were performed at an 8-m 3 vessel with attached pipes, varying the p ipe dimensions, ignition location, and mixture reac tivity. It is found that the model ca ptures the effects of the expe rimentally va ried parameters , and accurately predicts the time available for isolation systems to form a flame barrier . These results sho w how this model can help pre dict installation distances as p art of the overall performance limitations of explosion isolation systems, extending the range of use c ases and reducing the number of tests needed for system eva lua tion. Keywords: experiments, explosion isolation, modeling, vessel-pipe systems Nomenclature Symbol Description Units Values a Flame cone base radius (m) D P Pipe diameter (m) h Flame cone heig ht (m) k V Calibration p arameter for flame adv ection (-) 0.35 L P Pipe leng t h (m) p Vessel pressu re (Pa) R Vessel radius (m) r f Flame radius (m) R V Vent radius (m) u x Advection velocity at the leading flame-tip position (m/s) X Fuel con centration, by volume (%) x f Flame tip position (m) x f * Flame tip position in coordinates of the virtual origin (m) α Flame cone ang le (rad) γ V Ratio of h eat capacities of vented gases (-) ρ V Density o f vented gases (kg/m 3 ) 𝜁 Ignition locatio n param eter (-) 0.25 ; 1; 2 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 37 1. Introduction Explosion isolation is a key element of explosion prote ction that mitigates the effec ts of explosions in interconnected vessels (Hattwig and Steen, 2008). Figure 1 il lustrates an explosion originating in a primary vessel or piec e of equipment and prop agating through an interconnecting pipe toward a secondary vessel. Compared to isolated v essels, int erc onnected vessels present a mo re severe explosion hazard: As combustion occurs in the primary vessel, pressure i ncrea ses in both vessels ; pressure is elevated in th e secondary vessel when the flame ent ers, and the peak explosion pressure increases proportional ly . This effect is known as pressure piling. In addition, turbulence is enha nced in the pipe and in the secondary vessel due to the flow generated ahead of the flame, leading to higher rates of pressure rise, w hich may overwhelm explosion protection designed to protect indi vidual vessels. A highl y turbulent flame-jet entering the secondary vessel similarly enhances the rate of pressure rise. (Bartknecht, 2012; Eckhoff, 2016; Taveau, 2017) Fig. 1. Illustrati on o f explosion in interconnected vessels. Combustion produces turbulence ahead of the flame and elev ates pressure, resulting in pressure piling and high rate of pressure rise. Explosion isolation valve closes pipe c ross-se ction to preve nt flame spread. To prevent these ha zardous effec ts, and prevent fla me spread, explos ion isolation sys tems are installed at interconnecting pipes to form a flame barrier in the event of an e xplosion. Both active and passive systems are co mmonly used, which activate electronically ba sed on pre ssure or fla me detection, or are triggered by pressure waves and flow ahead of the flame, respectively. To successful ly isolate an explosion, an isolation device must satisfy three criteria, including: (i) pressure resistance: the isolation device must withstand the explosion overpressure; (ii) barrier efficacy: the flame barrier formed by the device must prevent trans mission of flame and/or hot gases/mater ial that may cause re -ignit ion downstrea m; (iii) rapid barrier formation: an effective flame ba rrier must be formed before the flame arrives at the device location and maintained for the duration of the e xplosion. As all isolation systems require some finite time to respond, t he third requirement translates into a minimum necessar y installation distance for an isolation system, the distance along the pipe between vessel and isolation device, which allows time for the barr ier to develop prior to flame arrival . The installation distance depends on the spe cific use c ase, de termined by p arameters such as vessel an d pipe dimensions, mixture reactivity, ignition location , and the presence of additional explosion protection measures suc h as venting or suppression . To determine installation dist anc es, existing testing standards, such as EN 15089, Explosion I solation Systems , require a significant number of Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 38 Fig. 5. Effect of mixture reactivity (propane concentration X) on pressure (top) and flame arrival (bottom). Experiments (grey lines and markers) and model predictions (red lines). Dotted lines mark static vent deployment pressure for each test. L P = 9.75 m, D P = 0.4 m, center ignition. The pea k pressures at bo th X = 3.0% and X = 3.5 % clearly exceed the static deployment pressure of the vent, indicated with black dotted lines. This is due to rapid pressure rise in both cases: The pressure exceedance above the static vent deployment pressure increases as the rate of pressure rise in creas es with increasing mixture reac tivity. It was observ ed that the pres enc e o f a pipe can lead to higher dynamic vent deployment pressures compared to s imply vented explosions, without pipes attached to the vessel. 4.3 Effect of ignition location Figure 6 shows experimental results and model predictions for dif ferent ig nition locations. For front ignition, the flame shape inside the vess el is approximately hemispherical, l eading to a mode rate rate of combustion and pr essure ris e. For this c ase, th e model over -predicts th e early p ressure peak, bu t still captures flame prop aga tion through the pipe accurately. Cases with front igni tion and lean mixtures are typically the most challenging from a modeling perspective, as small model deviations integrate over longer times and flame speeds are highly sensitive to mixture composition. Center ignition shows faster initi al pressure rise due to a spherica l flame shape and higher rate of combustion. After flame arrival at the pipe, pressure rises rapi dly, which is attribute d t o higher flow velocity in the pipe, incr easing the t urbulent burning rate in t he pipe, and a la rger flame surface area insi de the vessel at thi s time, produ cing faster pressure rise. For back -wall ignition, initial pressure rise is slow due to the hemispherical flame geometry. As the fla me a rrives at the pipe, higher vessel pressure and resulting higher flow v elocity in the pipe, c ompared to center ignition, lead to high turbulent burning rates in the pipe and the sharpest rise in vessel p ressure among t he three tests. Back-wall igni tion genera lly produces the highest peak pressures and flame speeds in the pipe, unless early vent deployment mitigates the explosion severity. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 45 Fig. 6. Effect of ignition location (front; center; back) on pressure (top) and flame arrival (bottom). Experiments (grey lines and markers) and model predictions (red lines). Dotted lines mark static vent deployment pressure for each test. L P = 9.75 m, D P = 0.4 m, X = 3.0%. 5. Discussion of predictive model capability Experiment-model comp arisons in Sec. 4 demonstrated that the model predicts time -resolved vessel pressure and flame progress well and captures the critical effects of geomet ry and mixture para meters. For broader validation, a total of 20 experiments with differe nt combinations of para meters are us ed in the following for comparison against model results. Figure 7 summarizes the various combinations of para meters covered by t hese tests. Fig. 7. Matrix of parameters used for model validation in Fig. 8. Marker color indicates pipe diameter; marker shape indicates pipe length. Figure 8, upper left panel, compares expe rimental and modeled flame -arrival times (relative to ignition) for all fl ame de tector posi tions along the pipe. Mode l p redictions capture all experimental results within an accuracy of ± 10%, which is comparable to the scatter of the experime ntal results. The next step is to evaluate the predictive capability of the model from the perspe ctive of explosion isolation. Active explosion isolation systems typica lly activate b ased on a pre ssure measurement using a set pressure threshold. Once the vessel pressure reaches this threshold, the system control unit triggers the explosion isolation d evice . Once the device is triggered, it takes a finite time for the device Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 46 to form a n effective flam e ba rrier, i.e., to close the pipe cross-section if the device is a mechanical valve, or to establish a sufficient concentration o f suppressant in the pipe in the case of chemica l isolation systems. The explosion dynamics model can be used to predict t he time of flame arrival, relative to the time of crossing a set vessel-pressure threshold, at various locations along the pipe. These arrival times can then be c ompared to the time a specific isolation system re quires to form a barrier, to determine the minimum installation distanc e for the specific system and use case. Fig. 8. Comparison of experimental and modeled flame-arrival times along the pipe. Top left: time from ignition; Top right: time from reaching p = 3.5 kPa; Bottom le ft: time from reaching p = 5.0 kPa; Bottom right: time from reaching p = 10.0 kPa. Figure 8, upper right and lower p anels, show comp arisons of experimental and modeled flame-arrival times along the pipe, relative to the time a given ac tivation pressure threshold is reac hed in the vess el in eac h test. Typi cal activ ation pressure thresholds are evalu ated, i.e., 3.5 kP a, 5.0 kPa , and 10.0 kP a. For each plot, the mean deviation, m , betw een model and experiments (soli d line), and the stand ard deviation, s (dashed lines ), are given. Th ese comp arisons show good agreement be tween the model predictions and experimental results throughout the entire range o f parameters studied . This indicat es that the model accurately captures both the early phase of flame propa gation within the vessel and it s associated pressure rise, which would activate th e isolation s ystem, as well as the later high -speed flame propagation through the pipe . The non-zero mean a nd standard dev iations can be considered for practica l applications as part of safe ty margins for installation distances. The sli ght over- prediction of flame -arrival time by 7 – 12 ms relative to pressure-based acti vation, is fairly consistent between differe nt configurations. This may be a result of the simplified tre atment for flame- entrance Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 47 into the pipe, whe re forward-jetting of the fla me along the ce nterline of the pipe is not considered. Note, that arrival times for some cases and pipe locations are negative in both experiments and model, which shows that pressure-ba sed detection cannot detect all explosions before the flames enter the pipe. This is particularly relevant for front-ignition cases, low mixture reacti vities, and lar ge ratios o f pipe diameter to vessel volume. Optical flame detection at the pipe entrance or large installation distances may be neede d to resolve such situations. For explosion isolation s ystems activating by mea ns other than pre ssure detec tion, the model can provide similar predictions. Active systems may use flame detectors at the pipe entranc e, which c an be sim ulated based on th e model predictions fo r flame progr ess. P assive sy stems such as flap or float valves activa te based on flow velocity in the pipe, which can also be predicted by the model. While these results can be used to determine the minim um installation di stances, it is important to mention that installation dist ance s are also bou nded by upper limits t o prevent deflagration- to - detonation transition (DDT) and excessive ov er-pressure. Model predictions, experimental results, and literature studies will be used in future wor k to establish such upper limit s, using maximum permissible values for f la me spee d and pressure buil d-up that elevate the likelihood of DDT. 6. Conclusions and future work This study validated a physics -based mod el for e xplosion dynamics in vented vessel-pipe system s using large-scale experimental data. Large-scale explosion tests were per formed in an 8 -m 3 vessel that provide a wide range of validation cases and varied parameters that govern th e explosion dynamics, including: pipe lengths of 4.88 m and 9.75 m; pipe diameters of 0.1 m, 0.2 m, and 0.4 m; ignition locations at the center of the vessel, at the back -wall, and at the pipe entrance; and propane conce ntrations in air of 2.8% ( Φ = 0.69 ), 3.0% ( Φ = 0.74), and 3.5% ( Φ = 0.86). It was shown that the model predicts pressure transients closely and captures flame -arrival times along the pipe at an accuracy better th an ±10%, ov er the full range of parameters studied. The model can be used to predict the time available for an explosion isol ation system to actuate and form a flame barrier, which can be compared with the activation dynamics of the sp ecific system. The explosio n dynamics model can support analyses that determine whether isolation systems will perform as needed in applications, i ncluding the prediction of minimum installation distances, while reducing the number of tests needed to determine the performance limitations of isolation systems. Additional work is need ed to develop comprehensive installation guidance for various types of explosion isolation systems. Practical test methods are currently being developed that determine the isolation system pressure-resistance a nd flame-bar rier e fficacy, and measure activation dynamics. Acknowledgements This work was funded entirely by FM Global and performed within the framework of the F M Global Strategic Research Program on Explosions and Mater ial Reactivity. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 48 Reference s Bartknecht, W. ( 2012). E xplosions: course, prevention, protection . Springer Science & Business Media. Bauwens, C. R., Bergthorson, J. M., & Dorofe ev, S. B. (201 5). Experimental study of sphe rical- flame acce leration mechanism s in large-scale propane-air flames. Proceedings of the Combustion Institute , 35 , 2059 – 2066. Boeck, L.R., Bauwens, C.R., & Dorofeev, S.B. (2018). A physics-based model for explosions in vented vessel-pipe systems. 12 th ISHPMIE, Kansas City, USA. Boeck, L.R., Bauwens, C.R., & Dorofeev, S.B. (2019). Propane-air explosions i n an 8-m 3 vessel with small vents: Dynamics of pressure and e xternal flame-jet. 9 th ISFEH, St. Petersburg, Russia. Eckhoff, R. K. ( 2016). E xplosion hazards in the process industries . Gulf Professional Publishing. EN 15089 (2009). Explosion Isolation Systems. Ferrara, G., W illacy, S. K., Phylaktou, H. N., Andrews, G. E., Di Benedetto, A., S alza no, E., & Russo, G. (2008). Venting of gas explosion through relief ducts: Interac tion between internal and e xternal explosions. Journal of Hazardous Materials , 155, 358-368. Goodwin, D. G., Moffat, H. K., & Speth, R. L. (2017). Cantera: An object-oriented software toolkit for chemical kinetic s, thermodynamics, and transport pr ocesses, v ersion 2.3 . Hattwig, M., & Steen, H. eds. (2008) . Handboo k of explosion preve ntion and protection . John Wiley & Sons. Lowry, W., de Vries, J., Krejci, M., Peter sen, E.L., Serinyel, Z., Metcalfe, W., Curran, & H., Bourque, G. (2010). Laminar flame speed measurements and modeling of pre alkanes and alkane blends a t elevated pressures. Journal of Engineering for Gas Turbines and Power , 133 (9) 091501. Molkov, V. V. (1994). Venting of defla grations: Dynamics of the process in systems with a duct and receiver . Fire Safety Science , 4, 1245-1254. Ponizy, B., & Leyer, J. C. (1999). Flame dynamics in a vented vessel c onnected to a duct: 1. Mechanism of ve ssel -duct interac tion. Combustion and Flame , 116, 259-27 1. Smith, G. P., Golden, D. M., Fre nklach, M., Moriarty, N. W., Eiteneer, B., Goldenberg, M., Bowman, T., Hanson, R. K., Song, S., Gardiner Jr., W. C., Lissianski, V. V. & Qin, Z. GRI- Mech 3.0. http://www.me.berke ley.edu/gri_mech/ . Taveau, J. (2017). Dust explosion propagation and isolation. Journal of L oss Pre vention in the Process Industries , 48, 320 – 330. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 49 13th International Symposium on Hazards, Prevention, and Mitigat ion of Industrial Explosions Braunschweig, GERMANY – July 27-31, 2020 CEQAT-DG HS interl aboratory tes ts for chemical safety: V alidation of la boratory tes t metho ds by determinin g the measureme nt uncert ainty and probability of incorrect class ificatio n including s o-called “Shark pr ofiles” Peter Lüth a , Steffen Uhlig b , Kirstin Frost c , Marcus Malow a , He ike Michael-Schulz a , Martin Schmidt a & Sabine Zakel d a Bundesanstal t für Mate rialfo rschung und -prüfung (B AM), Berlin, Germ any b QuoData G mbH, Berlin, Germany c QuoData G mbH, Dr esden, Germany d Physikalisch-Te chnische Bundesanstalt (PTB ), Brau nschweig, Ge rmany E-mail: peter.l ueth@bam .de Abstract Laboratory t est results a re of vital importance for correctly classifying and labelling chemica ls as “ hazardous ” as de fined in the UN Globally Harmonize d S ystem (GHS) / EC CLP Regulation or as “ dangerous goods ” as d efined in the UN Recommendations on the Transport of Dangerous Goods . Interlaboratory tests play a decisive role in assessing the reliability of laboratory test results. Interlaboratory tests pe rformed over the last 1 0 years have examined different laboratory test methods. After analysing the results of these int erla boratory tests, the following conclusions can be drawn: 1. There is a need for improvement and validation fo r all labor atory test methods examined. 2. To avoid any discr epancy concerning the classification and labelling of chemicals , the use o f validated laboratory test methods should be state of the art, with the result s accompanied by the measurement uncerta inty and (if applicable) the pr obability of incorre ct classification. This paper addresses the probability of correct/incorrect classification (for example, as dangerous goods) on the basis of the measurement deviation obtained fr om interla boratory tests performe d by the Centre for quality assurance for testing of dangerous goods and hazardous substance s (CEQAT- DGHS) to validate laboratory test methods. This paper outlines typical results (e.g. so -called “Shark profiles” – the probability of incorrect classification as a function of the true value estimated from interlaboratory test data ) as well as general conclusions and steps to be taken to guarantee that laboratory test results are fit for purpose and of high quality. Keywords: interlaboratory test, validation, measureme nt uncertainty, incorrect classification 1. Introduction Accidents such as explosions in chemica l plan ts and fir es on dangerous goods vesse ls can be caused in several wa ys. Preve ntion starts in the laboratory, where chemicals are tested for their hazardous properties in order to be able to assess the risks involved in their handling. F or this purpose, laboratory test methods have be en d eveloped and published (see e.g. laboratory test m ethods published by the European Union (2008) and by the United Nations (2019 )) ; these methods are currently a pplied throughout the world. La boratory test results (amongst others) are used to corre ctly classify and label Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 50 chemicals as “ hazardous ” as defined in the UN Globally Harmonized System (GHS) / EC CLP Regulation or as “ dangerous goods ” as d efined in the UN Recommendations on the Transport of Dangerous G oods. Safety experts, manufacturers, suppliers, importers, employers and consumers must be able to rely on the validit y of sa fety-relate d labora tory test met hods and on the ac curacy of laboratory test re sults and assessments. I nterlaboratory tests play a decisi ve role in assessing the reliability of laborator y test results. Pa rticipation in interla boratory tests is a crucial element of t he quality assurance of laboratories; for this reason, it is explicitly recommended in ISO/IEC 17025 (2017) ( assuming suc h interlaboratory tests are available). Interlaboratory tests are used in laboratory test method development a nd validation and ca n be used to determine mea surement un certa inties (Hässelbarth, 2004; ISO 21748:2017-04, 2017). Over the past 10 years, the Bundesanstalt für Materialforschung und - prüfung (BAM) and th e Physikalisch-Technische Bundesanstalt (PTB) in cooperation with QuoData GmbH have carried out interlaboratory comparisons to evaluate various laboratory test methods (Lueth et al., 2019). Significant differences betwe en the results of the participating laboratories were observed in all interlaboratory tests. The deviations in the laboratory test results were caused not only b y malfunctions in the laboratory equipment and labo ratory faults but also by deficiencies in the diff ere nt laboratory test methods ( see interlaboratory test r eports of the Centre for qu ality assurance for testing of dangerous goods and hazardous substances (CEQAT-DGHS), www.ceqat-dghs.bam.de ). After a nalysing the re sults of these interlaboratory tests, the following conclusions ca n be drawn: • To avoid any discrepancies in the classification and labelling of chemica ls, the use of validated laboratory t est methods should be state of the art, with the results accompanied by the measurement uncerta inty and (if applicable) the pr obability of incorre ct classification. • There is a need for improvement and validation for all laboratory test m ethods examined. Thus, interlaboratory te sts should initially foc us on the development, improvement and validation of the laboratory test methods (including the deter mination of the measurement uncertainty) and not on proficie ncy tests. • Laboratory management and the practical execution of laboratory tests need to be im proved in many laboratories. • The term “ experience of the examiner ” must be seen critically; long-term experience involving many laboratory tests does not necessarily guarantee correct results. However, it is not currently clear whether the measurement uncertainty is sufficiently considered in practice when a chemical is classified as a pa rticular packing group ( PG ) or as a hazard class based on the measurement results combined with the mea surement uncertainties. Statistical methods and graphical tools must th ere fore b e developed which will clearly define how the me asurement uncertainty should be used in prac tice. This paper addr esses so -calle d “Shark profiles”, graphica l indicato rs of the probability of incor rect classification of dangerous goods and haz ardous substances on the basis of the measurement uncertainty obtained from special interlaboratory tests performed by CEQ AT -DGHS to validate laboratory t est methods . “Shark pro files” we re de veloped to character i ze the quality and suitability of the laboratory test method(s) with respect to the laboratory test objective, i.e. the classifica tion of chemicals based on specific classification criteria ( Antoni et al., 2010, Antoni et al., 2011 and Kunath et a l., 2011). This objective is particularly important whe n there are several classification level s (e .g. packing groups PG or hazard c lasses ) and th e question of the reliability of the classific ation arises. The principle of the c alculation of “Shark profiles” for interlaboratory tests is explaine d by means of quantitative laboratory t est results. This paper outlines typical results as we ll as general conclusions and steps to be take n to guarantee that laboratory test results are fit for purp ose and of high quality. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 51 2. Experiments 2.1 CEQAT-DGHS interlaboratory tests and minimum number of participating laboratories As part of the CEQAT-DGHS programme, int erlaboratory tests were carried out on various laboratory test methods and either qualitative or qu antitative test results were obtained and evaluated. An overview of the l aboratory test methods curren tly listed in the CEQAT- DGHS interlaboratory tes t programme is shown in Fig. 1. This figure also indicates interlaboratory t ests already performed b y CEQAT-DGHS as well as the number of laboratories interested in taking part in these int erla boratory tests. Fig. 1. Laboratory test methods listed in the CEQAT-DGHS interlaboratory test programme, number of laboratories with interest in participation in CEQAT -DGHS interlaboratory tests and interlaboratory tests performed by CEQAT-DGHS since 2007 (RR = interlaboratory te st ) Since CEQAT-DGHS was founded in 2007, th e number of laboratorie s interested in taking part i n interlaboratory tests has steadily increa sed to about 98. The minimum number of participating laboratories required for statistical ly meaningful interlaboratory tests (i.e. for statisti cal reliability ) has been met for almost all laboratory test metho ds. Hence, int erlaboratory tests can now be carried out for almost all laboratory test methods listed in in Fig. 1. 2.1.1 Data verification (inspection upon receipt) in interlaboratory tests for method validation A special feature of the CEQAT-DGHS interlaboratory comparison programme for method validation is that the data supplied by the laboratories are subjected to a strict verification before being evaluated. This is necessary to ensure data quality. The specific re view o f th e data submitted by the laboratories includes the following points (Antoni et al., 2010, Antoni et al., 2011, Kunath et a l., 2011, Frost et al., 2016 and Lueth et al., 2019) : • Completeness of the data – check for missing data • Conformity – check for irregular deviations from the laboratory testing method and/ or from the interlabora tory test instructions • Plausibility – check for obvious incorre ctness of the data submitted (e.g. distorted data) • Consistency – check the c orrectness of the values in the data input for m submitted (e.g. by comparing them to raw data) 0 5 10 15 20 25 30 35 40 N u m ber o f l aborat ori es T es t m eth od 2007 2010 2014 2020 ☺ = RR co m pl ete d  = RR current ☺ = RR i n p re para t i o n ☺ ☺ ☺ ☺ ☺ ☺ ☺ ☺ ☺ ☺ ☺ ☺ ☺  ☺ ☺ Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 52 The data were verified/validated by different exp erts independently directly after rece iving the dat a from the laboratory and before starting the sta tistical analysis. If necessary and possible, faulty data were corrected after consulting the respective laboratory; laboratories were also asked to submit any missing data. A statistical evaluation including the determination of the measurement uncertainty and the proba bility of incorre ct c lassification with corre sponding “S hark profiles ” was carr ied out on this specially validated data. The test results constituted an adequate basis for the statistical evaluation and for reliable c onclusions. 2.2 Probability of incorrect classification and corresponding “Shark profiles” The probability of incorrect classification of a tested substance , as defined by the European Union (2008) and by the United Nations (2019), depends on the measurement uncertainty of the test result(s) and the difference be tween the test result and the classifica tion criterion. However, the pe rcentage of laboratories which have incorrectly classified the test samples may be only a very rough estimate of the ac tual probability of the incorrect classification. There fore, statistical methods a nd graphica l indicators must be developed which can provide the probability of incorrect classification re gardless of the substance and cha racterize the quality and t he suitability of the laboratory test method(s) with regard to the test objectiv e (i.e. the classification of chemicals based on spe cial classification criteria). For this reason, the concept of the proba bility of i ncorr ect classification and the visual shark profiles explained below w ere developed. The method s and development rel evant to this topic are demonstrated below using several examples. During the d evelopment of the “Shark p rofiles” the focus was on laboratory test methods whose quantitative test results are used to classify substances bas ed on cl assifica tion criteria defined by the European Union (2008) and the United Nations (2019). The interlaboratory tests evaluated the laboratory test methods listed in Table 1: Table 1 : Quantitative laboratory test methods of the CEQAT-DGHS interlaboratory tests, year in which each interlaboratory test was performed and name of report publi shed Laborato ry test met hod Year o f the interlabora tory test Report o f the interlabora tory test UN Test O.1 “T est for ox idizing solids” 2005 -2006 Anton i et al., (2010) UN Test N.5 “ Substances which in contact with wate r emit flamm able gases” /EC A.1 2 “Flam mability (contact with water)” 2007 Kunath et al., (2011) UN Test O.2 “T est for ox idizing liquids” / EC A.21 “Ox idizing Proper tie s (Liquid s)” 2009 -2010 Anton i et al., (2011) The labor atory test methods shown in Ta ble 1 we re ex amined in interlaboratory tests for the improvement and v alidation of laboratory test methods (i.e. there were n o proficie ncy tests) . For reasons of simplifica tion and better re adability, an alterna tive packing group designation – PG 1, PG 2 and PG 3 – is used in the following inst ead of the legally correct designation – PG I, PG II and PG III. 2.2.1 Measurement uncertainty of laboratory test methods The measu rement un certainty of the labo ratory test methods can be d etermined eff iciently vi a interlaboratory tests for method validation (Hä sselbarth, 2004, ISO, 2017-04), and can be expressed as shown in eq (1). ‘Labora tory result’ = ‘M ea surement result’ ± U, where U denotes the expanded measurement uncertainty U = k * u with u = s R (1) Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 53 The factor k c orresponds to the coverage factor k defined in GUM (Guide to Uncertainty in Measurement, Joint Comm ittee for Guides in Metrology, 2008) and the factor s R denotes the reproduc ibility standard deviation obtained in the interlaboratory tests for method validation. The reproducibility stan dard deviation s R is determined by means of standardized procedures and statistical methods (Antoni et al., 2011) . Thes e st andardized statistical cal culations are commonly used by statistical expert s in proficie ncy testing and therefore do not require any further explanation at this point. However, detailed information on the calculation methods can be found in the respective interlaboratory test reports (Antoni et al., 2010, Antoni et al., 2011, Kunath et al., 2011). 2.2.2 Calculation of the probability of incorre ct classification and “Shark profiles” The procedure used to calculate the probability of incorrect cla ssification and the “Shark profiles” is similar to statistical testing with a null hypothesis (e.g. the test sample b elongs to PG 1) and with an alternative hypothesis (e.g. the test sample belong s to PG 2). In g eneral, th e objective of a statistical test is to kee p the p robability of a false positive re sult (rejection of null hypothesis and acceptance o f alternative hypothesis) below a certain limit (statistica l significance level), whereas th e probability of the false n egative error ( acceptance of null hypothesis) can only be controlled with great difficulty . These probabilities a lways depend on the “true” value of the measurand (e.g. combustion time) for the specific test sample. Using int erla boratory test data, the percentage of laboratories which incorrectly classified test samples provides only a very rough estimate of the ac tua l probability of incorrec t classification. A more reliable estimation of this probability can be obta ined using the quantitative properties o f the data. Here, the probability of incorrect classification of the test samples is derived by evaluating the ratios between the laboratory results of the test sa mple a n d the classification criterion (e.g. a firmly defined threshold value taken from a legal regulation or a l abora tory test value obt ained by testing a reference substance whic h characterizes the c lassification criterion (threshold value)). The proce dure is as follows (and is initially carried out for each PG separately) : Step 1: Calculation of the cha racteristic statistical values and determination of the true va lue Based on interlaboratory test data, the mea n value and the standa rd deviation of the laboratory- specific ra tios between the test results of the test sample and those of the reference sample are determined using the robust Q/Hampel method (desc ribed e.g. in I SO 13528:2015). Thus, with the interlaboratory test data, a robust estimate of the “true” ratio value (which is typically not exactly known) will be obtained as well as the respe ctive reproducibility standard deviation regarding th e ratios. The advant age of using robust methods is that no special outlier test s are required. Especially the Hampel estimator is a weighted arithmetic mean, with lower weights for the outlying values. Obvious “outliers” have a weight of zero and no influence on the mean value. Step 2: Calculation of the proba bility of incorrect classification B ased on the “true” rati o value (i.e. the robust mean value), and the corresponding reproducibility standard deviation obtain ed in Step 1 above), the expected reproducibility standard deviation for the ratio between the test results of an arbitrary test sample and those of the refere nce sample ( which character izes the classifi cation li mit) can be estimated. Based on thi s estimation, the probability of the test sample b eing classified in a P G for more dangerous substances c an be calculated as a function of the ratio of the “ true” value of the mea surand (see F ig. 2). Here, the “ decision direc tion ” (i.e. from a PG for less da ngerous substances to a PG for more da ngerous substances ) must be taken int o account. F or the packaging groups, the orde r fr om PG 3 to P G 2 to P G 1 means a n increase in the dangerousne ss of th e substance. Once a certain classification threshold value (criterion fo r classification) has been reac hed, the substance is to be classifie d as a de fined PG and then remains in this P G until the next higher classification criterion (i.e. for more dangerous substances) has been reac hed. Once the next higher c riterion value (for the next PG for more dangerous substances ) has Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 54 resources available at the C EQAT-DGHS competence cent re limit the number of interlaboratory tests that can be offered to approximately one pe r year. Interested l abora tories can obtain detailed information and register for i nterlaboratory tests at the CEQAT-DGHS website (www.ceqa t-dghs.b am.de) . Acknowledgements The authors gratefully acknowledge the personnel and financial support provided by the Bundesanstalt für Materi alforsc hung und -prüf ung (BAM), Physikalisch- Technische Bundesanstalt (PTB), QuoData GmbH and by the laboratories participating in the interlaboratory tests. Reference s Antoni S., Kunath K., Lüth P., Schlage R., S imon K., Uhlig S., W ildner W ., Zimmerma nn C . (2010). Evaluation of the interlaboratory test on the method UN test O.1 'Test fo r oxidizing solids' with sodium perborate monohydrate 2005 / 06 Final r eport, BAM, Berlin. ISBN 978-3-9814281-2-4 https://opus4.kobv.de/opus4-bam/frontdoor/index/index/docI d/25091 . Antoni, S., Kunath, K., Lüth, P., Sim on, K., Uhlig, S. (2011). Evaluation of the int erlaboratory tes t on the method UN O.2 / EC A.21 Test for oxidizing liquids 2009 - 2010 Final report , BAM, Berlin, ISBN 978-3-9814634-0-8, https://opus4.kobv.de/opus4-bam/frontdoor/index/index/docId/25090 . European Union (2008). Comm ission Regulation (EC) No 440/2008 of 30 May 2008 laying down test methods pursuant to Regulation (EC) No 1907/2006 of the Eur opean Parlia ment and of the Council on the Registrati on, Evaluation, Authorisation and R estriction o f Chemicals (REACH), OJ L 142, 31.5.2008, p.1. Frost K., Lüth P., Schmidt M., Sim on K., Uhlig S . (2016). Evaluation of the int erla boratory test 2015- 2016 on the method DIN EN 15188:2007 “Determination of the spontaneous ignition behaviour of dust accumulations” Final report, BAM, Berlin . I SBN 978-3-9818270-0-2, https://opus4.kobv.de/opus4-bam/frontdoor/index/index/docI d/38734. Hässelbarth W. (2004). BAM-Leitfaden zur Erm ittlung von Messunsicherheiten bei quantitativen Prüferge bnissen. Forsch ungsbericht 266. BAM, Berlin. I SBN 3-86509-212 -8. ISO 13528:2015-08 Statistical methods for use in proficiency testing by interlaboratory comparison, Internationa l Organization for Standardization, Geneva. ISO/I EC 17025:2017-11 Gene ral requirements for the competence of testing and calibra tion laboratories, I nternational O rganiza tion for Standardization, Geneva. ISO 21748:2017 -04 Gui dance for th e use of repeatability, re producibility and trueness estimates in measurement uncerta inty evaluation, I nternational Organiza tion for Standardization, Geneva. Joint Comm ittee for Guid es in Metrology (2008) Evaluation of me asurement data — Guide to the expression of uncertain ty in measurement (GUM 1995 with minor corrections), B I PM < www.bipm.org/en/publications/guides/gum.html >, accessed 06.02.2020. Kunath K., Lüth P., Uhlig S. (2011). Interlaboratory test on the method UN test N.5 / EC A.12 'Substances which, in contact with water , emit flamm able ga ses' 2007 Short report, BAM, Berlin . ISBN 978-3-9814634-1-5, https://opus4.kobv.de/opus4-bam/frontdoor/index/index/docId/25094 . Lueth P., Frost K., Kurth L., Malow M., Michael-Schulz H., Schmidt M., Schulte P., Uhlig S., Za kel S. (2019). CE QAT - DGHS Inter laboratory Test Programme for Chemica l Safety - Need o f Test Methods Validation. CET Chemical Engineering Transactions , 77: 1-6. DOI: 10.3303/CET1977001 . United Nations (2019 ). Manual of Tests and Criteria, Seventh revised edition, United Nations , New York and Geneva, < www.unece.org/trans/danger/publi/manual/rev7/manrev7-files_e.html >, accessed 06.02.2020 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 61 13th International Symposium on Hazards, Prevention, and Mitigat ion of Industrial Explosions Braunschweig, GERMANY – July 27-31, 2020 CEQAT-DG HS Interla boratory tes ts for chemical safety – A ne w gravimetric procedure for the gas flo w measurement fo r fla mmable and toxi c gases Peter Lüth a , Marc us Malow a a Bundesanstal t für Mate rialfo rschung und – prüfung (BAM), Ber lin, Germa ny E-mail: marcus.malow@ba m.de Abstract Testing of hazardous m aterials and eva luating their ha zardous properti es concerning tra nsport, handling or use is essential for the prevention of incidents. For this purpose, test methods h ave been developed and published that are used worldwide today (European Union, 2008, United Na tions, 2019). For the evaluation of test results their corre ct measurement is of im portance. On basis of the interlaboratory tests carried out by BAM and P T B within the framework of the CEQAT -DGHS in the last years, it is shown that there is a need for im provement in all the test methods examined so fa r. In addition to the interlaboratory tests furth er quality measures are mandatory (ISO, 2017). For example, methods for verifying the test equipment used in the labora tories should be developed. The deve lopment of a verification method is demonstrated using the test method UN Test N.5 as an example. This test method is used to evaluate subs tance s which in c ontact with water emit flammable gases. The basic principle of the verification method is de monstrated by Lüth et al. (2019). Requirements and diff iculties during this development are discussed. This test procedure UN Test N.5 has now been modified at BAM so that it is possible to measure very small or larg e amo unts of both flammabl e and/or toxic g ases ov er a long period of time in a validated and verified te st apparatus, utilizing th e principle of a gas colle ctor. This allows us to determine slow a s we ll as fa st ga s evolution rates. The determina tion of the evolution rates (e.g. total gas amount or flow rate) of toxic gases is of special interest because of the ongoing discussion how to evaluate and qua ntify toxic gases which are formed from a substanc e due to c ontact with wa ter. Up today no validated t est procedure for the m easure ment of evolutio n rates for toxic g ases is described. I t is shown that the newly modified test method could help to solve this problem. Keywords: hazardous materials, transport regulations, flammable gases, toxic gases 1. Introduction The gas flow m ea surement procedure for substances which emit flammable gases in contact wit h water is described in the UN Test method N.5 (United Nations, 2019). The experiment involves the determination of the maximum gas evolution rate (GER) by means of a conical flask ( Erlenmeye r flask) and a dropping funnel. The volume of the g as produced when the chemical come s into contact with water should be m easured at appropriate time intervals by suitable measures. However, more precise in formation on measurement technique, how the GER should be determined, and corre sponding qu ality requirements are not spe cified in th is UN method. On basis of the c alculated maximum GE R, the dange rous goods classification of the chemica ls is carried out ac cording to the criteria listed in (United Nations, 2019). On the c riteria listed in Table 1, it can be stated that a relatively large measuring range must b e covere d by the test app aratus when determining the GER which is needed for the determination of the corre sponding Packing Group (PG). It is therefore to be assumed that for this large measuring Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 62 range a r elatively high effort is required in the control of the test quality r ealized with the respective testing apparatus. Table 1 : Criteria for Classification in Dangerous Goods Division 4.3 on the basis of the test results in accordance with UN Test N.5 (maximum G ER), [PG] = Packing Group (United Nations, 2019) Maximum gas evolution ra te Dangero us Goods Division PG  10 l / (kg *min) 4.3 I  20 l / (kg *h) and not  10 l / (kg *min) 4.3 II  1 l / ( kg*h) an d neither  20 l / ( kg*h) n or  10 l / (kg *min) 4.3 III ≤ 1 l / (kg * h) not 4 .3 No For exa mple, the volum e flows ("minute flow" and "hour flow") resulting in the respective PG decision criterion for a ty pical sample quantity of 10 g for the two evaluation intervals (1 mi nute or 1 hour) are shown in Table 2. Table 2 : Criteria for Classification in Dangerous Goods Division 4.3 on the basis of the test results in accordance with UN Test N.5 (maximum G ER), [PG] = Packing Group (United Nations, 2019) PG Gas evo lution rate PG - Decisions criterion Sample weig ht [g] "Minute-F low" per sample weig ht [ml/m in] "Hour-Flo w" per sample weig ht [m l/h] I 10 l / (kg*min) 10 100. 00 6 000 II 20 l / (kg* h) 10 3.33 2 00 III 1 l / (k g* h) 10 0.17 10 An essential question for the acceptance and the t rustworthiness of the test re sults according to UN test N.5 is the question of the traceability of the test r esults ( I SO, 2017 ). With a gravimetric test method, the traceability of the results can be de monstrated very easily, especially in the d aily laboratory routine. For th is reason and because of the simplicity of the procedure, the UN Test N.5 at BAM is carried out usi ng a gravimetri c test apparatus. The results of the ve rification of this gravimetric apparatus are explained in (Lüth et al., 2020) and are based on a gravimetric and thus easily traceable verifica tion process. 1.1 Gas evolution rate and total amount of gas evolution The GER is an important safety-related parameter for the risk assessment and the selection of protective me asure s. Acc ording to test method U N N.5, the maximum GER is a measure of the risk of ignition and is us ed to classify a substance in class 4.3. However, it can b e assumed that th e maximum GER determi ned over a maximum of 5 days according to UN Test N.5 is no t a safe ty - relevant parameter in all cases to p revent an a ccident (e.g. long -t erm transport in the container with the formation of an explosive and / or ga s concentration dangerous for inhalation). The chemical reaction usually does not run smoothly. Thus, t he gas evolution rate depends on the duration of the chemical re action with w ater. I t has to be considered that in some case s the maximum activity of the che mical reaction is only reached after 5 days or even more. However, i t should be noted that a low ga s formation rate is no guarantee for a safe situation a t the workplace or during transport. Even a small gas evolut ion rate with a correspondingly long duration Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 63 of re action with water can lead to a dangerous ga s conc entration. The following factors can pl ay a decisive role here: air exchange r ate, ratio of the amount of substance to the volume of the room , chemical r eac tion kinetics / rea ction course, reaction time, type of water (e.g. freshwater, seawater). The lower the air ex change rate a nd the closer the ratio of substance amount to the ro om volume, the faster a dangerous gas c oncentration is reac hed. This is problematic in terms of safety, e.g. during the long transport of reactive subst anc es in containers or in ship holds and other relatively small storage rooms. Therefore, in addition to the GER (as a limited para meter for the kinetics, velocity of the ch emical reac tion process), the a bsolut e amount of gas produced, and the duration of the re action must also be considered . The absolute amount of evolved gas can also be an imp ortant parameter for ris k assessment and for the s election of protective measures. A chemical mi xture with only a very small proportion, but which is a very strongly water- reactive chemical substan ce (with a high chemical reac tion rate / kinetics with water), can e.g. show a very high maximum gas evolution rate. However, due to the small amount of substance in the mixture, the tot al amount of gas will tend not to be as large compared to anoth er che mical mi xture which could react completel y with water, even if thi s other ch emica l mixture would react with wa ter much more slowly. The amount of gas is also a question of the duration o f exposure, i. e. how long the che mically reactive mixture has been in contact with water. In case of co ntact with moisture and water during long-term storage or tr ansport (e.g. in maritime traffic over seve ral weeks), the question of the tot al amount of gas generated could therefore be very important and possibly even more important than the gas evolution rate. From the B AM's perspective, both pa rameters (the maximum gas evolut ion rate and the tot al amount of gas) should therefore be determined f or the risk assessment and the de finition of protective mea sures. This might especially be importa nt for toxic gas release. Taking these aspects into acc ount, we have therefore modified the BAM test system for the UN Test N.5 syst em, i.e. with the help of a solenoid valve the system automatica lly de gasses the measuring system de pending on the filling level of the measuring container. The mo de of operation and some exemplary results are explained below. 2. Experimental 2.1 Modification of the testing ap paratus to d ete rminate both the gas evolution rate and the total amount of gas evolution As discussed above , t he GER at BAM is deter mined using a gravimetric ap proach as described in the following. The basic set -up continues to meet th e requirements of UN T est N.5 ( United Nations, 2019). The functional scheme is il lustrated in F ig. 1. The reaction chamber i s a closed system without direct cont act to atmosphere and basically c onsists of a conic al flask (1) and a d ropping funnel (2) . The evolved gas in the conical flask pre ssurizes the water column in the gas-water-displacement tank (4). This water column is displaced vi a the flexible connecting hose (5) to the water colle ction tank (6) (open to atmosphere ), located on a balance (7). Essentially the evolved gas is detected as the corre sponding w eight increase on the balance and can be processed digita lly. One drawback of thi s system is the discontinue d measuring ra nge. Th at mea ns if all water is displaced into the water collection tank no additional gas evolution can b e r ec orded. Th erefore, a modi fica tion of this set -up was implemented, i.e. a solenoid valve (3) wa s introduced into the set-up which opens automatically when a preset weight on t he balan ce is reached and degasses the system. Aft er deg assing and reaching a certain mass lev el on the balance, the valve automatically c loses. H ereby the devic e works like a continuous gas collector, despite the short degassing intervals that must be switched on automatically when a maximum fill level in the gra vimetric container is reache d. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 64 Fig. 1. Functional scheme of the modified gravimetric method and photo of the gravimetric test apparatus and the additional solenoid valve (3) used in the BAM for the determination of the gas flow according the UN Test N.5 method with: 1 = Erlenmeyer flask with the che mical test sample, 2 = dropping funnel, 3 = solenoid valve 4 = gas-water-displacement tank, 5 = connecting hose, 6 = water-collection tank and 7 = balance with coupling to a computer Table 3 summariz es the main parts and important para meters of the modified set-up as shown in Fig. 1. The changes in weight are recorded every 12 seconds using a computer. For the s et -up standar d laboratory glassware was use d. The connecting hose (4) was a flexible PVC hose. The balance was either a Mettler Toledo model B 2002-S or a model ME2992T. The water volume in the system is about 2 l. Th e solenoid valve (one -way v alve) aut omatically degasses the s ystem when the maximum load on the balance is reached. Th e degassing stops a nd the system is closed again when th e lower weight limit on the balance is reac hed. Based on the gravimetric measurement results, the GER at each ti me point and the maximum is determined. The GER was calculated using MS Excel. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 65 Table 3 : Special information abo ut the device s according to Fig. 1 and operating parameters Number a ccording to Fig . 1 Device Description 1 Erlenmey er flask Holds the chemical test sub stance 2 Dropp ing funnel By drop ping the water on to the chemical substance, the reaction is initiated 3 Solenoid valve Used for short deg assing the system in case of maximum load o n the balance is rea ched 4 G as -water-d isplacement tank Cylindrical* g as-water-d isplacement tank (may also be filled with oth er appropriate liqu ids) 5 Connectin g hose Flexible jun ction between the two tank s 6 Water collection tank Cylindrical* water collection tan k (may also be filled with other appropriate liquid s) 7 Balance Mettler Toled o model B 2002 -S or a m odel ME2992 T with measuring range up to 2 kg with a resolution with 2 decimal places with d ynamic weig hing * … the cylindrica l shape allows gas volume correction by calculating the pressure rise in the gas phase, which arises fr om the changes in the liquid height leve ls (height differe nces) in the cont ainers 4 and 6 according to Fig.1. 2.2 Verification of the modified testing app aratus The modi fied testing appara tus with the solenoid valve can be verified using t he newly developed verifica tion procedure (Lüth et al., 2020). This is important in order to also chec k th e suit ability of the solenoid valve or the gas -tight seal and the gas-tightness of the test system. The experimental setup for the verifica tion of the apparatus is shown in Fig. 2. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 66 Fig. 2. Functional scheme of the ve rification apparatus (left part of scheme) and test apparatus (right part of scheme) with: 1 = water reservoir, 2 = filter, 3 = nee dle valve, 4 = rotameter, 5 = water-gas-displacement tank, 6 = balance with coupling to a computer, 7 = Erlenmeye r flask with pipe connec tion, 8 = dropping funnel, 9 = solenoi d valve; 10 = gas-water-displacement tank, 11 = connecting hose, 12 = water-collec tion tank, 13 = balance with coupling to a computer For more details on the v erifica tion procedure and results please see Lüth et al. (2020). 3. Results and discussion 3.1 Results of data acquisition and calculation of gas volume and gas evolution rate s The results of the data acquisition and calculation of the gas quantity a nd the gas evolution rates as determined with this ne wly modified set-up a re explained using a test sample, which has been examined at BAM. The sample wa s an aluminum alloy which was in powder fo rm ( < 500 µ m, i.e. 20-63 µ m) and was examined at BAM without further processing. An amount of about 10 g was weighed into the fla sk and 30 ml of water were added. The examination time is approx. 333 h, i. e. almost 15 days, i.e. the time was significa ntly longer than required by UN Test N.5. The following Fig. 3 shows the increase in mas s on the scale for the test sample with automatic degassing of the system. In principle, the device works like a continuous gas collector, despite the short degassing int erva ls that are sta rted automatically when a ce rtain fill le vel (e.g. at mass 1500 g in Fig. 3 ) in the water-collection tank is reached. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 67 Fig. 3. Typical weight vs. time diagram of the test sample measured by balance of the test apparatus with automatic degassing of the system The strictly vertical lines of the mass- time -curves corr espond to the ope ning and closing of the valve and thus to the temporary degassing of the system (see marked example in Fig. 3). 3.2 Determination of the cour se of the developing gas volume The gas volum e is determined on the basis of the mass of the medium dis place d (e.g. water), taking into account the de nsity of the re spective gravimetric medium (e.g. water) and shown in Fig. 4. Fig. 4. Gas volume vs. time diagram of the test sample measured by balance of the test apparatus with automatic degassing of the system open ing of the valve closing of the valve Degassing b y opening and closing of the valve Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 68 As shown in Fig. 3 and Fig. 4 there were 8 automatic switching operations for degassing the syst em within the testing. Each dega ssing opera tion a ccounts to 9 int erva ls of 12 se conds (da ta acquisition rate is 12 s p er d ata point), which, given the corresponding rate of increase, means a loss of approx. 3.8 g of the water displaced during this time by further gas formation (approx. 3.8 ml gas). Based on the 500 g m easuring interval, this is a loss of less than 1% and was th er efore not conside red in Fig. 4. From Fig. 4 it can be s een that the signal - to - noise ratio in relation to the tota l gas evolution is perfectly sufficient. The tot al gas evolution measured by this modi fied method can be used as a suffici ent saf ety parameter for bo th the flamm able a nd the toxic safety questions. 3.3 Determination of the GER The calculation of the gas flow rate is based on the weight differences between the respective neighboring 12-second measuring interva ls as shown in Fig. 5, left axis. Fig. 5. Left axis: Typical gas flow rate vs. time diagram of the test sample measured by balance of the test apparatus with automatic degassing of the system; right axis: Typical GER (minute volume) vs. time diagram of the test sample measured by balance of the test apparatus with automati c degassing of the system Taking into account the weight of 10 g of test su bstance, we calculated th e normalized GER shown in Fig. 5, right axis. From Fig. 5 it c an be seen that the signal- to -noise ratio in relation to the c lassification criteria for PG I (10 l / (kg*min)) is perfectly sufficient a nd even low va lues of about 0.2 l / (kg*min) can be ac curately determined whereby the experimental limit of detec tion is not reached yet. For classifica tion purposes it is also necessar y to calcula te the GER on the hour b asis also. This is illustrated in Fig. 6. Degassing b y opening and closing of the valve Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 69 Fig. 6. Typical GER (hour volume) v s. time diagr am of the test sample measured by balance of the test apparatus with automatic degassing of the system The signal- to -noise ratio with re gard to the classification criterion for PG II I of 1 l/ (kg*h) is shown graphically in Fig. 7: Fig. 7. Enlarged Fig. 6 with enlarged y-axis to evaluate the signal- to -noise in relation to the classification criterion of PG III (1 l/(kg*h) It can be stated that also t he eva luation of the GER on basis of the hour value can be performed with our modified set -up accurately. From Fig. 7 it can be seen that the signal- to -noise ratio in relation to the classification criteria for all PG is perfectly sufficient. 0 2 4 6 8 10 12 0 24 48 72 96 12 0 14 4 16 8 19 2 21 6 24 0 2 6 4 28 8 31 2 33 6 Gas e v olu t io n r a te GER [ l/( kg h ) ] T i m e [h ] Degassing b y opening and closing of the valve Degassing b y opening and closing of the valve Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 70  Hydrogen has high specific energy, about 142 and 120 MJ kg -1 for the higher heating value (HHV) and lower heating value (LHV), respectively, compared to hydrocarbons such as methane (55 and 50 MJ kg -1 ), propane (50 and 46 MJ kg -1 ), as well as diesel and jet fuel (45-46 and 43 MJ kg -1 ). However, the low density, and hence low en ergy density, implies that compressed hydrogen (GH2) is stored and transported at considerably higher pressures ( ≈ 700 bar) compared to conventional energy carriers, such as CNG ( ≈ 260 bar), LPG ( ≈ 30 bar), as well as marine gas oil, heavy fuel oil, diesel and petr ol (atmospheric pressure in the tank space). As su ch, it is not straightforward to prevent los s of containm ent. The pressu rised vessels also represent a hazard, for instance in th e event of m aterial failure or exte rnal loads from im pacts, projectiles, fires, etc. Furthermore, the low boiling point im plies that li quefied hydrogen (LH2) is stored and transported at considerably lower temperatures ( ≈ -253 o C) compared to conventional energy carriers, including LNG ( ≈ -162 o C). This implies cryogenic hazards in the event of loss of containm ent.  Hydrogen is a gas at normal temperatures and pre ssure, but LH2 has very low flashpoint (< -253 o C) compared to conventional fuels such as LNG ( ≈ -188 o C), LPG ( ≈ -104 o C) and marine fuel oils (flashpo int > 60 o C). Since the flashpoint ind icates the lowest tem perature at which a volatile substance evaporates to form an ignitable mixtur e with air, a lower flashpoint generally im plies increased fire and explosion hazard. To this end, the United Nations (U N) classifies flamm able liquid and vapours according to their flashpoint (UN, 2011): o Extremely flammable liquid and vapour: flashpoint < 23 o C and initial boiling point ≤ 35 o C o Highly flammable liquid a nd vapour: flashpoint < 23 o C and initial boiling point > 35 o C o Flammable liquid and vapour: flashpoint ≥ 35 o C and ≤ 60 o C o Combustible liquid: flashpoint > 60 o C and ≤ 93 o C o Non-flammable liquid: flashpoint > 93 o C  Hydrogen-air mixtures are flammable over a consid erably wider range of concentrations (4-75 vol.%), compared to conventional fuels such as methane (5-15 vol.%) and propane (2.1-9.5 vol.%).  Hydrogen has very low m inimum ignition energ y (MIE ≈ 0.017 m J), compared to conventional fuels such as methane ( ≈ 0.28 mJ) and propane ( ≈ 0.26 m J). This implies that ignition sources that otherwise do not represent a significant hazard ma y readily ignite hydrogen-air mixtures (Skjold & van Wingerden, 2010). Further research will presum ably result in better understanding of ignition phenomena related to el ectrostatic discharges, m echanical impacts and the effect of external congestion and presence of dust or mist for spontaneous ignition of high-pressure releases (Wola ń ski & W ojcicki, 1973).  Table 1 summarises the classifi cation of gases and vapours for equipment used in explosive atmospheres (EN IEC 60079-0, 2018). The classification is based on the param eters maximum experimental safe gap (MESG), m inimum i gnition curr ent (MIC) and the auto-ignition temperature (AIT). The AIT of hydrogen ( ≈ 500 o C) is com parable to other conventional fuels in temperature class T1, su ch as natural gas ( ≈ 640 o C) and propane ( ≈ 466 o C). Table 1 : Classification of select ed flammable vapours and gases. Gas group Temperature class ‒ AIT ( o C) T1 T2 T3 T4 T5 T6 450 > 300 ‒ 450 200 ‒ 300 135 ‒ 200 100 ‒ 135 85 ‒ 100 IIA MESG > 0.9 mm Methane Propane Ammonia Methanol Ethanol Diesel Fuel oils Petrol ‒ ‒ ‒ IIB 0.9 ≥ MESG > 0.5 Coal ga s Ethy lene ‒ Diethyl ether ‒ ‒ IIC 0.5 mm > MESG Hydrogen Acetylene ‒ ‒ ‒ Carbon disulphide Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 77  The maximum lam inar burning velocity of hydrogen-air m ixtures is significantly higher ( S L ≈ 2.9 m s -1 ) than for most conventional fuels (Konnov et al. , 2018), such as methane ( S L ≈ 0.36 m s -1 ) and propane ( S L ≈ 0.41 m s -1 ). Furthermore, various hydrodynam i c, thermo-diffusive and thermo- acoustic instabilities may influence the flame pr opagation in hydrogen-air mixtures significantly (Clavin & Searby, 2016). In partic ular, the role of therm o-diffusive instabilities depends on mixture composition (Sánchez & W illiams, 2014) ‒ cellular instabilities oc cur at sufficiently fuel- lean conditions (negative Markstei n lengths), and pulsating instabili ties occur at sufficiently fuel- rich conditions (posi tive Markstein lengths).  Vapour cloud explosions (VCEs) in industry enta il turbulent f lame propagation in premixed fuel- air mixtures. The positive fee dback mechanism between expa nsion of combustion products, turbulence generation in the unreacted mixture (e.g. in wakes behind obstacles) , flam e folding caused by obstacles and/or instabilities, and enhan ced rate of turbulen t co mbustion results in flam e acceleration and pressure build-up, es pecially in congested and partly conf ined regions. The chain of events depends on th e reactivity of the m ixture, typica lly quantified by th e lam inar burning velocity mentioned above, as well as initial a nd boundary co nditions such as the turb ulent flow conditions, the size of the flammable cloud, and th e degree of congestion and confinem ent. The extreme reactivity of hydrogen-air m ixtures, relative to conventional fuels, results in a dram atic increase in flam e acceleration in the pres ence of obstacles and/or initial turbulence, with significant potential for escalation (Skjold et al. , 2019b; Shirvill et al. , 2019ab).  Hydrogen-air mixtures have a propensity for deflagration-to-detonation-transition (D DT) under specific conditions (C iccarelli & Dorofeev, 2008). The detonation cell size for stoichiometric hydrogen-air is about 10 mm, compared to 30 0 mm for methane and 70 mm for propane. This implies that sm all variations in the initial a nd boundary conditions can have dramatic influence on the chain of events in hydrogen explosions. Extensive research in the aftermath of the Buncefield explosion has resulted in increased aw areness of the critica l importance of considering detonation scenarios in risk assessments and design of industr ial facilities (Oran et al. , 2020).  Hydrogen can cause embrittlement in certain m ateri als, resulting in loss of physical properties, and possible loss of containment.  Several of the risk-reduc ing m easures frequently employed fo r systems with conventional energy carriers have limited applicability f or hydrogen, including deflagration venting and autom atic suppression.  Many application areas for hydrogen involve emerging technologies, and hence inherent uncertainty associated with the estimation of both event frequencies and consequences.  Hydrogen safety is an area of active res earch, which of ten addresses inh erently complex phenomena and em erging technologies. As such, th e strength of the knowledge available for risk assessments is inherently limited.  Many consultants that perform risk assessm en ts do not have specifi c competence on hydrogen safety, and may apply conventiona l consequence m odels and event frequencies derived from the hydrocarbon industry. This increases the uncertainty in the risk assessments considerably and can result in a false impression of safety am ongs t owners and operators of hydrogen systems.  It takes time to build competence on technical safe ty and to develop a hea lthy safety cultur e in an organisation. Furthermore, experience shows that accidents recur in the sam e industry, and even in the same organisation (Kletz, 1993).  As mentioned above, there is a fundam ental differ ence betw een the required level of safety for controlled production and use of hydrogen within industrial facilities, with access control and trained employees, and widespread us e of hydrogen in the public domain. There are some favourable features of hydrogen concerning safety, at least for specific conditions:  Buoyancy, due to the low density of hydrogen relative to air at the same tem perature, can prevent the formation and limit the duration of large fla mmable clouds in unconfined areas. This implies that natural convection in open geometries is th e preferred approach f or preventing and m itigating hydrogen explosions. However, in confined geom etries, buoyancy favours the formation of highly Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 78 reactive stra tified layers near the ceiling (Skjold et al. , 2019b). This implies that classical m eans of explosion protection are inherently difficult to implement for tunnels and parking facilities, as well as fuel preparation rooms, fuel cell room s and machinery spaces on ships. Furth ermore, in congested unconfined scenarios, a D DT event in a localis ed congested and/or confined region can trigger an aboveground detonation in the rising hy drogen-air plum e. The interaction between the original blast wave and the blast wave reflec ted from the ground m ay result in a Mach stem.  Pure hydrogen flames do not produce soot, which im plies less transport of heat by radiation, compared to hydrocarbon flames. However, jet flam es in realistic accidents may entrain components from volatile materials, such as paint, plastic, du st, etc. Without proper barriers in place, an external fire, e.g. a je t fire from an uncontrolled rel ease of hyd rogen, may com promise the tanks used for storing the fuel. Figure 3 illustrates a typ ical event tree f or the initiating event ‘re lease and dispersion ’ (i.e. loss of containment) of gaseous hydrogen, w ith associated probabilities ( P ) and consequences ( C ) for various scenarios. A com prehensive QRA will conside r mu ltiple initiating events, repres enting leaks with different release rates and durations, located in vari ous positions and pointing in different directions, for different wind conditions and/or with or without fo rced or natural ventilati on. It is essential to account for the variation in the release rate during the event, as well as the delay between the onset of the release and igniti on, since these param eters influence the size of the flamm able cloud, and hence the time-dependant probability of ignition and the consequences of poten tial explosions. Furthermore, there can b e numerous interm ediate or combined scenario s, such as fires initiated by explosions or vice versa, structural collapse caused by fires and explosions, blast waves, projectiles, etc. Nevertheless, the simple event tree in Figure 3 illustra tes the signifi can t challenge of reducing the risk for hydrogen installations to a level equivalent to or below that of conventional fuels. Fig. 3. Simplified event tree for loss of containment of hydrogen. For comparable en ergy systems, i.e. facilities with sim ilar inventory of fuel (chem ical energy), similar probabilities of initiating events, sim ilar levels of confinem ent and congestion, similar age and level of maintena nce, etc., the safety- rel ated properties of fuels mentioned above im ply that the overall risk often will be significantly higher for hydrogen sy st ems, compared to system s based on conventional fuels. In particu lar, the likely shift towards an increas ed fraction of accident scenarios involvin g strong deflagrations or detonati ons implies an increased lik elihood of major accid ents, and hence increased requirements for em ergency preparedness in society. Hence, to achie ve an acceptable level of safety for systems where hydrogen is the pr imary energy carrier, such as hydrogen refuelling stations or ferries, the design of the hydrogen systems must differ significantly from that of contemporary system s using conventional fuels. Furthe rm ore, from the perspect ive of societal safety it will be increasingly important to develop approp riate infrastructure and procedures for emergency response. In summ ary, the specific safety-related properties o f hydrogen imply that it often will be necessary to im plement additional o r specially desi gned safety m easures to reduce the risk to an acceptable level in systems and infrastructure that use hydrogen as an energy carrier. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 79 6. Learning from accidents The accident at a hydrogen refuel ling station in S andvika, Norway, on 10 June 2019 illustrates some of the challenges associated with the safe in troduction of hydrogen in society (Løkke, 2019). The accident took place arou nd 17:30 on Pentecost/Whit Monday, a public holiday in Norway. The chain of events started with a leak from the high-pressure system (about 900 bar) , followed by ignition and a strong explosions. The initial acci dent investigation did not identif y an obvious ignition source. T he accident did not resu lt in any fatalities, and only minor injuries to people when the blast wave from the explosion triggered airbags in car s. Plates of s heet metal c overed the exterior of the fire /pressure wall surrounding the compound with the process equipm ent for hydrogen production, compression and storage. Figure 4 illustrates how parts of the wa ll near the high-press ure unit and plates of sheet metal formed projectiles that landed on the opposite side of a four-lane road and a roundabout, after passing over a combined sidewalk and bicycle pat h. Hence, under slightly different circum stances this accident could have resulted in severe injuries a nd even fatalities.  Fig. 4. The hydrogen refuelling station in Norway a fter the explosion on 10 June 2019 (NRK, 2019). Apart from the considerable m aterial damage to the refuelling station and minor dam age to nearby buildings, the accident resulted in loss of share holder value and a consid erable setback for the development of the hydrogen infrastructure, at least in Norway. The operator of the station closed down all hydrogen refuelling stations in Norway shortly after the acc ident, and put all plans fo r new stations on hold. It is difficult to predict whethe r this acciden t will influence public perception an d widespread adoption of hydrogen as an energy carri er in a long-term perspective. However, the accident demonstrates several important asp ects of hydrogen safety, from an industrial (commercial) as well as a societal perspective:  Risk assessments involving explosion scenarios are often lim ited to pressure and thermal loads, combined with sim plified criteria for effects (h arm ) , and do not necessarily cons ider escalating accident scenarios that may involve structu r al response, fires following explosions or vice versa , failure of pressurised vessels due to fire loads, formation of projectiles du e to explosion or blast loads, etc. Nevertheless, most fire and explosi on accidents that result in considerable dam age involve one or several of these aspects.  The thresholds for individual risk for members of the public are typically tw o orders of m agnitude lower than for onsite personnel in industry. As such, it is essential to adopt sufficiently conservative risk acceptance criteria when memb ers of the public can access the sy stem under consideration, such as hydrogen refuelling st ations, cars, buses, trains, f erries, etc. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 80  The risk assessment for a given facility will not n ecessarily consider all th e indirect losses related to an accident scenario, such as downtim e, loss of production, and delay ed implementation or termination of emerging technologies. A transiti on towards widespread use of hydrogen as an energy carrier in indust ry and society will require massive investments in technology and infrastructure, and the p rospect of se rious accide nts repres ent a significant risk to investors and insurance companies. There are several examples of acci d ents that have had widespread and long-lasting im pact on the development of other industries. Th e Piper Alpha disaster on 6 July 1988 (167 fataliti es) resulted in significant changes to the safety regulations in the offshore petroleum industry. The explosion and fire at the drilling rig Deepwater Horizon on 20 Apr il 2010 (11 fatalities) initia ted the largest m arine oil spill in the history of the petroleum industry, the Macondo bl owout, as well as expensive legal proceedings and sentencing for owners and operators. The accidents at Three Mile Island on 28 March 1979, Chernobyl on 26 April 1986 (42 acute and delaye d fatalities) and Fukushim a-Daiichi on 11 March 2011 (one delayed fatality ) had severe impact on the devel opment of the nuclear industry. Regarding hydrogen, events such as the Hindenburg disaster on 6 Ma y 1937 (36 fatalities), the Space Shuttle Challenger disaster on 28 January 1986 (sev en fatalities), and the hydrogen explosions at Fukushima-Daiichi influence public percep tion. So me recent accidents involving hyd rogen include:  the explosion and fire at the AB Specialty Sili cones Facility in Waukega n, Illinois, on 3 May 2019 (four fatalities, one injured),  the explosion in the Gangwon Technopark in Gangneung, South Korea, on 23 May 2019 (two fatalities, six injur ed),  the explosion at the Air Produc ts facility in Santa Clara, California, on 1 June 2019,  the explosion and fire at the Uno-X refuelli ng station in Sandvika, Norway, on 10 June 2019,  the explosion at the Airg as plant in Waukesha, W isconsin, on 12 December 2019,  the explosion at the OneH2 hydrogen plant in Catawba County, North Carolina, on 7 April 2020,  the explosion at the Praxair hydrogen production plant in Texa s City, Texas, on 11 June 2020 . Although it is not possible to asse ss the level of safety for the emerging hydrogen technologies from a limited number of accidents, m ostly in industrial f acilities, there is reason for concern regard ing the prospective increase in exposure to h ydrogen systems for m embers of the public. 7. Conclusions The inherent complexity of the physical and ch emical phenom ena involved in accidental fires and explosions represents a significant challeng e for the comm unication of hazards and risks for hydrogen systems. This has im plications for risk perception, in industry as well as soci ety. The inherent lack of relevant experience data for the emerging hydrogen technologies implies signi ficant uncertainty in the estimation of event frequencie s for hydrogen system s with a non-t rivial level of com plexity. The level of conservatism in the desi gn of hydrogen installations should re flect this uncertainty. However, optimal design requires reliable and efficient cons equence m odels that reproduce important trends observed in experiments. To this end, future re search on hydrogen safety should focus on developing and validation consequence models for industrial applications. Further bl ind-prediction benchmark studies are arguably the only reliab le way of documenting the pred ic tive capabilities of advanced model systems, as well as the users of such m odels. Large-scale experiments in realistic g eometries are essential for model validation, as well as for supporting the blind-predicti on studies. Certification of modellers can also be an eff ective means of reducing the u ncertain ty in risk assessm ents. A more holistic approach to risk analysis, risk assessmen t and risk management as tools for decision support will be required for ‘nourishing ’ of the em erging hydrogen technologies through the rem a ining technology readiness levels, and even tually towards widespread comm ercial use in society. In its current state, the developm ent of the ‘hydrogen econo my’ is fragile, and new accidents, especially in the public domain, m ay delay or term inate further development. It is imperative that all stakeholders realise that the implications of future severe accidents in hyd roge n system s will not be limited to the companies that own or o perate the affected installations. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 81 References Aven, T. & Kristensen, V. (2019). How the dis tinction between general knowledge and specific knowledge can improve the foundation and practi ce of risk assessment and risk-inform ed decision-making. Reliability Engineering and System Safety , 191: 9 pp. DOI: https://doi.org/10.1016/j.ress.2019.106553 Aven, T. (2015). Implications of black swans to the foundations and prac tice of risk assessment and management. Reliability Engineering and System Safety , 134: 83-91. DOI: https://doi.org/10.1016/j.ress.2014.10.004 Ciccarelli, G. & Dorofeev, S.B. (2008). Flame acceler ation and transition to detonatio n in ducts. Progress in Energy and Combustion Science , 34: 499-550. DOI: https://doi.org/10.1016/j.pecs.2007.11.002 Clavin, P. & Searby, G. (2016). Combustion waves and fronts in flows . Cam bridge University Press, Cambridge. Clean Sky 2 JU & FCH 2 JU (2019). Hydrogen-powered aviation: a fa ct-based study of hydrogen technology, economics, and climate impact by 2050 . Report prepared by McKinsey for Clean Sky 2 Joint Undertaking and Fuel Cells and Hydrogen Joint Undertaking: 96 pp. Cracknell, R.F., Alcock, J.L., Rowson, J.J., Shirvill, L.C. & Ungut, A. (2002). Safety considerations in retailing hydrogen. SAE Technical Paper 2002-01-1928: 922-926. DOI: https://doi.org/10.4271/2002-01-1928 Crowl, D.A. & Jo, Y.-D. (2007). Th e hazards and risks of hydrogen. Journal of Loss P revention in the Process Industries , 20: 158-164. DOI: https://do i.org/10.1016/j.jlp. 2007.02.002 Dawood, F., Anda, M. & Shafiullah, G.M. (2020). Hydrogen production for energy: an overview. International Journal of Hydrogen Energy , 45: 3847-3869. DOI: https://doi.org/10.1016/j.ijhydene.2019.12.059 DNV GL (2019). Maritime forecast to 2050 ‒ Energy transition outlook 2019 . DNV GL Report. DoE (2009). Human performance improvement handbook, Volume 1: concepts and principles . US Department of Energy (DoE), Report DOE-HDBK-1020e2009, Washington DC. DoE (2011). The Department of Energy hydrogen and fuel cells program plan . US Department of Energy (DoE), September 2011. Ehrenfeld, J.R. (1996). Risk assessment and m anagement: a critique of current practices and policy implication s. Industrial & Environmenta l Crisis Quarterly , 9: 376-404. URL: https://doi.org/10.1177/108602669600900305 EN IEC 60079-0 (2018). Explosive atmospheres ‒ Part 0: Equipment ‒ General requirements . European Committee for Electrotechnical S tan dardization (CENELEC), Brussels, July 2018. FCH 2 JU (2019). Hydrogen Roadmap Europe ‒ a sustainable pathway for the European energy transition . FCH JU, January 2019. Floristean, A. (2019). HyLAW Deliverable 4.4: EU regulations and directives which impact the deployment of FCH technologies . February 2019. Flynn, R. & Bellaby, P. (2007). Risk and the public acceptance of new technologies. Edited by R. Flynn and P. Bellaby, Palgrave Macmillan. DOI: https://doi.org/10.1057/9780230591288 Frewer, L. (1999). Risk percepti on, social trust, and public part icipation in strategic decision making: implications fo r em erging technologies. Ambio , 28: 569-574. Hord, J. (1976). Is hydrogen safe? US Department of Commerce, National Bureau of Standards Technical Note 690: 38 pp. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 82 IEA (2019). The future of hydrogen ‒ seizing today’s opportunities . International Energy Agency (IEA), June 2019. IMO (2016). IGF Code ‒ International code of safety fo r ships using gases or other low- flashpoint fuels . IMO, London. Kletz, T. (1993). Lessons from disaster: How organizations have no memory and accidents recur . Gulf Professional Publishing. Kletz, T.A. (1978). What you don’t have, can’t leak. Chemistry and Industry , 6 May: 287–292. Kletz, T.A. (1999). The origins and history of loss prevention. Process Safety and Environmental Protection , 77: 109-116. DOI: https://doi.org/10.1205/095758299529938 Konnov, A.A., Mohammad, A., Kishore, V.R., Kim, N.I., Prathap, C. & Kumar, S. (2018). A comprehensive review of m easurements and data analysis of lam inar burning velocities for various fuel+air mixtures. Progress in Energy and Combustion Science , 68: 197-267. DOI: https://doi.org/10.1016/j.pecs.2018.05.003 Lemkowitz, S.M. & Schotte, R.M. ( 1999). Explosion theory for dumm ies ‒ Using simple theory to predict how process changes affect gas explosion risk. NPT Procestechnologie , March- April 1999: 19-24. Lemkowitz, S.M. & Taveau, J.R. (2019). A m ode l structuring dust, mist , gas/vapour and hybrid explosion behavior: the Chemical-Engineering Model. Chemical Engineering Transactions , 77: 961-966. DOI: https://doi.org/10.3303/CET1977161 Løkke, J.A. (2019). The Kjørbo incident . Presentation, press conf erence 28 June 2019: 22 pp. NRK (2019). Eksplosjon ved hydrogenstasjon [‘Explosion at hydrogen station’]. Online news article from the Norwegian Broadcasting Com pany (NRK). In Norwegian. Available 30.06.2020: https://www.nrk.no/nyheter/eksplosj on-ved-hydrogenstasjon-1.14582944 Oran, E.S., Chamberlain, G. & Pekalski, A. (2020) . Mechanisms and occurrence of detonations in vapor cloud explosions. Progress in Energy and Combustion Science , 77: 100804. DOI: https://doi.org/10.1016/j.pecs.2019.100804 Paté-Cornell, M.E. (1994). Quantitative safety goal s for risk management of industrial facilities. Structural Safety , 13: 145-157. DOI: https://doi.org/10.1016/0167-4730(94)90023-X Rigas, F. & Amyotte, P. (2013). Myth s and facts about hydrog en hazards. Chemical Engineering Transactions , 31: 913-918. DOI: https://doi.org/10.3303/CET1331153 Sánchez, A.L. & William s, F.A. (2014). Recent advances in understand ing of flammability characteristics of hydrogen. Progress in Energy and Combustion Science , 41: 1-55. DOI: https://doi.org/10.1016/j.pecs.2013.10.002 Shirvill, L.C., Roberts, T.A., Royle, M., W ill oughby, D.B. & Sathiah, P. (2019a). Experimental study of hydrogen explosion in repeated pipe congestion ‒ Part 1: Effects of increase in congestion. International Journal of Hydrogen Energy , 44: 9466-9483. DOI: https://doi.org/10.1016/j.ijhydene.2018.04.193 Shirvill, L.C., Roberts, T.A., Royle, M., W illoughby, D.B. & Sathiah, P. (2019b). Experimental study of hydrogen explosion in repeated pipe congestion ‒ Part 2: Part 2: Effects of increase in hydrogen concentration in hydrogen methane-air m ixture. International Journal of Hydrogen Energy , 44: 3264-3276. DOI: https://doi.org/10.1016/j.ijhydene.2018.12.021 Skjold, T. & van Wingerden, K. (2010). A fatal accident caused by bacterial hydrogen production in an atmospheric storage tank . Proceedings Sixth Intern ational Sem inar on Fire and Explosion Hazards (6 ISFE H), Leeds, 11-16 April 2010: 516-525. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 83 Skjold, T., Hisken, H., Bernard, L., Mauri, L., Atanga, G., Lakshmipathy, S., Lucas, M., Carcassi, M., Schiavetti, M., Rao, V.C.M., Sinha, A., Tolia s, I.C., Giannissi, S.G., Venetsanos, A.G., Stewart, J.R., Hansen, O.R., Kumar, C., Krum enacker, L., Laviron, F., Jambut, R. & Huser, A. (2019a). Blind-prediction: estimating the consequences of vented hydrogen deflagrations for inhomogeneous m ixtures in 20-foot ISO containers. Journal of Loss P revention in the Process Industries , 61: 220-236. DOI: https://doi.org/10.1016/j.jlp.2019.06.013 Skjold, T., Hisken, H., Lakshmipathy, S., Atanga , G., van W ingerden, M., Olsen, K.L., Holme, M.N., Turøy, N.M., Mykleby, M. & va n Wingerden, K. (2019b). Vented hydrogen deflagrations in containers: effect of congestion for homogeneous and inhom ogeneous mixtures. International Journal of Hydrogen Energy , 44: 8819-8832. DOI: https://doi.org/10.1016/j.ijhydene.2018.10.010 Skjold, T., Pedersen, H.H., Bernard, L., Middha, P., Narasimhamurthy, V.D., Landvik, T., Lea, T & Pesch, L. (2013). A matter of life and d eath: validating, qualifying and documenting models for simulating flow-related accide nt scenarios in the process industry. Chemical Engineering Transactions , 31: 187-192. DOI: http://dx.doi.org/10.3303/CET1331032 Skjold, T., Siccama, D., Hisken, H., Bram billa, A, Middha, P., Groth, K.M. & LaFleur, A.C. (2017). 3D risk managem ent for hydrogen installations. International Journal of Hydrogen Energy , 42: 7721-7730. DOI: https://doi.org/10.1016/j.ijhydene.2016.07.006 Skjold, T., Souprayen, C. & Dorofeev, S. (2018). Fires and explosions. Progress in Energy and Combustion Science , 64: 2-3. DOI: http://dx.doi.org/10.1016/j.pecs.2017.09.003 Taylor, J.R. (2016). Can proce ss plant QRA reduce risk? – Experience of ALARP from 92 QRA studies over 36 years. Chemical Engineering Transactions , 48: 811-816. DOI: http://dx.doi.org/10.3303/CET1648136 UN (2011). Globally harmonized system of classification and labe lling of chemicals (GHS) . United Nations, New York and Geneva. Wola ń ski, P. & Wojcicki, S. (1973). Investigation in to the mechanism of th e diffusion ignition of a combustible gas flowing into an oxidizing atm osphere. Proceedings fro m the Combustion Institute , 14: 1217–1223. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 84 13th International Symposium on Hazards, Preven tion, and Mitigation of Industr ial Explosions B raunschweig, GERMANY – July 27-31, 2020 Review of the HySEA project Trygve Skjold a,b , Melodia Lucas b , H elene Hisken b , Gordon Atanga b , Sunil Lakshmipathy b,k , Laurence Bernard b,h , Matthijs van Wingerden a,b , Kees van Wingerden b , Jennifer X. Wen c , Vendra Chandra Madhav Rao c , Anubhav Sinha c,l , Marco Carcassi d , Martino Schiavetti d , Tommaso Pini d , Jef Snoeys e , Arve Grønsund Hanssen f , Changjian W ang g , Simon Jallais h , Elena Vyazmina h , Derek Miller i , Carl Regis Bauwens j a University of Bergen, Berg en, Norway b Gexcon, R&D Department, Bergen, Norway c University of Warwick, Coventry, UK d University of Pisa, Pisa, Italy e Fike Europe, Herentals, Belgium f Impetus Afea, Flekkefjord, Norway g Hefei University of Technology, Hefei, China h Air Liquide, R&D, Paris Innovation Cam pus, Paris, France i Air Products, Central Process Safety Team , Allentown, PA, USA j FM Global, Research Division, Norwood, MA, USA k Jensen Hughes, Richmond, BC, USA l Indian Institute of Technology, Bana ras Hindu University, Varanasi, India E-mail: [email protected] Abstract Explosion protection by venting is a f requently used measure f or mitigating the c onsequences of deflagrations in confined systems. The overall goal of the project “Improving hydrogen safety for energy applications through pre-normative research on vented deflagrations” (HySEA) was to provide recommendations for European and intern ational standards on hydrogen explosion venting mitigation sy stems. The HySEA project received f unding from the Fuel Cells and Hydrogen 2 Joint Undertaking (FCH 2 JU) under grant agreemen t No 671461. The official project started on 1 September 2015 and ended on 30 Novem ber 2018. The members of the HySEA consortium were Gexcon (coordinator), University of Warwick (UWAR) , University of Pisa (UNIPI), Fike Europe (FIKE), Impetus Afea (IMPETUS) and Hefei Univ ersity of Technology (HFUT). The research activities in the projec t included developm ent of em pirical and semi-empirical correlations, as w ell as computational fluid dynam ics (CFD) and fini te element (FE) m odels, and verification and validation of the various models against data fr om experim ents performed in 20-foot shipping containers and smaller enclosures with industry-repr esentative obstacles. As foreseen in the project proposal, the project resulted in a hi erarchy of predictive tools for th e safe design of venting devices for hydrogen deflagrations. The HySEA project dem onstrated that explosion protection by venting can be a valuable means of risk reduction for hydr ogen installations, isolated or in combination w ith measures such as natural ventilation, forced ve ntilation, gas detection and inerting. Projectiles generated during explosions in we ak enclosures represen t a signi ficant hazard, but adequately designed venting devices can prevent rupture and fragmentation of contai ners, even for strong hydrogen deflagrations. This paper reviews the resu lts from the HySEA project, including selected research activities conduct ed after the official end date of the project. The discussion highlights some of the practical implications of the pr oject and suggestions for further work. Keywords: Vented hydrogen deflagrations, maximum reduced explosion pressure, structural response, pre-normative resear ch, computational fluid dynam ics, empirical correlations Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 85 1. Introduction It is common practice in the i ndustry to install equipment for hydrogen energy applications in containers and smaller enclosures. Fires and e xplosions represent a significant hazard for such installations, and specific measures are generally re quired for reducing the risk to a tolerable level (Hord, 1976; Strehlow, 1980; Dorofeev et al. , 1994; Cracknel et al. , 2002; Crowl & Jo, 2007; Rigas & Amyotte, 2013; Skjold et al. , 2017a; Fuster et al. , 2017; S kjold et al. , 2018a; Moradi & Groth, 2019). Explosion venting is a frequently used m easure for mitigating the consequences of hydrogen deflagrations in confined syst ems. This p aper summ arises the m ain results f rom the pro ject “Improving hydrogen safety for energy applica tions through pre-normati ve research on vented deflagrations” (HySEA). The HySEA project received f unding from the Fuel Cells and Hydrogen 2 Joint Undertaking (FCH 2 JU) under grant agreement No 671461. Fi g. 1 shows the logo for the HySEA project. Fig. 1. The logo for the HySEA project (www.hysea.eu). The members of the HySEA consortium were Ge xcon (coordinator), Univ ersity of Warwick (UWAR), University of Pisa ( UNIPI), Fike Europe (FIKE), Im petus Afea (IMPETUS) and Hefei University of Technology (HFU T). The project started on 1 Septem ber 2015 and ended on 30 November 2018. The research activities were organised in five work packages (WPs):  WP1 Engineering models and standards  WP2 Experimental campaigns  WP3 Advanced modelling  WP4 Exploitation, dissemination and communication  WP5 Project managem ent The overall goal of the HySEA project was to c onduct pre-normative research on vented hydrogen deflagrations with an aim to provide recommenda tions for European and in ternational standards on hydrogen explosion venting mitigatio n systems. Furt herm ore, the members of the HySEA consortium developed and validated empirical and semi-empirical correlations, as well as computational fluid dynamics (CFD) and finite elem ent (FE) models, and verified and validated se lected models against data from experiments perform ed in containers a nd smaller enclosures with industry-representative obstacles. Additional objectives of the HySEA project included:  To characterise different venting devices (i.e. co ntainer doors, plastic film and commercial vent panels), including measurem ents of the structur al response of the wall s of the enclosures.  To broaden the participation in the project by in viting the scie ntific and indus trial co mmunity engaged in hydrogen safety, and he nce a variety of modellers using different m odels, to participate in blind-prediction benchmark studies based on full-scale experiments in vented enclosures.  To develop, verify and validate models, including CFD-based tools, for reliable predictions of the reduced maxim um overpressure in vented hydrogen deflagrations.  To develop and validate empirical and FE-based tools for predicting the structural response of enclosures during vented hydrogen deflagrations, and derive empi rical pressure-impulse (P-I ) diagrams for typical hydroge n energy applications.  To use the validated CFD models to explore vent ed hydrogen deflagrations in larger enclosures, such as warehouses with fork lift trucks powered by hydrogen.  To formulate recommendations to the standa rdising comm i ttees for EN 14994 (Europe) and NFPA 68 (USA), as well as recommendations for a harmonised intern ational standard on hydrogen explosion vent ing mitigation system s. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 86 Fig. 7: Maximum explosion pressures for 30 vented de flagration tests in 20-foot containers (Skjold, 2019) and predictions accordi ng to the UWAR model (Sinha & Wen, 2019) . Fig. 8: Geometry models from FLACS for a) the bottle basket obstacle, b) the pipe rack obstacle, and c) the high congestion (HC) geometry. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 93 Tab. 2. Total external surface area for the obstacle configurations in F ig. 8. Obstacle Cylinders (m 2 ) Boxes (m 2 ) Total (m 2 ) Bottle basket 18.1 4.9 23.0 Pipe rack 4.8 9.6 14.4 High c ongestio n (HC ) 4.4 3.9 83.0 Unambiguous estimates for the relevant external su rface area of obstacles, or any other parameter used for representing internal c ongestion, represent a sign ificant source of uncertainty in the semi- empirical correlations for the reduced explosions pr essure in vented deflagrations. Other lim itations include reliable prediction of the pressure impulse, as well as realistic rep resentation of the effect of the static opening pressure P stat of the explosion venting device on the reduced explosion pressure P red . The latter is particularly im porta nt for weak enclosures, such as buildings, ships and containers, where even the loading resulting from the opening pr essu re of a commercial vent panel (typically 0.1 bar) may result in perm anent deformation. The UW AR m odel has not been validated for hydrogen concentrations exceedin g about 25 vol.% in air, and there is significant uncertainty in the estimates for concentrations exceeding this value in Fig. 7. Fig. 9 shows examples of empirical pressure-im pulse ( P-I ) diagram s for 20-foot shipping containers, where the damage criteria, i.e. the P-I curves, are based on specific le vels of permanent deformation, as well as whether the containers were dam aged beyond repair (Skjold et al. , 2019a). The plot on the left resembles a class ical P-I diagram for ideal blast waves (Fri edlander, 1946; Krauthammer, 2008; Dewey, 2010), while the plot on the right illustrates typical P-I curves that account for the effect of the finite rise time of the pressure loads (Baker et al. , 1983). Fig. 9: Empirical pressure-impulse diagram s for 20-foot shipping containers. From the point of view of vented deflagrations, 20-f oot ISO containers are rela tively weak structures, and the structural response measurem ents reveal significant deflection of the container walls even for relatively modest pressure loads. Hence, the volume of the enclos ures varied significantly during some of the tests, especially th e more violent explosions. The vent panels opened simultaneously in most of the tests with co mmercial panels. The main s tructure of the con tainer remained inta ct in all tests, except from test 9 where the hinges br oke when the doors opened. One door bounced off the gravel on the side of the container, hit the hi llside some 10 m above th e ground, and landed about 30 m from the container. This observation dem onstrates the hazard posed by projectiles and highlights the importance of securing attached structural elem en ts such as doors, louvre panels and ventilators. The experiments with 20-foot cont ainers included measurements of external blast pressures (Skjo ld et al. , 2017c; Skjold, 2018b). Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 94 The container doors do not represent prop er expl osion venting devices according to the European standard (EN 14797, 2006). The containe rs walls ruptured in some of the tests that produced high overpressures, e.g. test #9 in Fig. 5-7. The smoothi ng frequency used for processing the experimental pressure-time curves influenced the results (Skjold, 2018ab; Skjold et al. , 2019d), but prim arily for the maximum rate of pressure rise. The finite rise time of the pressu re lo ads and the high level of plasticity of the corrugated plates in the walls a nd roof complicate the analysis of the structural response (Baker et al. , 1983). Furthermore, it was not strai ghtforward to calculate unambiguous values for pressure im pulse for tests with m ultiple pressure peaks. Fig. 5-7 illustrates that the ma ximum reduced explosion pressure s increase cons istently with increasing fuel concentration for all obstacle and vent configurations. This is reason able since most tests involved lean mixtures, i.e. less than 30 vol.% hydrogen in air. Tests 34 (O) and 69 (P) with rich mixtures (42 vol.% hydrogen) were included to explore near worst-case conditions for a modest degree of congestion (P2). The maximum pressures in crease m ore rapidly for tests with internal congestion, compared to the predictions by EN 14994 (2007). The results for tests 71 and 72 demonstrate the strong ef fect that higher levels of congestion (HC) can have on the m aximum reduced explosion pressure, even for a modest increase in concentra tion (from 12 to 15 vol.%) for lean hydrogen-air mixtures. Although no detonations were obser ved in any of the 66 tests, tests with more reactive mixtures in th e HC geom etry would likely have resulted in deflagration-to-detonation- transition (DDT). Loss of containment of gaseous hydrogen in confined spaces will typically result in b uoyant releases and stratified fuel-air clouds. The significantly higher overpressures obtained for stratified mixtures, compared to lean homogeneous m ixtures with the same total m ass of fuel, imply that m odels for vented hydrogen deflagrations should account for the effect of inhom ogeneous fuel-air clouds. The second blind-prediction benchmark ex ercise in the HySEA project expl ored the predic tive capabilities of consequence models for sim ulating vented de flagrations resulting fr om stratified hydrogen-air mixtures ignited to def lagration in 20-foot ISO cont ainers (Skjold et al. , 2018b; Skjold et a l. , 2019b). Although some of the results were encouraging, especially for the m odelling of release and dispersion scenarios, there is significant room for improvin g the p redictive capabilitie s of both model systems and modellers. For well-defined vented deflagration scenarios , the spread in the explosion pressures predicted by different CFD tools covered two orders of m agnitude, and none of the predictions were within a factor two of th e experimental values. Large-scale experiments by HFUT HFUT designed a test vessel with dimensions 12 .0 m × 2.5 m × 2.5 m, i.e. approximately the sam e dimensions as a standard 40-foot ISO container (Rui et al. , 2020). The walls of the enclosure were made from 25 mm steel plates, a nd 20 rectan gular vent panels could be fitted on the roof. To accommodate tests with inhomogeneous m ixtures, 20 noz zles located near th e ceiling could b e used for creating well-defined concentration gradients. The experimental setup allowed resea rchers to study the effect of the internal layout (obstruction s) , structural response, and the effect of m itigating measures. By July 2019, HFUT has com pleted 92 vent ed deflagration tests in the 40-foot enclosure:  47 tests vented through the opening at the e nd wall, covered by a thin plastic film  45 tests vented through openings in the roof, covered by a thin plastic film HFUT has also completed 24 vented tests in the 20 -foot enclosure (half of the 40-foot enclosure):  12 tests vented through the opening at the end wall, covered by a thin plastic film  12 tests vented through openings in the roof, covered by a thin plastic film. Fig. 10 shows selected frames from a test in the 40-foot enclosure, vented through the opening at the end wall. The experimental m atrix included varia tion in parameters such as hydrogen concentration, ignition locations, obstacles (tanks), and the number of vent openings on the roof. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 95  Fig. 10: Vented deflagration test in the 40-foot enclosure . 4. Advanced modelling – CFD and FE (WP3) This section summarises the research ac tivities in W P3 of the HySEA project. UWAR (Rao & Wen, 2017ab; Rao & Wen, 2019bcd) , in cooperation with HFUT (Wang & Wen, 2017), developed the in-house CFD solver HyFOAM, based on the open-source solver OpenFOAM. HFUT developed and validated the CFD code Explosion FOAM, also within the OpenFOAM framework (Wang & Wen, 2017). Gexcon (Lakshmipathy et al. , 2017; Lakshmipathy et al. , 2019; Lucas et al. , 2019ab) developed the CFD solver FLACS™, including the special versi on FLACS-Hydrogen for sim ulating dispersion and explosion phenomena involving hydrogen. The two vers ions of FLACS-Hydrogen that were released during the project period for the HySEA project focused on validation of the Flacs-2 solver and did not include extensive changes to the model system . Lucas et al. (2019a) described th e changes to the model system for the new CFD so lver developed during the HySEA project (Flacs-3). Fig. 11 shows selected results f or both solvers for tw o experiments with stratified mixtures. Fig. 11: Simulated and experimental pressure -time histories for tests 57 and 59 with stratified mixtures and commercia l vent panels (Lucas et al., 2019a). The new Flacs-3 solver includes se veral fundamental changes (Lucas et al. , 2019a), including a new two-equation Reynolds-averaged Navier Stokes (RANS) turbulence model (the kskL model), a new correlation for turbulent burning veloc ity that includes the ef fect of th e stretch-rate Markstein number, and a transport equation for flame folding. Gexcon and UWAR used the impr oved model system to simulate vented hydrogen deflagra tions in larger enclosures. Fig. 11 shows the geometry model for a hypothetical warehouse (Lucas et al ., 2019b). Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 96 In connection with the first HySEA blind- prediction study, Gexcon and IMPETUS (Atanga et al. , 2017; Atanga et al. , 2019) developed a methodology for one-w ay coupling between the CFD solver FLACS and the Impetus Afea FE solver. The p rimary lim itation of this approach is the inherent uncertainty in the prediction of the transient press u re loads on the walls of th e enclosure during vented hydrogen deflagrations (Skjold et al. , 2017b; Skjold et al. , 2018; Skjold et al. , 2019bc). Fig. 11: Hypothetical warehouse geometry (Lucas et al., 2019b) . UWAR (Rao & Wen, 2017ab; Ra o & W en, 2019bcd) implem ented several model improvem ents in HyFOAM during the HySEA project:  Flame wrinkling com puted from an algebrai c m odel or by solving a transport equation.  Lewis number effects for the turbulent burning velocity in lean hydrogen-air m ixtures.  An analytical expression for estim ating the eff ect of the Darrieu s–Landau and thermo-diffusive instabilities on flame propagati on in lean hydrogen-air mixtures. Rao & Wen (2018, 2019a) modelled the interaction between the pressure loads predicted by HyFOAM and the structural response of the container walls computed by fluid-structure interactions (FSI) for vented hydrogen deflagrations. To evaluate the dynamic displacem ent of the container walls based on the peak overpressure insi de the container, a single d egr ee of freedom m odel approximated the structural response obse rved in the experiments. 5. Exploitation, dissemination and communication (WP4) This section summarises selected activ ities from WP4 of the HySEA project. The HySEA project included extens ive activitie s related to dissem ination and communication of results ‒ 33 of the 80 deliverables from the project addressed various aspects of dissemination. The project website ( www.hysea.eu ) includes a comprehensive list of variou s dissem ination activitie s, including publications and newsletters. The memb ers of the HySEA consor tium made significant efforts to involve the intern ational research community on hydrogen safety, including active interaction with the Internationa l Association f or Hydrogen Safety (IA HySafe) and International Energy Agency (IEA) Hydrogen Task 37 on Hydr ogen Safety. UNIPI and Gexcon organised two blind-prediction benchmark exerci ses during the project (Skjold et al. , 2017b; Skjold et al. , 2018; Skjold et al. , 2019bc). Gexcon Software rel eased an updated version of the CFD tool FLACS for hydrogen applications in 2017 (F LACS v10.7), and an updated beta version in 2018 (FLACS v10.8 beta). Through the cooperation with UNIPI, IM PETUS and Gexcon, Fike explored innovative solutions for venting of hydrogen deflagrations in c ontainers and smaller enclosures. UWAR released two versions of the HyFOAM software. HFUT us ed the network and competence acquired through the participation in the HySEA consortium to acquire resear ch funding from the Chinese governm ent. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 97 The final dissemination workshop was organised in cooperation with the Fire and Blast Inform ation Group (FABIG). Gexcon, UWAR and Air Liquide pres ented results from the HySEA project at the FABIG technical meetings ‘Developments in Fire & Explosion Engineering towards a Hydrogen Economy’ in Aberdeen on Wednesday 26 Septembe r and in L ondon on Thursday 27 September 2018. The ultimate goal of the HySEA projec t was to provide recommendations f or a harmonised international standard on vented hydrogen deflagrations, and thereby support safe and widespread introduction of hydrogen as an energy carrier in industry and society. The call explicitly m entioned the European standard EN 14994 (2007) on “Gas explosion venting protective systems” . To this end, UWAR and Gexcon communicated relevant results to the European Comm ittee for Standardization (CEN) Technical Committee 305 (T C305) Working Group 3 (WG3) ad -hoc group on gas explosions. 6. Project management (WP5) This section summarises selected activ ities in WP5 of the HySEA project. Gexcon organised the kick-off meeting for the HySEA project in Bergen on 14-16 Septem ber 2015. The subsequent progress and advi sory board meetings were organi sed by FIKE (Her entals, Fe bruary 2016), Gexcon (Bergen, September 2016), UNIPI (Pis a, February 2017), Gexcon (Bergen, Septem ber 2017), Air Liquide (Paris, April 2018), HFUT (H efei, July 2018), and finally by Gexcon (London, September 2018). Gexcon presented results from the HySEA project at the FCH 2 JU Programm e Review Days in Brussels in 2016, 2017 and 2018. 7. Impact The ultimate goal of the HySEA pr oject was to deliver amendm ents on vented hydrogen deflagrations to the European standard EN 14994 “Gas explosion venting protective systems” and the American standard NFPA 68 “Standard on explosion protec tion by deflagration venting” , as well as a potential new international standard on “Hydrogen explosion venting mitigation systems” from the International Organization for Sta ndardization (ISO). The relevant results f rom the HySEA project were not available in tim e to be considered for the latest revision of NFPA 68 (2018). Furthermore, there was no initiative within ISO for a common international standard on vented deflagrations. Hence, the efforts towards RCS focused on EN 14994 (2007). As a result of th e constructive dialogue and exchange of information between members of the HySEA consortium and the ad-hoc group on gas explosions in CEN TC305 WG3, a version of the semi-em pirical model developed by UWAR will likely be included as an appendix to the nex t revised version of EN 14994. As such, the HySEA project achieved its prim ary objective, at least to the extent that m embers of the HySEA consortium could influence the internal pro cesses of the relevant standardising committee during the projec t period. The HySEA project resulted in a hier archy of predictive tools for th e safe design of venting devices for hydrogen deflagrations. UWAR developed a semi-empirical model for vented hydrogen deflagrations (Sinha & Wen, 2019a b), and UWAR with support from HFUT developed an improved version of the CFD solver HyFOAM (Rao & Wen, 2019abcd), Gexcon developed an improved version of the commercially availa ble software product FLACS (Atanga et al. , 2019; Lakshmipathy et al. , 2019; Lucas et al. , 2019ab), and IMPETUS validated the co mmercially available Im petus Afea FE solver for structural response calculations in connection with hydrogen e xplosions in enclosures (Atanga et a l. , 2019; Pini et al. , 2019; Skjold et al. , 2019c). Although the research activities in the HySEA project resulted in im prove d models and m odel systems, it is necessary to improve the predictive capabilities of conseque nce models for vented hydrogen defl agrations further. In particular, there is a need for empirical and s emi-empirical models that can predict both the maximum pressure and the impulse of vented hydrogen deflagrations, and at the sam e time acc ount for the effect of internal congestion and the opening pre ssure of commercial vent panels. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 98 The experimental data generated in the HySEA project will be of va lue to end-users, as well as modellers from industry and academ ia. Several mode llers and research groups participated in the blind-prediction benchmark studies or ganised during the project (Skjold et al. , 2019bc), and the results from the vented deflag ration experiments have been published extensively (Carcassi et al. , 2018; Pini et al. , 2019; Schiavetti et al. , 2019; Skjold, 2019; Skjold et al. , 2019abcd). The outcome of the blind-prediction studies is likely to increa se the awareness amongst developers and users of consequence models for vented deflagrations concer ning the inherent limita tion and uncertainty of the model predictions. In a longer pe rspective it is foreseen that th is awareness will result in m odel improvements, includ ing m ore user-friendly interfaces, as well as updated docum entation and guidelines, and possibly some kind of certification or appr oval of users of advanced model systems. In principle, it should be possibl e to com plement the experimental results from the HySEA project with data from CFD and FE sim ulations of vented deflagrations (Atanga et al. , 2019). However, the results from the two blind-predic tion benchm ark studies conducted as part of the HySEA project demonstrate that it is not straightforward to simulate vented hydrogen de flagrations in 20-foot shipping containers with in ternal congestion (Skjold et al. , 2019ac). Combustion in stratified hydrogen-air mixtures and vented deflagrations in weak enclos ures are inherently com plex phenomena. The lam inar burning velocity of hydrogen- air mixtures is signifi cantly higher than for most conventional and alternative fuels (Astbury, 2008; Konnov et al. , 2018), and various hydrodynamic, thermo-diffusive and therm o-acoustic instabilities m ay influence the flame (Clavin & Searby, 2016). To this end, it is nece ssary to further im prove the pred ictive capabilities of advanced consequence models for vented hydr ogen deflagrations. It is particul arly im portant to improve the modelling of flam e acceleration in highly congeste d g eom etries and to im plement models that describe the opening of realistic explosion venting devices with sufficient accuracy. Safe design of container-based in stallations for hydrogen energy appl ications should consider the hazard of projectiles generated from vented deflag rations. As such, the experim ents with 20-foot shipping containers on the HySEA project demonstrated that the container doors do not represent suitable explosion venting devices (Skjold et al. , 2017c; Skjold et al. , 2019d). Properly installed and certified explosion venting devices shall not represent a hazard (EN 14797, 2006), but it may be necessary to secure structural elemen ts such as louvre panels and f ans for natural or f orced ventilation. The experiments perform ed as part of the Hy SEA project did not resu lt in any DDT events. Nevertheless, the possibility of DDT and detonati on should be accounted for in risk assessm ents for systems with significant inventory of hydrogen and high degree of congestion. In general, it can be difficult to rule out the possibility of detonations for large flamma ble clouds with more than 18-20 vol.% hydrogen in air if the initial and boundary conditions support sign ificant flame acceleration. The results from the two HC scenarios in Fig. 5-7 demonstrate the extrem e e ffect internal congestion can have on vented hydrogen deflagrations. In summ ary, it is not straightforward to elim inate the possibility of DDT for industrial hydrogen systems th at include high levels of congestion inside the enclosure, especially in the event of hi gh-pressure releases impinging on congestion. The results from the blind-prediction benchm ark exercises in the HySEA project (Skjold et al . , 2019bc) demonstrated severe limitati ons in the predictive capabilities of current state-of-the-art consequence models for vented hydrogen deflagrations. At the sam e time, m odern hydrogen energy applications represent emerging technologies w here there are limited statistic al data to support the estimation of event frequencies. In light of the increasing awareness of the importance of reflecting knowledge and lack of knowledge in the understa nding, assessment and m anagement of risk (Aven & Kristensen, 2019), the inherent uncertainty in risk assessm ents for hydrogen systems should be reflected in a conservative approach to both the location and the design/layout of new installations. In this context it is importan t to keep in mind that th e criteria for tolerable o r acceptable risk for hydrogen systems located in populated areas is typically two orders of magnitude lower than for industrial facilities (Paté-Cornell, 1994). As such, robust, or resilient, design based on a credible worst-case approach should be considered as an alternative to the classica l probabilistic approach. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 99 The HySEA project demonstrated th at explosion protection by venting can be a valuable addition to other means of risk reduction for hydrogen installations, such as natural ven tilation, forced ventilation, gas detection and inerting. Projectiles represent a signifi cant hazard, but adequately designed venting devices can prevent rupture and fragmentation of contai ners, even for strong hydrogen deflagrations. For a given mass of hydrogen, stratified mixt ures result in significantly higher pressures, compared to lean hom ogeneous mixtures. Although it is not straightforward to predict the maximum reduced explosion pressures in vented hydrogen deflagrati ons, a total vent area of 8 m 2 on the roof have been show n to give reasona ble protection of 20-foot ISO containers with limited internal congestion, even for highly reactive m ixtures (Skj old, 2019; Skjold et al. , 2019ad). To this end, the results from the HySEA project have resulted in significant changes in the approach to explosion protection in relevant industries. Acknowledgements The HySEA project received funding from the Fuel Cells and Hydrogen 2 Joint Undertaking (FCH 2 JU) under grant agreement No 671461. This Joint U ndertaking receives support from the European Union’s Horizon 2020 research and innovation programme and Hydrogen Europe and Hydrogen Europe Research. The members of the HySEA c onsortium gratefully acknowledge the valuable contributions from the m embers of the HySEA Advi sory Board: Simon Jallais and Elena Vyazmina from Air Liquide, Derek Miller from Air Produc ts and Carl Regis Bauwens from FM Global. References Astbury, G.R. (2008). A review of the propertie s and hazard s of some alternative fuels. Process Safety and Environmental Protection , 86: 397-414. DOI: https://doi.org/10.1016/j.psep.2008.05.001 Atanga, G., Lakshmipathy, S., Skjold, T., Hisken, H. & Hanssen, A.G. (2017). Structural response for vented hydrogen deflagrations: coup ling CFD and FE tools. Proceedings Seventh International Conference on Hydrogen Safe ty (ICHS 2017), Paper # 224, Hamburg, 11-13 September 2017: 378-387. ISBN 978-88-902391. URL: https://www.hysafe.info/wp-co ntent/uploads/2017_papers/224.pdf Atanga, G., Lakshmipathy, S., Skjold, T., Hisken, H. & Hanssen, A.G. (2019). Structural response for vented hydrogen deflagra tions: coupling CFD and FE tools. International Journal of Hydrogen Energy , 44: 8893-8903. DOI: https://doi.org/10.1016/j.ijhydene.2018.08.085 Aven, T. & Kristensen, V. (2019). How the dis tinction between general knowledge and specific knowledge can improve the foundation and practi ce of risk assessment and risk-inform ed decision-making. Reliability Engineering & System Safety , 191: 9 pp. DOI: https://doi.org/10.1016/j.ress.2019.106553 Baker, W.E., Cox, P.A., Westine, P.S., Kulesz, J.J. & Strehlow, R.A. (1983). Explosion hazards and evaluation . Elsevier, Amsterdam. Bauwens, C.R., Chaffee, J. & Dorofeev, S.B. ( 2010). Effect of ignition lo cation, vent size, and obstacles on vented explosion overp ressures in propane-air m ixtures. Combustion Science and Technology , 182: 1915- 1932. DOI: http://dx.doi.org/10.1080/00102202.2010.497415 Bauwens, C.R., Chao, J. & Dorofeev, S.B. (2012). Effect of hydrogen concentration on vented explosion overpressures from l ean hydrogen-air deflagrations. International Journal of Hydrogen Energy , 37: 17599-17605. DOI: https://doi.org/10.1016/j.ijhydene.2012.04.053 Carcassi. M., Schiavetti, M. & Pini, T. (2018). Non-homogeneous hydrogen deflagrations in small scale enclosure: Experim ental results. International Journal of Hydrogen Energy , 43: 19293-19304. DOI: https://doi.org /10.1016/j.ijhydene.2018.08.172 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 100 Clavin, P. & Searby, G. (2016). Combustion waves and fronts in flows . Cam bridge University Press, Cambridge. Cracknell, R.F., Alcock, J.L., Rowson, J.J., Shirvill, L.C. & Ungut, A. (2002). Safety considerations in retailing hydrogen. SAE Technical Paper 2002-01-1928: 922-926. DOI: https://doi.org/10.4271/2002-01-1928 Crowl, D.A. & Jo, Y.-D. (2007). The hazard and risks of hydrogen. Journal of Loss Prevention in the Process Industries , 20: 158-164. DOI: https://doi.org/10.1016/j.jlp.2007.02.002 Dewey, J.M. (2010). The shape of the blast wave: studies of the Friedlander equation. Proceedings Twenty-first Internation al Symposium on Military Aspects of Blast and S hock (21 MABS), Jerusalem, 3-8 October 2010: 9 pp. Dorofeev, S.B., Kochurko, A.S., Efimenko, A.A. & Chaivanov, B.B. (1994). Evaluation of the hydrogen explosion hazard. Nuclear Engineering and Design , 148: 305-316. DOI: https://doi.org/10.1 016/0029-5493(94)90116-3 EN 14797 (2006). Explosion venting devices . European Committee for Standardization (CEN), Brussels. EN 14994 (2007). Gas explosion venting protective systems . European Committee for Standardization (CEN). Friedlander, F.G. (1946). The di ffraction of sound pulses I: Diffrac tion by a semi-infinite plate. Proceedings of the Royal Society A , 186: 322-344. DOI: https://doi.org/10.1098/rspa.1946.0046 Fuster, B., Houssin-Agbomson, D., Jallais, S., Vyazmina, E., Dang-Nhu, G., Bernard-Michel, G., Kuznetsov, M., Molkov, V., Chernyavskiy, B., Shentsov, V., Makarov, D., Dey, R., Hooker, P., Baraldi, D., Weidner, E., Melideo, D., Pa lmisano, V., Venetsanos, A. & Kinderen, J.D. (2017). Guidelines and recommendations for indo or use of fuel cells and hydrogen system s. International Journal of Hydrogen Energy , 2: 7600-7607. DOI: https://doi.org/10.1016/j.ijhydene.2016.05.266 Groth, K.M. & Hecht, E.S. (2017). HyRAM: A methodology and toolkit fo r quantitative risk assessment of hydrogen system s. International Journal of Hydrogen Energy , 42: 7485-7493. DOI: https://doi.org/10.1016/j.ijhydene.2016.07.002 Guo, J., Liu, X. & Wang, C.J. (2017a). Experi m ents on vented hydrogen-air deflagrations: the influence of hydrogen concentration. Journal of L oss Prevention in the Process Industries , 48:254-259. DOI: http://doi.org/10.1016/j.jlp.2017.05.013 Guo, J., Sun, X., Rui S., Cao, Y ., Hu, K., & Wang, C.J. (2015). Effect of ignition position on vented hydrogen-air explosions. International Journal of Hydrogen Energy , 40: 15780-15788. DOI: http://doi.org/10.1016/j.ijhydene.2015.09.038 Guo, J., Wang, C.J., Liu, X. & Chen, Y. (2017b) . Explosion venting of rich hydrogen-air mixtures in a sm all cylindrical vess el with two symmetrical vents. International Journal of Hydrogen Energy , 42: 7644-7650. DOI: http://doi.org/10.1016/j.ijhydene.2016.05.097 Hisken, H., Atanga, G., Skjold, T., Lakshm ipathy, S. & Middha, P. (2016). Validating, documenting and qualif ying models used for consequence assessm ent of hydrogen explosion scenarios. Proceedings Eleventh Interna tional Symposium on Hazards, Prevention and Mitigation of Industrial Explosions (11 ISHPMIE), Dalian, 24-29 July 2016: 1069-10 86. DOI: https://doi.org/10.5281/zenodo.581649 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 101 Holtappels, K. (2006). Report on th e experimentally determ ined explosion lim its, explosion pressures and rates of explos ion pressure rise, Part 1: methane, hydrogen and propylene, SAFEKINEX Project Deliverable No. 8, Federal In stitute for Materials Research and Testing (BAM), Germany: 149 pp. URL: https://www.morechemistry.com/SAFEK INEX/deliverables/44.Del.%20No.%208.pdf Hord, J. (1976). Is hydrogen safe? US Department of Commerce, National Bureau of Standards Technical Note 690: 38 pp. Konnov, A.A., Mohammad, A., Kishore, V.R., Kim, N.I., Prathap, C. & Kumar, S. (2018). A comprehensive review of m easurements and data analysis of lam inar burning velocities for various fuel+air mixtures. Progress in Energy and Combustion Science , 68: 197-267. DOI: https://doi.org/10.1016/j.pecs.2018.05.003 Krauthammer, T. (2008). Modern protective structures . CRC Press, Boca Raton. Lakshmipathy, S., Skjold, T., Hisken, H. & Atanga, G. (2017). Consequence m odels for vented hydrogen deflagrations: CFD vs engineering m odels. Proceedings Seventh International Conference on Hydrogen Safety (ICHS 2017), Paper # 222, Hamburg, 11-13 September 2017: 615-626. ISBN 978-88-902391. URL: https://www.hysafe.info/wp-co ntent/uploads/2017_papers/222.pdf Lakshmipathy, S., Skjold, T., Hisken, H. & Atanga, G. (2019). Consequence m odels for vented hydrogen deflagrations: CFD vs engineering models. International Journal of Hydrogen Energy , 44: 8699-8710. DOI: https://doi.org/10.1016/j.ijhydene.2018.08.079 Li, H., Guo, J., Yang, F. Wang, C., Zhang, J. & L u, A (2018). Explosion venting of hydrogen-air mixtures from a duct to a vented vessel. International Journal of Hydrogen Energy , 43: 11307-11313. DOI: https://doi.org /10.1016/j.ijhydene.2018.05.016 Lucas, M., Hisken, H. & Skjold, T. (2019a). S imulating vented hydrogen deflagrations: improved modelling in the CFD tool FLACS-Hydrogen. Accepted for oral presentation at Eig hth International Conference on Hydrogen Safety (ICHS 2019), Adelaide, 24-26 September 2019. Lucas, M., Skjold, T. & Hisken, H. (2019b). CFD simulations of hydrogen releases and vented deflagrations in large en closures. Subm itted to Journal of Lo ss Prevention in the Process Industries. Moradi, R. & Groth, K.M. (2019). Hydrogen storage and delivery: re view of the state of the art technologies and risk an d reliability analysis. International Journal of Hydrogen Energy , 44: 12254-12269. DOI: https://doi.org /10.1016/j.ijhydene.2019.03.041 NFPA 68 (2018). Standard on explosion protection by deflagration venting . National Fire Protection Association (NFPA), Quincy, Massachusetts, USA. Pate-Cornell, M.E. (1994). Quantitative safety goal s for risk management of industrial facilities. Structural Safety , 13: 145-157. DOI: https://doi.org/10.1016/0167-4730(94)90023-X Pini, T., Hanssen, A.G., Schiavetti, M. & Carca ssi, M. (2017). Experimental measurem ents of structural displacement during hydrogen vented deflagrations for FE model validation. Proceedings Seventh International Conference o n Hydrogen Safety (ICHS 2017), Paper # 292, Hamburg, 11-13 September 2017: 400-411. ISBN 978-88-902391. URL: https://hysafe.info/wp-content/uploads/2017_papers/292.pdf Pini, T., Hanssen, A.G., Schiavetti, M. & Carca ssi, M. (2019). Small scale experim ents and FE model validation of struct ural response during hydr ogen vented deflagrations. International Journal of Hydrogen Energy , 44: 9063-9070. DOI: https://doi.org/10.1016/j.ijhydene.2018.05.052 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 102 T able 1 : Obstacle configurations of the test rig Blockage ratio BR in % Spacing s in mm Obstructed part L obs in m Abbre viation 60 100 3.95 BR60S100L 300 3.95 BR60S300L 1 30 100 3.95 BR30S100L 300 3.95 BR30S300L 1 Results for this configuration are only mentioned briefly . A more detailed discussion can be found in Heilbronn et al. (2019) unit. Due to a pressure of 8 bar in the fuel supply system, the critical flo w condition in the nozzles ensures a uniform fuel distrib ution along the explosion channel. The fuel is injected into the test rig through 27 holes per section in the channel ceiling. In order to prepare the channel for one shot, the channel is e v acuated to a specific sub-atmospheric pressure at first. By opening the fuel v alv es for a set injection time, the channel is filled with the H 2 /CO-mixture until ambient pressure is reached. The inflo wing gas is deflected at deflection plates which are inte grated in the obstacles in the obstructed part or which are mounted at the ceiling in the unobstructed of the channel, as sho wn in fig. 2. At the unobstructed part of the channel the deflection plates induce a BR of about 2 %. Due to the deflection a stratified mixture is established after the injection. Dif fusion allows the fuel to distrib ute ov er the channel height, mixing with air present in the channel. Since only homogenous mixtures are in vestigated, the dif fusion time between the end of injection and ignition is set to t D = 60 s. In vestig ations at a model of the test rig used by V ollmer et al. (2010) sho wed that a uniform fuel distribution can be achie v ed by setting this dif fusion time. The fuel content of the channel is measured based on the method of partial pressures by two static pressure transducers of type SW A09 and WIKA S20. The mixture is ignited using a con ventional spark plug centred at the end plate of the channel. The spark plug is operated for approximately 15 ms. After each shot, the channel is flushed with air for 4 min. t D H 2 /CO-Mixture Air Obstacle Deflection plate H = 60 mm F ig. 2 : Injection mechanism In order to track the flame trajectory , 36 photodiodes are located in the ceiling of the explosion channel as indicated by the red dots in fig. 3. The UV -sensiti v e diodes are of type Hamamatsu S1366-18BQ and are mounted slightly abo v e the centreline. A quarz glass co ver protects each photo diode from the flame, yielding an 10 ◦ angle of vie w . Kno wing the position and time of arri v al at each diode, the trajectory of the flame is obtained. The mean flame v elocity u f between two diodes located at positions x 1 and x 2 is e v aluated from the time of arri v al at each diode t 1 and t 2 : Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 109 u f  x 2 − x 1 2  = ∆ x ∆ t = x 2 − x 1 t 2 − t 1 (1) The dynamic pressure is recorded by eight pressure transducers of type Kistler 601A as indicated by the green squares in fig. 3. Sev en transducers are mounted in the ceiling of the channel, one is mounted at the end plate. The faces of the transducers are co v ered by a thin layer of silicone in order to protect the transducers from thermal shocks. The pressure and photo diode signals are recorded by a data logging system operated at a sample rate of 225 kHz. Pressure transducer Photo diode Spark plug x F ig. 3 : Con ventional measur ement system 3 Results In table 2 the desired fuel contents x F as well as the expansion ratios σ for fuel mixtures of 100/0, 75/25 and 50/50 H 2 /CO are listed. The expansion ratios for an adiabatic comb ustion are calculated with Cantera (Goodwin et al. (2018)). The fuel content x F is gi ven in v ol . -% in air . The fuel injection mechanism of the channel does not allo w to set an exact fuel content. A v ariance of ∆ x F = ± 0 . 5 v ol . -% with re gards to the desired fuel content is accepted. The data presented in the follo wing section is based on the a v erage of at least three e xperiments per fuel and fuel content. T able 2 : Expansion ratios at the desir ed fuel contents x F [v ol . -%] 15 17.5 20 22.5 25 27.5 30 35 40 σ 100 / 0 4.64 5.14 5.62 6.06 6.47 6.81 7.02 6.88 6.58 σ 75 / 25 4.73 5.24 5.72 6.16 6.55 6.85 7.02 6.93 6.62 σ 50 / 50 4.82 5.34 5.82 6.25 6.62 6.89 7.04 7.00 6.70 3.1 Relative terminal velocity The relati ve terminal v elocity ˜ u term is e v aluated by a v eraging the v elocity at the unobstructed part of the channel for each experiment. The av eraged velocity is normalized by the Chapman-Jouguet v elocity D CJ for the gi ven fuel and fuel content calculated by the SD-T oolbox (Bro wne et al. (2008)). The obtained relati ve terminal v elocities are a veraged re garding the fuel and desired fuel content. The relati ve terminal v elocities ˜ u term plotted ag ainst the fuel content for BR30S100L are sho wn in fig. 4. T w o re gimes are identified. F or ˜ u term ≈ 1 flames are found in the detonation re gime, while flames at ˜ u term ≈ 0 . 6 are found in the fast flame re gime. F or BR30S100L all fuels at a fuel content of x F = 15 vol . -% are found in the f ast flames re gime. If the fuel content is increased the 100/0 H 2 /CO fuel transits to the detonation re gime at x F = 17 . 5 vol . -%, while fuels of 75/25 and 50/50 H 2 /CO are still found in the fast flame re gime. If the fuel content is further increased all fuels are found in the detonation re gime. As soon as the detonation regime is reached, ˜ u term is highgest for 100/0 H 2 /CO with a slight decrease at the boundaries of the detonation re gime at x F = 17 . 5 and 30 vol . -%. In case of 75/25 and 50/50 ˜ u term is lo wer than for 100/0 H 2 /CO. For x F ≤ 27 . 5 v ol . -% the highest terminal v e- locities are found for 100/0 follo wed by 75/25 and 50/50 H 2 /CO. At a fuel content of x F ≥ 30 vol . -% Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 110 transition from the detonation to the fast flame re gime is observ ed for 100/0, while 75/25 and 50/50 H 2 /CO-fuels remain in the detonation re gime. For x F = 40 v ol . -% only 50/50 H 2 /CO-fuels are found in the detonation re gime. In this case ˜ u term is slightly belo w the unity . The configuration BR30S100L is similar to the one used by V eser et al. (2002) with regard to blockage ratio and spacing. As sho wn in fig. 4, ˜ u term for x F = 15 v ol . -% is similar for all fuels. This agrees well with the findings of V eser et al. (2002). Both studies show similar velocities ( u ≈ 1400 m s − 1 ) in the obstructed part. The e xperiments conducted by V eser et al. (2002) did not feature an unobstructed section. T erminal velocities were found to be at u term ≈ 1150 m s − 1 for 100/0, u term ≈ 900 m s − 1 for 75/25 and u term ≈ 800 m s − 1 for 50/50 H 2 /CO. When leaving the obstructed part of the channel, the v elocities in the present study are decreasing, leading to lo wer terminal velocities, which are in order of the speed of sound of the products. 15 20 25 30 35 40 0 0 . 2 0 . 4 0 . 6 0 . 8 1 1 . 2 1 . 4 Fuel content x F in v ol.-% Relati ve terminal v elocity ˜ u t er m 100/0 H 2 /CO 75/25 H 2 /CO 50/50 H 2 /CO F ig. 4 : Relative terminal velocity ˜ u term over fuel content x F for BR30S100L The relati ve terminal v elocities for BR30S300L are sho wn in fig. 5. Similar to BR30S100 the fast flame and the detonation re gime can be distinguished. For 100/0 H 2 /CO ˜ u term is found in the f ast flame re gime at x F =15 v ol . -%. The transition to the detonation regime is found at x F = 17 . 5 v ol . -%. In case of 75/25 H 2 /CO the transition is less clear . For x F = 17 . 5 v ol . -% one out of four experiments is found in the detonation re gime. For x F = 20 v ol . -%, three out of four e xperiments reached the detonation re gime. For 50/50 H 2 /CO-fuels a clear transition is reached at a fuel content of x F = 22 . 5 v ol . -%. In the detonation re gime 100/0 H 2 /CO sho ws the highest relati ve terminal v elocity with little v ariation o v er the fuel content. The relati ve terminal v elocities for 75/25 are found similar to the ones obtained for 100/0 H 2 /CO. In contrast 50/50 H 2 /CO sho w lower relati ve terminal v elocities in the detonation re gime. In case of fuel rich mixtures, the transition to a fast flame is found between 35 vol . -% < x F < 40 vol . -% for 100/0 and 75/25 H 2 /CO. For 50/50 H 2 /CO ˜ u term is still found in the detonation re gime for x F = 40 vol . -% . Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 111 15 20 25 30 35 40 0 0 . 2 0 . 4 0 . 6 0 . 8 1 1 . 2 1 . 4 Fuel content x F in v ol.-% Relati ve terminal v elocity ˜ u t er m 100/0 H 2 /CO 75/25 H 2 /CO 50/50 H 2 /CO F ig. 5 : Relative terminal velocity ˜ u term over fuel content x F for BR30S300L In fig. 6 the relati ve terminal v elocities ˜ u term for the configuration BR60S100L are sho wn. Compared to BR30S100L and BR30S300L, ˜ u term remains significantly smaller than the Chapman-Jouguet ve- locities for the gi ven mixtures. In case of BR60S100L, ˜ u term ranges from ˜ u term = 0 . 14 to 0 . 33. The highest terminal v elocity is found at a fuel content of x F = 25 v ol . -%. A distincti v e influence of the fuel as sho wn in fig. 4 and 5 is not displayed. When compared to the speed of sound of the products a Pr , the results show , that almost all e xperiments are found in the fast flame re gime in the obstructed part of the channel. In the unobstructed part the flame is found belo w a Pr . Hence, the flame is de- celerating when entering the unobstructed part. This might be linked to reflected w av es from the end plate. A similar beha viour was observ ed by V ollmer et al. (2011) at a similar obstacle configuration. 15 20 25 30 35 40 0 0 . 2 0 . 4 0 . 6 0 . 8 1 1 . 2 1 . 4 Fuel content x F in v ol.-% Relati ve terminal v elocity ˜ u t er m 100/0 H 2 /CO 75/25 H 2 /CO 50/50 H 2 /CO F ig. 6 : Relative terminal velocity ˜ u term over fuel content x F for BR60S100L 3.2 Dynamic pr essur e Fig. 7 sho ws the maximum dynamic pressure p max for the configuration BR30S100L. The a v er - age peak pressure for all in v estigated fuel contents is found at ¯ p max = 28 . 16 bar for 100/0, ¯ p max = Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 112 24 . 78 bar for 75/25 and ¯ p max = 26 . 95 bar for 50/50 H 2 /CO. The lo west dynamic pressure is found at x F = 15 vol . -% for 75/25 H 2 /CO. As the fuel content is increased the peak dynamic pressure in- creases. When increasing the fuel content to x F = 17 . 5 v ol . -%, the peak pressure for 100/0 H 2 /CO almost doubles, while the increase for 75/25 and 50/50 H 2 /CO is small. For 75/25 and 50/50 a similar beha viour is obtained when the fuel content is increased to 20 v ol . -%. The sharp increase in p max matches well with the transition from fast flames to detonations as described in 3.1. The peak pres- sure further increases with increasing fuel content. The highest pressure is found at p max = 49 . 46 bar at x F = 27 . 5 vol . -% for 100/0 H 2 /CO. If the fuel content is further increased the peak pressure de- creases. Again the transition from detonation to fast flames agress well with the decrease in the peak pressure. While 75/25 and 50/50 H 2 /CO are still found in the detonation regime at x F = 35 v ol . -%, 100/0 H 2 /CO is in the fast flame re gime. This results in a lo wer dynamic pressure for 100/0 when compared to 75/25 and 50/50 H 2 /CO. For x F = 40 v ol . -% a similar beha viour is observ ed for 75/25 and 50/50 H 2 /CO. 15 20 25 30 35 40 0 10 20 30 40 50 Fuel content x F in v ol.-% Maximal pressure p max in bar 100/0 H 2 /CO 75/25 H 2 /CO 50/50 H 2 /CO F ig. 7 : Maximum dynamic pr essur e p max over fuel content x F for BR30S100L In fig. 8 the maximum dynamic pressure p max for the configuration BR30S300L is plotted against the fuel content x F . The a v erage pressure is ¯ p max = 24 . 66 bar for 100/0, ¯ p max = 24 . 17 bar for 75/25 and ¯ p max = 25 . 10 bar for 50/50 H 2 /CO slightly belo w the mean pressures obtained for BR30S100L. The lo west pressure is obtained for 75/25 H 2 /CO at x F = 15 v ol . -% at p max = 10 . 90 bar. The peak pressure increases as the fuel content is increased. A high increase is observ ed between x F = 15 vol . -% and x F = 17 . 5 vol . -% for 100/0 and 75/25, which might be related to the transition from fast flames to detonations in one of the e xperiments leading to a strong increase in peak pressure. A similar beha viour is obtained at x F = 20 v ol . -% as the experiments with detona tion increase the peak pressure. W ithin the detonation regime an increase in fuel content leads to an continuous rise in peak pressure. Compared to BR30S100L the increase in peak pressure is smaller . The highest peak pressure is found at a fuel content of x F = 35 vol . -% for 50/50 H 2 /CO. Although all fuels are found at the detonation re gime at x F = 35 vol . -% (see fig. 5), the peak pressure for 75/25 and 50/50 H 2 /CO is increased when compared to x F = 30 vol . -%. A similar increase in pressure was observ ed at the configuration BR60S300L at which a significant increase in dynamic pressure was obtained at x F = 35 v ol . -% for 50/50 H 2 /CO. In this case 50/50 H 2 /COwas the only fuel found at the detonation re gime for x F = 35 vol . -% (Heilbronn et al. (2019)), while for BR30S300L 75/25 H 2 /CO is found in the detonation re gime too. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 113 15 20 25 30 35 40 0 10 20 30 40 50 Fuel content x F in v ol.-% Maximal pressure p max in bar 100/0 H 2 /CO 75/25 H 2 /CO 50/50 H 2 /CO F ig. 8 : Maximum dynamic pr essur e p max over fuel content x F for BR30S300L In fig. 9 the maximum dynamic pressure for the configuration BR60S100L is displayed. The av erage pressure is ¯ p max = 10 . 17 bar for 100/0, ¯ p max = 10 . 06 bar for 75/25 and ¯ p max = 10 . 38 bar for 50/50 H 2 /CO. The lo west peak pressure is found at x F = 15 vol . -% for 100/0 H 2 /CO. When increasing the fuel content a slight increase in the peak pressure is found. The maximum peak pressure is obtained at x F = 25 vol . -% for 75/25 H 2 /CO at p max = 12 . 15 bar. Although the flames are found in the fast flame re gime at short run-up distances, DDT was not observ ed in this configuration (see sec. 3.1). This might be related to the detonation cell width. Due to the high blockage ratio and the lo w spacing, no detonations are obtained within the obstructed part of the channel. Hence the pressures obtained at BR60S100L are mainly the result of F A and therefore lo wer compared to BR30S100L and BR30S300L. 15 20 25 30 35 40 0 10 20 30 40 50 Fuel content x F in v ol.-% Maximal pressure p max in bar 100/0 H 2 /CO 75/25 H 2 /CO 50/50 H 2 /CO F ig. 9 : Maximum dynamic pr essur e p max over fuel content x F for BR60S100L 3.3 Run-up distance Since reaching the speed of sound of the isobaric comb ustion products a pr is one of the critical condi- tions for DDT , the run-up distance is a k ey parameter for safety applications (Ciccarelli & Dorofee v Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 114 (2008)). Fig. 10 displays the run-up distance for the BR30S100L configuration. The mean run-up distance is ¯ x a pr = 1 . 00 m for 100/0, ¯ x a pr = 1 . 20 m for 75/25 and ¯ x a pr = 1 . 18 m for 50/50 H 2 /CO. For v ery lean mixtures of x F = 15-17 . 5 vol . -% the run-up distance for 75/25 and 50/50 fuels is shorter than for 100/0 H 2 /CO. This trend is rev ersed as soon as a fuel content of x F = 20 v ol . -% is reached. W ithin the range of x F = 20-27 . 5 vol . -% the run-up distances of 75/25 and 50/50 H 2 /CO are v ery similar and about 0 . 4 m longer than for 100/0 H 2 /CO. For fuel rich mixtures another beha viour is observed. While for x F = 30-35 vol . -% all fuels beha ve quite similar , the gap between the run-up distance increases at 40 vol . -%. When compared to the results obtained by V eser et al. (2002), a shorter run-up distance is obtained. While V eser found a run-up distance of x a pr ≈ 1 . 5 m for 100/0, 2 m for 75/25 and 2 . 6 m for 50/50 H 2 /CO the run-up distances for the GraV ent configuration are shorter . This might be related to a slightly dif ferent geometry . The tube used by V eser used orifice plates of BR = 30 %. In contrast, the obstacles in the GraV ent configuration allow the flame to transit through the obstructed part through the whole channel width. 15 20 25 30 35 40 0 0 . 5 1 1 . 5 2 2 . 5 3 3 . 5 Fuel content x F in v ol.-% Run-up distance x a pr in m 100/0 H 2 /CO 75/25 H 2 /CO 50/50 H 2 /CO F ig. 10 : Run-up distance x a pr to speed of sound of the pr oducts a Pr over fuel content x F for BR30S100L The run-up distance for BR30S300L is sho wn in fig 11. The mean run-up distance is ¯ x a pr = 1 . 42 m for 100/0, ¯ x a pr = 1 . 30 m for 75/25 and ¯ x a pr = 1 . 38 m for 50/50 H 2 /CO. Similar to BR30S100L a shorter run-up distance compared to 100/0 is found for x F = 15 v ol . -% for 75/25 and 50/50 H 2 /CO. The run- up distance for 75/25 and 50/50 H 2 /COincreases significantly if the fuel content is increased to x F = 17 . 5 vol . -%. F or 100/0 H 2 /COthe run-up distance in a range of x F = 15-30 vol . -% sho w only v ery little v ariation. A similar trend can be observ ed for 75/25 and 50/50 in an range of x F = 22 . 5-27 . 5 vol . -%. In the fuel rich range of x F = 30-35 vol . -% an increase in the run-up distance is observ ed. This trend is re versed when the fuel content is further increased to x F = 40 v ol . -%. In total the run-up distance for BR30S300L is slightly longer than for BR30S100L, which might be related to less turbulence due to fe wer obstacles. The influence of the fuel is less clear . Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 115 15 20 25 30 35 40 0 0 . 5 1 1 . 5 2 2 . 5 3 3 . 5 Fuel content x F in v ol.-% Run-up distance x a pr in m 100/0 H 2 /CO 75/25 H 2 /CO 50/50 H 2 /CO F ig. 11 : Run-up distance x a pr to speed of sound of the pr oducts a Pr over fuel content x F for BR30S300L Fig 12 displays the run-up distance for BR60S100L. The mean run-up distance is ¯ x a pr = 0 . 80 m for 100/0, ¯ x a pr = 0 . 73 m for 75/25 and ¯ x a pr = 0 . 78 m for 50/50 H 2 /CO. At x F = 15 v ol . -% the shortest run-up distance is found for 100/0 H 2 /CO at x a pr = 0 . 41 m. When the fuel content is increased the run-up distance for 100/0 increases to x a pr = 0 . 90 m at x F = 27 . 5 v ol . -%. In fuel rich mixtures x a pr slightly decreases before increasing sharply at x F = 40 vol . -%. A similar trend is observed at 75/25 H 2 /CO. The slope of the run-up distance ov er the fuel content seems to be slightly shifted to higher fuel contents as the local maximum of the run-up distance is observed at x F = 30 v ol . -%. In contrast to 100/0 a short run-up distance is found at x F = 40 vol . -%. 15 20 25 30 35 40 0 0 . 5 1 1 . 5 2 2 . 5 3 3 . 5 Fuel content x F in v ol.-% Run-up distance x a pr in m 100/0 H 2 /CO 75/25 H 2 /CO 50/50 H 2 /CO F ig. 12 : Run-up distance x a pr to speed of sound of the pr oducts a Pr over fuel content x F for BR60S100L F or 50/50 the slope of the run-up distance is dif ferent than for 100/0 and 75/25 H 2 /CO. A local min- imum is observ ed at x F = 17 . 5 vol . -%. If the fuel content is increased, a marginal increase in x a pr is observ ed. In the range between x F = 17 . 5-27 . 5 vol . -%, 50/50 H 2 /CO-fuels ha v e the shortest run-up distances for the in vestigated fuels. A sharp increase is observ ed at x F = 30 vol . -%. By further increas- ing the fuel content, x a pr decreases and is shorter than for 100/0 H 2 /CO at x F = 40 v ol . -%. Compared Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 116 to BR30S100L and BR30S300L the run-up distances at BR60S100L are the shortest, due to the high blockage ratio and the lo w spacing. In general the trends of the run-up distance are dif ferent between the in vestig ated fuels. This is true especially for v ery lean mixtures of x F = 15-17 . 5 vol . -% at which 75/25 and 50/50 sho w shorter run- up distances than 100/0 H 2 /CO. This can not directly be explained by thermodynamic properties of the mixtures. The Le wis-Numbers are well belo w unity for all in vestigated fuels at lean fuel con- tents. Hence thermo-dif fusi ve instabilities, which lead to a higher inte gral fuel consumption should be present in all lean mixtures (Katzy et al., 2017). Furthermore, the total energy content as well as the e xpansion ratios σ are similar for each fuel at a gi ven fuel content (compare table 2). Hence, the dif ferent behaviour might be link ed to flow phenomena induced by the higher momentum of jets passing through obstacles in CO containing fuels. 4 Conclusion and Outlook Results of e xperimental studies on F A and DDT in homogenous H 2 -CO-air mixtures in a partially ob- structed channel ha v e been presented. Three different obstacle configurations ha ve been in v estigated: a lo w obstruction (30 %) at a small (100 mm) and lar ge spacing (300 mm), a high obstruction (60 %) at a small spacing in addition to the high obstruction and lar ge spacing which was published in Heil- bronn et al. (2019). Each obstacle configuration featured a 3 . 95 m long obstructed and a 2 . 05 m long unobstructed section in a rectangular e xplosion channel. The fuel was v aried between 100/0, 75/25 and 50/50 H 2 /CO. The fuel content was v aried from x F =15-40 v ol . -% in air . The experiments were analysed with re gard to terminal v elocity , peak pressure and run-up distance to the speed of sound of the products. Whether detonations occur depends on the obstacle configuration, the fuel and the fuel content in air . F or high obstruction and small spacing, detonations are not observed. For a lo wer obstruction detona- tions occur at slightly leaner conditions at 100/0 compared to 75/25 and 50/50 H 2 /CO. A similar trend is observ ed at a less frequently obstructed geometry . A significant change in the detonation limits is found in fuel rich mixtures. In lo w obstructed obstacle configurations fuels of 75/25 and 50/50 are found in the detonation regime, while 100/0 H 2 /CO remain in the fast flame re gime. This supports the findings in Heilbronn et al. (2019), sho wing that H 2 /CO-fuels show a dif ferent behaviour than 100/0 H 2 /CO in fuel rich mixtures. In addition to V eser et al. (2002) it can be concluded that CO addition can ha v e an impact in fuel rich mixtures. The highest peak pressures are found at a lo w obstruction and a small spacing. For rich fuel mixtures, peak pressures of 75/25 and 50/50 H 2 /COin lo w obstructed configurations are higher than the ones observ ed for the same fuel contents of 100/0 H 2 /CO, hence leading to a higher potential damage. On a v erage, peak pressures at lo w obstructions are slightly lo wer for a lar ger spacing. Since DDT was not observ ed a high obstruction and a small spacing, the measured peak pressures are very similar for all fuels and contents. Compared to a lo wer obstruction, the peak pressures are found to be lo wer . The longest run-up distances to the speed of sound of the products were found at a lo w obstruction and a lar ge spacing. In this case, all fuels sho wed a similar run-up distance. For a smaller spacing, run-up distances for 75/25 and 50/50 were slightly shorter than for 100/0 H 2 /CO. The shortest run-up distances were found for a high obstruction and a small spacing. The influence of the CO-content in the fuel on the run-up distance could not be identified by the measurement systems used for the present study . Hence are more detailed in vestigation of the early stages of flame propag ation using shado wgraphy and OH-PLIF will be used in order to gain further insights. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 117 Acknowledgements This project is funded by the German Federal Ministry of Economics and T echnology (BMW i) on the basis of a decision by the German Bundestag (project no. 1504545A) which is gratefully acknowl- edged. Refer ences Barfuss, C., Heilbronn, D., Sattelmayer , T . (2019). Simulation of deflagr ation-to-detonation transiti on of lean h2-co-air mixtur es in obstructed channels . In International Confer ence on Hydr ogen Safety . Barfuss, C., Heilbronn, D., Sattelmayer , T . (2020). Impact of local flame quenc hing on the flame acceler ation in h2-co-air mixtur es in obstructed channels . In 13th International Symposium on Hazar ds, Pr evention, and Mitigation of Industrial Explosions . Bro wne, S., Ziegler , J., Shepherd, J. (2008). Shoc k and detonation toolbox . GALCIT -Explosion Dynamics Laboratory , Pasadena, CA. Ciccarelli, G., Dorofee v , S. (2008). Flame acceleration and tr ansition to detonation in ducts . Progress in ener gy and comb ustion science, 34(4):499–550. Demirbas, A. (2005). P otential applications of r ene wable ener gy sour ces, biomass combustion pr ob- lems in boiler power systems and comb ustion r elated en vir onmental issues . Progress in Energy and Comb ustion Science, 31(2):171 – 192. ISSN 0360-1285. Gauntt, R., Kalinich, D., Cardoni, J., Phillips, J., Goldmann, A., Pickering, S., Francis, M., Robb, K., Ott, L., W ang, D., et al. (2012). Fukushima daiichi accident study (status as of april 2012) . Sandia Report Sand, 6173. Goodwin, D., Speth, R., H.K., M., W eber , B. (2018). Canter a: An object-oriented softwar e toolkit for c hemical kinetics, thermodynamics, and transport pr ocesses . https://www .cantera.or g. V ersion 2.4.0. Heilbronn, D., Barfuss, C., Sattelmayer , T . (2019). Deflagr ation-to detonation tr ansition in h2-co-air mixtur es in a partially obstructed channel . In International Confer ence on Hydr ogen Safety . Katzy , P ., Hasslber ger , J., Boeck, L. R., Sattelmayer , T . (2017). The ef fect of intrinsic instabilities on ef fective flame speeds in under -r esolved simulations of lean hydr og en–air flames . Journal of Nuclear Engineering and Radiation Science, 3(4). K umar , R., K oroll, G., Heitsch, M., Studer , E. (2000). Carbon monoxide–hydr ogen comb ustion char - acteristics in se ver e accident containment conditions . Or ganisation for Economic Co-operation and De velopment, P aris, France, Report No. NEA/CSNI. Lieuwen, T ., Y ang, V ., Y etter , R. (2009). Synthesis gas comb ustion: fundamentals and applications . CRC press. V eser , A., Stern, G., Grune, J., Breitung, W ., Bur geth, W . (2002). Co-h2-air combustion tests in the fzk-7m-tube . T echnical report, Institut für Kern- und Ener gietechnik. V ollmer , K. G., Ettner , F ., Sattelmayer , T . (2010). Influence of concentr ation gr adients on flame acceler ation in tubes . In 8th International Symposium on Hazar ds, Pr evention, and Mitigation of Industrial Explosions . V ollmer , K. G., Ettner , F ., Sattelmayer , T . (2011). Influence of concentr ation gr adients on flame acceler ation in tubes . Science and T echnology of Energetic Materials, 72(3-4):74–77. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 118 Independent whether the flame under goes DDT or remains in the chocked flame re gime, the flame accelerates to the maximum v elocity possible in a deflagrati ve comb ustion, i.e. Chapman-Jouguet condition behind the flame. The flame tip velocity in this state is close to the speed of sound in the comb ustion products a pr (Lee, 2008). The run-up distance to the speed of sound in the products x a pr is characteristic for the flame acceleration phase (Ciccarelli and Dorofee v, 2008). It also represents the intensity of acceleration and, hence, gi ves an idea of the DDT probability . If the distance is shorter , w a ves are emitted from the flame at a higher rate and stronger accumulated pressure wa ves e volv e leading to an earlier onset of detonation. While flame trajectory results hav e been reported pre viously for the GraV ent test rig (Barfuss et al., 2019), here, the influence of v arying blockage ratios and spacings shall be easily apparent. Therefore, the characteristic x a pr is used for comparison of numerical and e xperimental results. In table 3 the run-up distances of the dif ferent quenching limit treatments are listed together with e xperimental findings from the GraV ent e xperiments. Additionally , a relati v e de viation of the two approaches, dimensional term and interpolation table, is presented. The de viation is normalized by the the e xperimental v alue. T able 3 : Simulated run-up distances x a pr of dif fer ent strain limit tr eatments and experimental values for 22.5 vol.-% fuel in air . T est rig configuration Dimensional term Interpolation table Quenching neglected Rel. deviation g cr , Dim − g cr , 1D Experiment BR60S100L 1.012 0.9319 0.9297 10.73 % 0.7464 BR60S300L 1.133 0.9523 0.9523 17.82 % 1.0139 BR30S100L 1.252 1.2510 1,2510 0.08 % 1.3215 BR30S300L 1.280 1.3430 1.3700 4.62 % 1.3645 The e xperimental results in table 3 are a v eraged v alues of three repetitions with the same experimen- tal setup (Heilbronn et al., 2020). All numerical results remain within the confidence range of the e xperimental findings. The run-up distances obtained with the interpolation table approach and with ne glected quenching are almost identical for all obstacle configurations. This suggests that flame stretch limits from 1D simulations are almost ne ver reached. This is due to the much coarser grid in the RANS based simulations in comparison to the DNS simulation of figure 3. In contrast, suf ficient strain e xists for the smaller limit from the dimensional quenching term resulting in a generally fur - ther do wnstream position to reach a pr . As an exception, the flame tip velocity in the configuration BR30S300L rises abo v e a pr in the jet passing an obstacle, but drops noticeable belo w a pr during the ex- pansion after the obstacle. Hence, a Chapman-Jouguet deflagration has not been reached sustainably and further acceleration due to increasing heat release is still possible. Results from v arying quenching limit approaches differ most at lar ge blockage ratios. V ariation of obstacle spacing has a smaller impact on the contrib ution of quenching in comparison to blockage ratio v ariations. Since numerical results without quenching or with interpolation tables better match the e xperiments, it is concluded, that flame quenching can be neglected in small scale geometries. Due to the greater de viation from experiments, the e v aluation of g cr with the dimensional term is not considered suitable for accelerating H 2 -CO-air flames with a Le wis number less than unity . The ac- celerating and, hence, stabilizing ef fects on a flame by a Le wis number less than unity are not captured suf ficiently by the v ariables of the dimensional approach (Poinsot and V eynante, 2005). In contrast the interpolation table approach can be used, in order to maintain a threshold of maximum turb ulence intensity leading to quenching, which is almost ne ver reached in the RANS based simulations in ves- tigated in this study . Application of ef ficienc y functions correcting for the spectrum contrib uting to flame stretch do not appear to be useful, because in general coarse meshes are used in safety analysis in vestigations (Mene v eau and Poinsot, 1991). Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 125 2.2.2 Impact of geometry dimensions Realistic accident scenarios often happen on much lar ger geometrical dimensions than the GraV ent test rig. The impact of quenching has to be in v estigated on medium-scale or lar ge-scale geometries also. Therefore, the GraV ent configuration BR60S100L has been scaled-up by the factor ten in its geometrical dimensions with respect to the cross section. The dimensional scale-up is denoted with the abbre viation BR60S1000 in the follo wing. The length of the channel is 9 m and the obstructed section at 5.5 m. Hence, the number of obstacles reduces to 6. Simulations with the two quenching limit treatments, the dimensional term approach and the interpolation table, ha ve been conducted with the same gas mixture of 22.5 v ol.-% fuel in air (75/25 H 2 /CO). The initial mesh of the scale-up consists of cubic cells with a cell size of ∆ x cell = 40 mm. This roughly relates to the initial mesh of the small-scale GraV ent in vestigation after adapti ve mesh refinement is applied twice. T able 4 presents the e v aluated run-up distances x a pr . In Case BR60S1000 , both simulations e xpe- rienced a subsequent DDT after reaching a pr . Because there is no experimental v erification in the scale-up geometry , only the relati ve deri v ation can be discussed. The de viation is normalized here with the run-up distance obtained from the interpolation table approach. T able 4 : Run-up distances x a pr in the generic Gr aV ent BR60S1000 scale-up. T est rig configuration Dimensional term Interpolation table Relativ e deviation BR60S100L 1.012 0.932 8.69 % BR60S1000 4.055 4.005 1.25 % The relati ve de viation is se ven times smaller in the scale-up than in the original configuration. The smaller de viation in comparison to the original GraV ent experiment can be e xplained by the lar ger a v eraged cell size relating to lo wer strain rates. Because turbulence does not scale linearly with the geometrical dimension, it can be assumed that volumes of unb urned gas behind obstacles isolated by potential shear layers do not scale linearly either . According to this assumption, the impact of quenching on the flame trajectory decreases in lar ger geometries. The smaller de viation between the two stretch limit treatments in the scale-up is proof of that beha vior . Strain rates from both modeling approaches are not high enough to result in significant flame quenching here. 3 Conclusions In this study , the presence of local flame extinction by flame stretch has been presented for H 2 -CO-air mixtures under going flame acceleration to fast flames. DNS simulations hav e been carried out for 22.5 v ol.-% H 2 in air as well as for 75/25 v ol.-% H 2 /CO in air . The simulation results sho w ho w v ortices, shedding from the obstacle edges, stretch the flame and separate sections of the flame from the b ulk flame. The strain rate gro ws in smaller vortices leading to strong stretching of the flame. Locally the flame’ s reacti vity is reduced or it e ven e xtinguishes. Starting from flame tip velocities of 300 m/s, first vortices were observ ed, which ef ficiently quench the flame. W ith increasing velocities the intensity of said v ortices is assumed to gro w . Furthermore, in CO containing fuels the momentum of the jet passing the same obstacles is higher than in pure H 2 fuels. The increased momentum allo ws for equally strong strain as in H 2 fuels at lo wer flame tip velocity . As a consequence, the observ ed impingement depth of v ortices shedding from the rear edge of the obstacle is larger in comparison to pure H 2 fuel. The ef fect of isolating a fresh gas in a pocket behind obstacles is therefore stronger in CO containing fuel. F or the purpose of an application oriented modeling approach for safety analysis, three dif ferent approaches for e v aluating critical stretch rates are in vestigated in an RANS based compressible CFD solv er . The CFD solv er has already been v alidated for subsequent F A and DDT of H 2 -CO-air mixtures in one geometry configuration (Barfuss et al., 2019). A variation of obstacle spacing and blockage ra- tio allo ws for a more general statement on the necessity to consider quenching in ef ficient comb ustion Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 126 modeling approaches. Simulated run-up distances to the speed of sound in the comb ustion products hav e been compared with e xperiments. Simulations neglecting quenching sho w better agreement than simulations with Zimont’ s dimensional term for the flame stretch limit. While run-up distances of the different g cr approaches significantly de viate in small-scale geometries with high blockage ratios, the differences become irrele v ant on medium to lar ge-scale geometries. Therefore, the combustion phenomenon of partial flame quenching does not need to be taken into account in TFC comb ustion models for accelerating H 2 -CO-air flames in obstructed channels. Even though it can be ne glected, the authors suggest to still apply a quenching term to maintain an upper limit of turb ulent intensity as DNS data has sho wn general presence of partial quenching. A good practice for H 2 -CO-air flames appears to be using strain rate limits of flame extinction from 1D counter -flo w flame simulations with detailed chemistry , which take the increasing pressure of transient flame acceleration into account. Acknowledgements The presented work is funded by the German Federal Ministry of Economic Af fairs and Ener gy (BMW i) on the basis of a decision by the German Bundestag (project no 1501545A) which is grate- fully ackno wledged. Refer ences Barfuss, C., Heilbronn, D., Sattelmayer , T . (2019). Simulation of deflagration-to-detonation tr an- sition of lean h2-co-air mixtur es in obstructed c hannels . In Pr oceedings of the 8th International Confer ence on Hydr og en Safety . Adelaide, Australia. Ciccarelli, G., Dorofee v , S. (2008). Flame acceleration and tr ansition to detonation in ducts . Progress in Ener gy and Comb ustion Science, 34(4):499 – 550. ISSN 0360-1285. doi: https://doi.or g/10.1016/j.pecs.2007.11.002. Da vis, S. G., Joshi, A. V ., W ang, H., Egolfopoulos, F . (2005). An optimized kinetic model of h2/co comb ustion . Proceedings of the combustion Institute, 30(1):1283–1292. Dorofee v , S. B., K uznetso v , M. S., Aleksee v , V . I., Efimenko, A. A., Breitung, W . (2001). Evalu- ation of limits for ef fective flame acceler ation in hydr og en mixtur es . Journal of Loss Pre v ention in the Process Industries, 14(6):583 – 589. ISSN 0950-4230. doi:https://doi.org/10.1016/S0950- 4230(01)00050-X. Gamezo, V . N., Oga wa, T ., Oran, E. S. (2007a). Effect of obstacle spacing on flame acceler ation and ddt in obstructed c hannels . 21st ICDERS. Gamezo, V . N., Oga wa, T ., Oran, E. S. (2007b). Numerical simulations of flame pr opagation and ddt in obstructed c hannels filled with hydr ogen–air mixtur e . Proceedings of the Comb ustion Institute, 31(2):2463–2471. Goodwin, D. G., Speth, R. L., Mof fat, H. K., W eber , B. W . (2018). Canter a: An object- oriented softwar e toolkit for chemical kinetics, thermodynamics, and transport pr ocesses . doi: 10.5281/zenodo.1174508. V ersion 2.4.0. Hasselber ger , J. (2017). Numerical Simulation of Deflagr ation-to-Detonation T r ansition on Industry Scale . Ph.D. thesis, TU München. Heilbronn, D., Barfuss, C., Sattelmayer , T . (2019). Deflagr ation-to detonation tr ansition in h2-co-air mixtur es in a partially obstructed channel . In Pr oceedings of the 8th International Confer ence on Hydr og en Safety . Adelaide, Australia. Heilbronn, D., Barfuss, C., Sattelmayer , T . (2020). Influence of geometry on flame acceler ation and ddt in h2-co-air mixtur es in a partially obstructed channel . In In Pr oceedings of 13th International Symposium on Hazar ds, Pr evention, and Mitigation of Industrial Explosions . Braunschweig, Ger - man y . In print. Lee, J. H. S. (2008). The Detonation Phenomenon . Cambridge Uni v ersity Press, Cambridge. ISBN 978-0-521-89723-5. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 127 Li, X., Y ou, X., W u, F ., Law , C. K. (2015). Uncertainty analysis of the kinetic model pr ediction for high-pr essur e h2/co combustion . Proceedings of the Comb ustion Institute, 35(1):617–624. Mene veau, C., Poinsot, T . (1991). Str etching and quenc hing of flamelets in pr emixed turbulent com- b ustion . Combustion and Flame, 86(4):311–332. Menter , F . R. (1994). T wo-equation eddy-viscosity turb ulence models for engineering applications . AIAA journal, 32(8):1598–1605. Poinsot, T ., V eynante, D. (2005). Theor etical and numerical comb ustion . R T Edwards, Inc. T oro, E. F ., Spruce, M., Speares, W . (1994). Restoration of the contact surface in the hll-riemann solver . Shock wa v es, 4(1):25–34. V elikorodny , A., Studer , E., Kudriak o v , S., Beccantini, A. (2015). Combustion modeling in lar ge scale volumes using eur ople xus code . Journal of Loss Prev ention in the Process Industries, 35:104–116. Xiao, J., Breitung, W ., Kuznetso v , M., Zhang, H., T ravis, J. R., Redlinger , R., Jordan, T . (2017). Gasflow-mpi: A new 3-d par allel all-speed cfd code for turbulent disper sion and comb ustion sim- ulations: P art i: Models, verification and validation . international journal of hydrogen ener gy , 42(12):8346–8368. Zimont, V . L. (2000). Gas pr emixed combustion at high turb ulence. turb ulent flame closur e combus- tion model . Experimental thermal and fluid science, 21(1-3):179–186. Zimont, V ., Polifke, W ., Bettelini, M., W eisenstein, W . (1998). An ef ficient computational model for pr emixed turbulent comb ustion at high r e ynolds numbers based on a turb ulent flame speed closur e . Journal of engineering for gas turbines and po wer , 120(3):526–532. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 128 Obser vations of DDT in narr o w channels Yv es Ballossier , Josué Melguizo-Gavilanes & Florent V irot Institut Pprime, UPR 3346 CNRS, ISAE–ENSMA, Futuroscope Chasseneuil, France E-mail: [email protected] Abstract Experiments are conducted in a smooth 10 × 10 mm square cross-section, 1-m long channel, closed at the ignition end and open at the other end. Simultaneous two-direction schlieren visualization is used to in vestigate the three-dimensional dynamics of transition to detonation for a stoichiometric H 2 -O 2 mixture. Results sho w the e xistence of two distinct structures before detonation onset: (i) asymmetric, composed of an oblique shock trailed by a flame, that runs preferentially along the wall, and seems to get ignited inside the boundary layer de veloped by the precursor shock; (ii) symmetric, referred to as str ange wave in literature, propagating at the speed of sound in comb ustion products. The combined ef fect of shock induced preheating and viscous heating near walls seem to be responsible for the formation of the complex flame-shock interactions observ ed. Furthermore, simultaneous two- direction optical access allo ws to map the exact location of detonation onset, sho wing that 78 % of cases e xploded in corners, highlighting the role of corner flows and boundary layers in transition to detonation at this scale. K eywords: DDT , narr ow smooth channels, hydr o gen safety , flow visualization, strang e wave 1 Intr oduction T o comply with international agreements limiting CO 2 emissions, dif ferent solutions hav e been pro- posed to reduce the share of fossil fuels in electricity generation and transportation (Masson-Delmotte et al., 2018). The increased used of rene wable (e.g. wind, solar) requires efficient/lasting storage so- lutions due to its intermittent nature. Among them battery banks or electrolyzers to produce hydrogen (H 2 ). H 2 has the potential to be a ke y player in enabling a sustainable ener gy transition. Its widespread use in transportation, for instance, poses risks due to leaks in refueling stations for vehicles (such as the recent accident in Sandvika, Norway) or in fuel cell/battery enclosures that can lead to strong e xplosions via deflagration-to-detonation transition (DDT). Note that the typical gaps in these com- partments are of the order of millimeters, therefore understanding of DDT at this scale is needed to mitigate the aforementioned hazards through impro v ed, research-based design. T o wards that goal, in general, we focus on a canonical configuration, namely a 1-m long, optically accessible, square cross-section channel (10 mm × 10 mm) to study all stages of DDT : from ignition, flame propagation and acceleration, to transition to detonation. Ho we v er , here, we are only interested in characterizing the dynamics before and during detonation onset (i.e. formation of flame-shock comple x es) and re vealing their three dimensional structure. Studying this stage is essential to under - stand ho w to promote/mitigate DDT in narro w channels. Pre vious numerical and experimental work on flame-shock interaction by (Khokhlo v et al., 1999, Thomas et al., 2001) ha v e identified interesting structures that lead to DDT ; Y anez and Kuznetso v (2016) also observed similar structures b ut using narro w channels in which very wrinkled funnel-shape flames trail behind planar shocks. These quasi- steady flame-shock comple xes propag ate at speeds of the order of the speed of sound in comb ustion products; the authors called this re gime str ange wave . In the present work, tw o channels are used to study the formation of strange w a ves. They allo w one- and two-direction simultaneous schlieren visualization, respecti v ely . 13 t h International Symposium on Hazar ds, Pr evention and Mitigation of Industrial Explosions Braunsc hweig, GERMANY - J uly 27-31, 2020 ISHPMIE 2020 Braunschweig, Germany Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 129 2 Experiments 2.1 Apparatus and pr ocedur e The e xperimental apparatus was designed to allo w fle xibility re garding optical access. Its total length is 1030 mm; 91 % of which is optically accessible. T o switch from one-direction (1-D V) to two- direction visualization (2-D V) only the center piece of the channel needs being interchanged (see cross sectionals vie w in Fig. 1). In both configurations, 15-mm thick polycarbonate plates are used as windo ws. Aluminum flanges on top and bottom a v oid bending of the polycarbonate during detonation propagation. For the tw o direction configuration, spacers pre vent bending on the sides (one of them is visible in Fig. 8 bottom vie w). None of the tests performed sho wed signs of deformation. Structural inte grity is ensured with bolts scre wed on the aluminum in a "sandwich" configuration. The e xperiments were conducted as follo ws: (i) the channel was v acuumed to an absolute pressure belo w 1 mbar (controlled by a MKS 220D A pressure sensor); (ii) a plastic cap, held on a serv o motor’ s arm, sealed the open end of the channel; (iii) subsequently , the reacti v e mixture (stoichiometric H 2 - O 2 ) was fed into the channel until it reached atmospheric pressure. (i v) At this point, the plastic cap was automatically remo ved by the serv o motor; one second later the mixture is ignited ( E ign < 1 mJ) at the closed end by the electric arc that forms between the electrodes. This procedure ensured adequate control of the mixture composition ( Φ = 1 ± 0 . 01). The waiting time between filling and ignition w as also carefully controlled; 15 s were found suf ficient to a void an y impact of initial flo w unsteadiness on early stages of flame propagation. After each test, the channel was v acuumed for a fe w minutes to remo v e visible condensation from b urned gases. Closed end (igniti on) Open end 10 mm 10 mm 10 mm 10 mm Aluminum fl anges Electrodes sup port Center piece aluminum part Polycarbonate optical access F ig. 1 : Schematic of narr ow channel (top) and cr oss-sections for 1-D V (bottom left) and 2-D V (bottom right). Red squar es indicate ef fective section of the c hannel; white spaces optically accessible . 2.2 Flow visualization Schlieren visualization (SV) is performed on both channels. T wo sets of cameras were used: (i) T w o black and white Photron F ASTCAM SA-Z set at 140k frame per second (fps) (158 ns e xpo- sure), together with (248 ns exposure) for 2-D V ; (ii) Shimadzu HPVX camera, with acquisition rates of 1 million (500 ns e xposure) to 10 million fps (100 ns e xposure), together with a Shimadzu HPV2 camera set at 1 million fps (500 ns e xposure) for 2-D V . Dif ferent light sources were utilized, two 100-mm diameter collimated LED, and a mercury light, coupled with a collimating mirror also of 100 mm in diameter . While the collimated LED of fers compactness, the use of a mercury light pro vides a smaller focal point that results in a lar ger range of observ able gradients (the side vie w in Fig. 8 is the only case using the latter). For the remaining Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 130 cases sho wn, the collimated LED set-up is used with a razor blade occulting 50 % of the light to highlight gradients in the direction of flame propagation. Our schlieren set-up allo ws to visualize a section up to 100 mm; 2-D V is performed on the same section when done simultaneously . Figure 2 sho ws a simplified schematic of the global optical set-up. T o trigger high speed camera(s), a BPW34 photo-diode, placed before the section of interest detects the arri v al of the flame. Collimated whit e LED/ Mercury li ght Lens Filter 45 ° mirror Side view Bottom view Channel F ig. 2 : Schematic of global optical set-up for one- and two-dir ection visualisation. 3 Results and discussion F or the sake of clarity in the presentation of the results that follo w , Fig. 3 presents a schematic of the visualization orientations. Note that “Bottom” and “Side” faces refers to the relati ve position of the cameras; the actual position of the bottom face along the z − axis (or side face along y − axis) is arbitrary , as SV only provides inte grated information along the line of visualization. The chosen rep- resentation is con venient when interesting dynamics occur in the lo wer right corner as seen from the ignition end. Post-processing typically consists of contrast adjustment, and sometimes, background subtraction to highlight important flo w features. A total of 56 tests were performed of which 26 used 1-D V and 30 used 2-D V . Propagation direction Bottom face Side face y z x Ignition end F ig. 3 : Schematic of the visualization dir ections. “Side” and “Bottom” faces r efer to the camera position. 3.1 Strang e wave dynamics 3.1.1 Early stages and formation Figure 4 sho ws the structure observed during formation of the strange wa ve using 1-D V . T ime 0 µ s is arbitrary , and simply represents the first instance in the sequence. The tip of a very corrugated Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 131 flame (A) is seen to propagate from left to right at t = 0 µ s. Expansion of b urned products and the acceleration of the flame itself, generate perturbations within the fresh but shock ed mixture ahead of the flame (weak (B) and strong (C) pressures wa ves are visible in this frame). At 27.5 µ s, strength- ening of flame acceleration and successi ve preheating from precursor shocks/pressure w a ves result in ne w , stronger shocks, forming closer to the flame. The flame starts to show a rather symmetric shape, preferentially b urning near walls, resembling a funnel-shape (D) presumably filled with un- b urned mixture, finally , the flame catches up with the shock at 55 µ s (not sho wn as it is outside of the vie wing windo w). Note that an e xplosion emanating some where on the bottom face is observ ed, remnants of which are visible as a discontinuity tra v eling left (E). Not enough information is a v ailable to tell whether this local e xplosion lead to DDT . Front speeds computed from the images v ary from 1200 − 1500 m/s for the corrugated flame, and from 700 − 1000 m/s, for the precursor shock. t = 0 µ s Side x z t = 27.5 µ s Side x z t = 55 µ s Side x z F ig. 4 : 1-D V : symmetric formation of strang e wave; section 454 to 544 mm fr om ignition; conditions: stoic hiometric H 2 -O 2 at p 0 = 100 kP a, and T 0 = 290 K. Raw imag es. A: corrugated fr ont; B: weak pr essur e waves; C: pr ecursor shoc k; D: funnel-shaped flame surface; E: left moving discontinuity . t = 0 µ s Side x z t = 28 µ s Side x z t = 55 µ s Side x z F ig. 5 : 1-D V : asymmetric formation of strange wave; section 454 to 544 mm fr om ignition; condi- tions: stoichiometric H 2 -O 2 at p 0 = 100 kP a, and T 0 = 290 K. Raw ima ges. F: oblique shoc k; G: second ignition. Interestingly , in some cases using the same configuration and vie wing section, another structure was observ ed (see Fig. 5). In this case no perturbations seem to be visible ahead of the flame, instead, a front propagating within the boundary layer at the bottom w all forms an oblique shock (F). Upon reflection from the top wall, a second ignition within the boundary layer tak es place (G). On the last Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 132 frame, at 55 µ s, a symmetric shape starts to form. This shape is very similar to that observ ed experi- mentally by Thomas et al. (2001) and numerically by Khokhlov et al. (1999), and Oran et al. (2002) when the y studied shock-flame interactions. The precursor shock in this case propagates at the same speed as that reported abo v e. t = 0 µ s Bottom x y Side x z t = 7.1 µ s Bottom x y Side x z t = 14.2 µ s Bottom x y Side x z F ig. 6 : Simultaneous 2-D V of the formation of strang e wave; section 281 to 353 mm fr om ignition; conditions: stoichiometric H 2 -O 2 at p 0 = 100 kP a, and T 0 = 290 K. Backgr ound substracted to highlight important flow featur es. H: left mo ving discontinuity . Figure 6 sho ws 2-D V during formation of the strange wa ve. Note that in this channel this stage of DDT occurred in a dif ferent section, 281 mm ≤ x ≤ 383 mm, instead of 454 mm ≤ x ≤ 544 mm. At t = 0 µ s, this configuration sho ws a similar structure as in Fig. 4 preceded by a succession of pressure wa ves. The main dif ference lies on the fact that the corrugated flame does not catch up with the pertur - bations, instead the mixture seems to ignite within the boundary layer , likely due to a flame-boundary layer interaction, resulting in a front that propagates near the wall (similar to that seen in Fig. 5). Si- multaneous SV allo ws to pinpoint the exact location in the channel cross-section where ignition takes place (bottom right corner looking from the ignition end). At 7.1 µ s, the front in the boundary layer propagates f aster than the corrugated flame ( ∼ 3000 m/s), subsequently it decelerates to 2400 m/s, between 7.1 and 14.2 µ s. In the side view , at 14.2 µ s, another fast front starts to propagate within the boundary layer at the lo wer edge of this frame. The ignition of this front seems to be the result of an oblique shock-boundary layer interaction. Ho we v er , based on a preliminary surve y of all the results collected, ignition within the boundary layers arise more frequently due to flame-boundary layer interactions. The exact reason of spontaneous ignition within the corner is not clear , additional tests and analysis are currently being carried out to rule out uncertainties concerning manufacturing imperfections. The DNS of Dziemi ´ nska and Hayashi (2013) and LES of Zhao et al. (2017) present a similar scenario of DDT , b ut in our case, the flame-shock complex propagates more than 100 mm before it actually transits to detonation. A discontinuity (H), similar to that seen in Fig. 4 at 55 µ s, is also visible at 7.1 and 14.2 µ s. The ov erall dynamics obtained closely resembles that for 1-D V . Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 133 3.1.2 Quasi-steady pr opagation t = 0 µ s Side x z t = 10 µ s Side x z t = 20 µ s Side x z F ig. 7 : 1-D V of strang e wave; section 454 to 544 mm fr om ignition; conditions: stoichiometric H 2 -O 2 at p 0 = 100 kP a, and T 0 = 290 K. Raw images. The structure of the strange wa ve is visible in Fig. 7. It is very similar to that observ ed by Y anez and K uznetso v (2016), except that in our case it propagates at a speed ∼ 29% higher ( ∼ 1800 m/s). As mentioned abo ve, it is composed of a flat shock with a very wrinkled funnel-shaped flame surface responsible for sustaining the very high b urning rates observed. The 2-D V shown in Fig. 8, re veals additional information. At 0 µ s, the Bottom vie w and Side vie w e xhibit v ery dif ferent shapes. From the side, a similar shape as that observed in the first configuration, with a planar shock ahead of a funnel-shaped flame, whereas from the bottom, only a planar shock is visible, just before the black spacer . Even though this flame-shock comple x is supersonic with respect to reactants ahead, it is not the quasi-steady strange wa ve observ ed by Y anez and Kuznetso v (2016). The planar shock propagates around 1200 m/s and it is quickly caught up by the funnel-shaped flame. When the y mer ge near the center , at 18 µ s, the comple x accelerates to ∼ 1500-1600 m/s and starts to flatten, sometimes staying curv ed in one corner until detonation onset. The Bottom vie w at 18 µ s, is evidence of the con v oluted three-dimensional structure that forms when both oblique flame-shock comple x es mer ge. 3.1.3 Detonation onset T ransition to detonation systematically occurs follo wing the strange wa ve structure described abo v e. Figure 9 sho ws the results obtained using the 1-D V configuration. An ignition kernel de velops at the bottom wall at t = 0 µ s. The initially planar shock becomes curv ed, as the shock generated from the e xplosion tra v els to wards the top wall, consumi ng the fresh shocked mixture present do wnstream of it. Finally , the frame at 3.2 µ s, sho ws a quasi-planar shock of an incipient detonation. Run-up distances computed from the schlieren images (i.e. DDT location) lied between 38 - 67% of the total length of the channel (sample size: 14 tests). The high temporal resolution used allo wed us to resolve the detonation onset in great detail (5 million fps). Figure 10 sho ws the results obtained with 2-D V configuration, follo wing the flattening of the comple x described in Fig. 8, whose propagation speed is ∼ 2000 m/s, a strong explosion tak es place. Note that to be consistent with the schematic in Fig. 3, the planes x − y and x − z were translated, and subse- quently rotated o ver the x − axis; the DDT point is actually in the top left corner . Run-up distances lied between 42-47% of total length (sample size: 20 tests). SV performed on the same section for 1-D V configuration (454-544 mm) include four dif ferent struc- tures (Figures 4, 5, 7 and 9); fully de veloped detonations, as well as slo wer deflagrations were also observ ed b ut not included here. The 2-D V channel, on the other hand, provided repeatable results Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 134 Figure 2 . Li -ion cell physical formats. Energy storag e speci fic l i-ion cells tend to be phy sically larger than consumer cells because th eir size is not limited by the need to fit within consumer product enclosures, though there a re manufacturers who do develop ene rgy storage systems (ESS) from cells used in consumer products. The fiv e materi al components of a common li -ion cell consist of a cathode material, an anode material, a separa tor material between the two electrodes, an electrolyte, and an outer case. To manufacture a cell, cathode, separ ator, an d anode mate rials are e ither wound together simultaneously or a re as sembled in a stack of alternating layers (Schröder et al., 2017). This “winding” or “jellyroll” is inserted into a case manufactured from aluminium or steel, or int o metallicized polymer pouch, depending on the cell format. I n orde r to ena ble winding or stacking, cathode and anode m ater ials are coated onto substra tes of alumini um and copper foil, respectively (Warner, 2015). Electrol ytes are injected into any of these formats at the conclusion of assembly (Schröder et al., 2017). (a) (b) Figure 3. a) Cells enclosed within a module housing, b) open and enclosed ESS units. Cells are electrically inte rconnec ted in series and parallel from s everal cells to thousands of cells to achieve increases in v oltage and output current. A manufactured group of cells that are hous ed within an enclosure that typically includes hardware (e.g., fuses) and s oftware (e.g., battery Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 141 management) electrical s afety fe atures is referred to as a module. Modules c ontain tens to hundr eds of kilograms of cells and may or may not have th erma l runaway propagation mitigating features. Figure 3 (a) shows a schematic of a module with cylindrical cells designed by UL to demonstrate the potential for cell- to -cell propagation of thermal runaway, off-gassing and flaming beha viour of li-ion modules. Complete residential ES S typically do not exceed the module scale and m ay include all ancillary equipment within a single housing. I n commercial and industrial applications, modules are typically installed within racking physically similar to data systems a nd telecommunications equipment. Bus bars co nnect modules, to meet volt age and current design objectives. B us bar voltage ranges fr om hundreds of volts DC to more than 1000 VDC. Complete racks ar e defined as “units” and may be unenclosed or h ave enclosures that aid in me eting thermal management, electrical and mech anica l safety, and visual appearance objectives. Units typically weigh from tens to thousands of kilograms. Rack arrange ments are shown in Figure 3 (b). An ESS installation can consist of a single unit , especially in r esidences, and may b e installed indoors or outdoors. Co mmerc ial and industrial i nstallations can scale from single to hundreds o f ESS units. I t is common for ESS units to be installed in an inte rmodal ship ping c ontainer which contains ancillary ESS equipment and a complete fire protection system (e.g., detection, alarm, and extinguishing systems), which enables manuf acturers/installers to perform some pre -assembly before shipping. Similarly, custom large weather p roof enclosures are built to house one more ESS units. Dedicated ESS buildings constructed on -site are similar in layout to intermodal containe rs. Figure 4 shows the basic layout of ESS equipment commonly found at installations. Figure 4. Examples of outdoor ESS installations. The layout of ESS unit s in installations is like the layout of data and telecommunications equipment racking. The installation volume may be orders of magnitude larger than outdoor ESS enclosures. Indoor ESS installations are not com mon in North America, as codes and standards curre ntly r equire indoor equipment configurations to be first tested to develo p fire and explosion safety performance data using UL 9540A before a n Authority Having Ju risdiction (AHJ) approves Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 142 a specific installation (NFPA 855, 2019; Internationa l Code Council, 2018; Underwrite rs Laboratorie s, Inc., 2019). 1.2 Phenomenological description of li -ion cell thermal runaway and cell materials The basic design of a li -i on cell is uniqu e in t erms of fire and e xplosion hazards in that a single product can possess all of the elements necessary to ini tiate and sustain the process of combustion: fuel, heat, and some of the oxygen needed for partial oxidation of combusti ble cell materials . Thermal runaway occurs when a li-ion cell incre ases it s own temperature through self-heating, uncontrollably, until the rate of heat gen era tion exceeds the ra te o f heat dissipation. Th e thermal runaway proce ss often initiates in response to an abuse mechanism that destabilizes electrical and chemical potential energy stored by the cell. Abuse mechanisms can be thermal (e.g., external heat exposure), me chanical (e.g., crushing, penetration), electric al ( e.g., external short circuit, overcharge, over dischar ge), or inte rnal short c ircuits ( e.g., fo rmation of int erna l growths that compromise the separator, manufacturing defects, and contamination). Electrica l potential energy is released as part of thermal runaway when a los s of separator integrity results in an internal short circuit and Joule heating. The contribution of electrical potential e nergy to thermal runawa y severity and heat release rate has been documented by Golubkov, et al., (2015), Said, et al. (2018), and others. Chemical potential ener gy is released as part of thermal runaway wh en cell materials are sufficiently heated to initiate exothermic decomposit ion reactions, which lead to further heating and additional exothermic decomposition reactions. Feng, et al. (2018) identified six re actions that occur during a thermal runawa y pro cess. All but the separator melting are exother mic r eactions. Although some reac tions occur in parallel, the following order is given: 1. Decomposition of the solid-electrolyte inte rphase, which can happen as low as 57 °C. 2. Melting of the separator. 3. Further decomposition and regenera tion of the sol id-electrolyte interphase. 4. Cathode decomposition. 5. Electrolyte dec ompositio n. 6. Decomposition of the graphite/carbon anode with elec trolyte. In terms of mate rials, li -ion cells used in commercial and residential e nergy storage applications are effectively the same as those used consumer products (Warner, 201 5). Comm on cathode materials are: Lithium Cobalt Oxide (LCO), Lithium I ron Phosphate (L FP), Lithium Manganese Oxide (LMO), Lithi um Nickel Cobalt Aluminum Oxide (NCA), Lithium Nickel Manganese Cobalt Oxide (NMC), and Lithium Titanate (LTO). Anodes typi cally consist of graphitic carbons, although newer anode ma terials now include li thium alloyed metals. (Me konnen et al., 2016; Nitta et al., 2015; Doughty et al., 2012; Warner, 2015) . Within the first f ew charge cycles, a coating referre d to as the solid -electrolyte interphase forms on the anode surface by reactions with th e electrolyte. S eparator materials are made fro m polyethylene, polypropylene and ceramic- Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 143 polyolefin composite membr anes (Orendorff, 2012; Baldwin, 2009; Ya ng et al., 2012; Zhang et al., 2016). The thickne ss is t ypically on the order of 25 μ m but may be less. Electrolytes can be aqueous, non-aqueous, ionic li quids (salts with low boiling points), gel and solid polymers, or hybrid elec trolytes. Most electrolytes used in commercial cells are non -aqueous and consist of a solution of dis sociated lithium sa lt (LiPF 6 ) in hydroc arbon solvents (D oughty et al., 2012; Mekonnen et al., 2016). The hydrocarbon solvents are typically mixtures of ethylene carbona te, dimethyl carb onate, propylen e c arbonate, diethyl carbonate, an d ethyl methyl carbonate (Li, et al., 2016; Aurbach, et al., 2004). Table 1, summarize d from Somandepalli et al., shows an example of the material composition of an LCO cell by mass ra tio (Somandepa lli et al., 2014). Table 1. Materials used in a lithium cobalt oxide pouch cell. Chemical Comp ound Molecular Formula ∆ H c (kJ · g -1 ) Mass % Purpose Lithium Cobalt Oxide LiCoO 2 Inorganic 42.5 Cathode Lithium-interca lated Graphite LiC 6 Inorganic 34.9 Anode Diethyl Carbon ate C 5 H 10 O 3 21.15 7.2 Electrolyte Nylon 6 (C 6 H 11 NO) n 28.74 2.2 Case material Ethylene Carb onate C 3 H 4 O 3 12.3 1.1 Electrolyte Propylene Carbon ate C 4 H 6 O 3 15.53 0.9 Electrolyte Polyprop ylene (C 3 H 6 ) n 42.66 9.0 Separator Dimethyl Carbonat e C 3 H 6 O 3 13.4 0.3 Electrolyte Polyester Tereph thalate (C 10 H 8 O 4 ) n 22 2.0 Case material Table 1 demonstrates that material composition deter mines cell stability and the severity of thermal runaway. For example, the temperature of int ernal short circuit c an be increased selecting a separa tor material with a higher transition temperature, which can prevent breaching and distortion. A polyethylene separa tor m aterial h as a tra nsition temperature of 135 ° C , whereas composite separators ma y have transition tempe ratures in ex cess of 220 ° C (Ore ndorff, 2012) . Cathode selection can affect th ermal run awa y s everity. Less heat is released du ring thermal runaway with an LFP cathode than with an LCO cathode (Feng, et al., 2018). Material composition a lso de termines the off -gassing constituents o nce the rmal runaway o ccurs . The thermal energy generation of thermal runaway results in dissociation and vaporization of nearly all cell materials, typically leaving only r esidual soot , metals used for the cell case, and anode and ca thode substr ates. The g aseous/vapor products of thermal runaway are oft en r elea sed through the safety v ent. The safety vent may or may not pr event catastrophic rupture of the cell case during thermal runa way. With higher energy density cell chemistries , cell material surface temperatures achieved d uring thermal runa way can ignite the mixture, bu t off-gassing can occur Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 144 entirely without flaming, especially when thermal runaway occurs in confined volume such as in a module, unit or in a battery room/outdoor container. 1.3 Thermal runaway vent gas composition and release volume UL 9540A has standardized a methodology for deter mining the thermal runaway gas composition by applying an abuse me chanism to a cell contained in an 82 L nitrogen-inerted pressure vessel, shown in Figure 5. This approac h was developed from that used by Somandepalli et al. (2014) . Figure 5. 82-Liter gas composition test apparatus at UL. Application of thermal energy to the external cell surface by flexible film heater at 4 °C/min to 7 °C/min is one of the abuse mecha nisms used, as the results of the thermal abuse provide data which quantifies the surface t emperature at which the cell safety vent operates, and the surface temperature of the onse t of thermal runaway. The inert environment inside the test appara tus prevents combustion durin g thermal runawa y off- gassing, and facilitates collection of unreacted thermal runaway g ases 1 . The accumulated gas es are identified and quantified by gas chromatogra phy. Table 2 provides a typical makeup of thermal runaway off-gas compositi on for common li -ion c hemistries. The gas composition results is used to de velop the synthesis of thermal runaway gas for follow -on explosibility testing, described in section 2. The volume of gas released by a cell thermal runaway in the test apparatus , assumed to follow the ideal-ga s law, is given by: (1) (2) 1 exclud ing reaction of th e rmal runaway gas components with an y cathode -bound oxygen. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 145 Where : V is gas volum e in m 3 , n BVG is the number of moles of battery vent gas, R is the universal gas constant in J/mol· K , P is pressure in P a, T is temperature in °K, ⱷ O2 is the oxygen volume fraction in the test chamber prior to testing, and the subscripts “i”, “f”, and “∞” denote test apparatus initi al, final and ambient conditions at Normal Temperature and Pressure (NTP), respectively. UL 9540A standard requires testing of energy storage modules if a cell demonstrates thermal runaway and releases flamm able gas. P ropagation of thermal runaway in a m odule testing may result in a larger release of thermal runaway ve nt gas. Testing of energy storage units is necessary if modu le level testing demonstrates that module to module thermal ru nawa y propagation is possible. Quantifica tion of thermal runaway g as volume release at both module and unit level testing is achieved by col lecting the offgas and tot alizing the product of the exhaust flow rate and concentration measure ments of CO, CO 2 , tot al hy drocarbons, and hydrogen. 1.4 Propagation of thermal runaway within ESS The thermal exposure imposed on a djacent cells by cells in ther mal runaway is often enough to start the decomposition reactions described by Feng et . al ., cause an I SC and induce a cascading event. Figure 6 shows the time evolution for propagation of thermal runaway in a non-electrically interconnec ted array o f twenty-five 18650 cylindrical li-ion cells. One cell i n the bottom layer w as heated with a flexible film heater until thermal runaway. Within 1.5 s, thermal runaway occurred in adjac ent cells and the velocity of gas release was fast e nough to ca use jetting and fla me lift-off. After 8.5 s thermal runaway propagated to a ll cells in the array. From 11.5 s until after 22 seconds, the surface temperatures of the cell array a re at a high enough temperature t o emit visible light. Within 65 s the arra y was exhausted of thermal r unaway gases and residual combustible material. The gas composition obtained by testing one o f the c ells from Figure 6 with the 82-L gas composition test apparatus using the UL 9540A cell level test methodology is shown in Table 2. Values h ave b een rounded to the nearest int eger and correspond well wi th values tabulat ed by Baird et al. (2019). Table 2.: Repre sentative gas composition f rom ce ll level thermal runaway tests Gas Percent Carbon Dioxide 36 Carbon Monoxide 22 Hydro carbons 10 Hydro gen 32 Others < 1% Similar thermal runaway propagation spe ed and severity results for 18650 cells in air have been measured by (Said et al., 2019). Said e t. al. showed that the int ensity of cell to c ell he at transfer in arra ys can support thermal runaway propagation even without flaming in a nitrogen environment. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 146 Figure 6. Propagation of thermal runaway from one to 24 adjacent 18650 cells. The thermal energy gen era ted by cell to cell thermal runa way prop agation, as well as high momentum flaming combustion of thermal runawa y can enable module to module thermal runaway propagation. Figure 7 shows a UL 9540 A module level test demonstration conducted at UL based on the module design shown in Figure 3 a. The module contain ed 27.2 kg of cells, with a capacity of approximately 3.2 kWh. The pressure generated within the module enclosure was high enough to vent flaming gases more than 1 m, as shown in Figure 7c. (a) (b) (c) Figure 7. Demonstration of module (a) at t = 0 (b) during a 20-min period of off-gassing without flaming (c) immediate ly after igniti on of thermal runaway gas. Figure 3b shows that typical rack configurations tend to have minimal air gaps between modules, which may not provide insulation enough to inhibit heat transfer and pre vent propagation between adjace nt modules in a unit. Propagation of thermal runaway is also possible between nearby units if spacing is not ade quate to li mit heat transfer from unit to unit , particularly considering the potential for jet flaming fr om modul es in ther mal runaway. Without sprinkler prote ction, FM Global recommends spa cing unit s a minimum of 1.8 m – 2.7 m apart b ased on full-scale tests of LFP and NMC ESS units, respec tively (Ditch, 2019) . 1.5 Deve lopment of an explosi on hazard from accumulated battery gas Explosion hazards can de velop where unbu rned thermal runaway vent gas a ccumulate in a confined volume. Figure 3 and Figure 4 show confined volumes defined by the module, unit and 0 s < 0.5 s 1.5 s 5 s 8.5 s 15 s 22 s 65 s Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 147 battery room enclosures. Therefore, deflagration hazards can occ ur at sev era l different produc t levels. Figure 8 shows the volume flow rates me asured during the UL 9540A modul e level test demonstration. Volume flow rates were determin ed by measuring gas con centra tions within the ductwork of an oxygen consumpti on calorimeter t hat collect ed the gases released by the module. CO and CO 2 were measu red by nondispe rsive infr are d gas analysis and total hydrocarbons were measured using flame ion ization detection. The mo dule relea sed a total of 1 m 3 of thermal runaway gas before flaming. The demonstra tion also showed that the module enclosure forced under -ventilated conditions in which the gases were rel ease d without flaming. A re lease volume of 1 m 3 volume is a dequate to crea te a mixture a bove the lo wer flammability limit (LFL) in volumes up to approxima tely 16 m 3 , based upon a range of LFLs fr om 6.1 v% to 11.8 v% calculated for thermal runa way gas by B aird et al. (2019). The severity of the deflagration hazards are g reater for smaller volumes, as the fuel concentration can reach a worse-c ase scenario n ear stoichiometric fuel -air conditions. The gas composition data of the cell are used to dev elop the deflagration parameters lower fl ammability limit (ASTM E-918), laminar burning velocity (ASHRAE 34), and P max (EN 15967) Figure 8. Volume flow rates of thermal runaway off gas. 2. Explosibility parameter values for thermal runaway gases The gas explosibility parameters needed for e xplosion protection design depend on the protection method and associated d esign standard. The pri mary parameters for N FPA 68 (2018) e xplosion vent sizing are the gas mixture closed vessel deflagration pressure, P max , the gas mixture laminar burning ve locity, S u , a nd the ratio of the gas mixture volume, V mix , to the enclosure volume. The European normative standard for gas explosion ve nting, EN 14494 (2007), uses these parameters plus the gas mixture normalized rate-of-pressure rise, K G . Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 148 In applying these explosion venting standards, th e protection design is usually based on a worse - case, near- stoichiometric, gas mixture. Therefore, determina tions of P max and S u require testing or calculations over a range of gas conce ntrations. UL 9540A testing requires determination of P max and S u over a range of fuel concentrations for the thermal runaway vent ga s mixture measured at the cell level in the ine rte d environment. An exa mple of P max test data obtaine d over a range of fu el c oncentrations is shown in F igure 9. Thermal runaway vent gases c orrespond to the mixture documented in Ta ble 2. P m ax data for the thermal runaway gas es was obtained in a spherical 5 L test vessel using a 10 -15 joul e squib igni ter. A P max of 6.4 bar-g was measured at a fuel concentration of 2 9 vol%. Data from Somandepalli et al. (2014) obtained for thermal runaway vent gases from 7.7 Wh ce lls at 100% SOC are also shown. The Somandepalli et al. tests were conducted in a 20 L test vessel using a 250 J chemic al igniter. The Somandepa lli et al. data show a P max of 7.0 bar-g at a fuel concentration of about 15 vol%. The slightly higher P m ax for the smaller cell thermal runaway gases may be due to larger test vessel and more energe tic igniter since the gas compositions were similar. Figure 9. Closed vessel pressure versus concentration for two different size c ells. Gas mixture bu rning velociti es for the UL 9540A thermal runaway ga ses were obtained by measuring flame propag ation speeds and flame f ront surface areas in a 4-cm diameter, 150 cm long glass tube and a djusting the flame speed by the ratio of the fla me base cross-sec tional area to the flame surface area per Jabbour (2004 ). Data obtained for the same thermal runawa y gases as used to obtain the Figure 9 data show a maximum burning ve locity of roughly 35 cm/s at a fue l concentration of 26 v%, with burning velocities decreasing to about 20 cm/s at a concentration of 17 v% on the le an side and about 38 v% on the fuel rich side of the worst- case conc entra tion. This data is not necessarily representative of results for other UL 9540A runa way ga s samples, some of which have produce d significantly higher burning velocities. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 149 Johnsplass e t a l. (2017) have c alculated burning velocities using measured thermal runa w ay gas compositions and widely used open-source softw are for chemical reactions and burning velocities. Their results for diffe rent lithium ion electrolytes show peak values of S u in t he range 35-50 cm/s at equivalence ratios of 1.1 to 1.2. The lower end of this range is about 10% higher than the peak S u for the gas fr om Table 2, while the uppe r end of that ra nge is roughly comparable to the highe r values of S u measure d from gases produced in U L 9540A tests. Since laminar burning velocities increase rapidly with increasing gas mi xture temperature, the room-tempera ture measured v alues of S u need to be adjusted upward for ES S explosion venting applications in which the thermal runaway produ ces elevated gas temperat ures. Fernandes et al. (2018) have calculated values of S u for gas compositions formed from vented Li -ion cell overcharge tests. They show the value of S u for s toichiometric mixtures increasing from 30 cm/s to almost 110 cm/s as the gas mixture initial tempera ture increases from 300 K to 600 K. Normalized rate- of - pressure rise, K G , data are als o obtained for the UL 9540A thermal runaway gases using the same test method as for P max values. Data for the T able 2 ga s mixtures resulted in a K G of 84 bar-m/s at a fuel c oncentra tion of 2 9 v%. This is 29% gre ater than the K G value measured by Somandepalli et al. of 65 bar-m/s for their tests with 100% SOC thermal run awa y gases. Ho wever, the use of the 5 -L and 20- L test vessels for this data may limit its applicability because of the well-kno wn test -vessel large-volume flame acceleration effects mentioned by Somandepalli et al. The gas mixture volume to be use d for explosi on v enting applications can be either the entire ES S enclosure volume for wo rse-case scenarios, or a smaller volume for performance-b ased scenarios involving thermal runaway conditions represented by UL 9540A testing. In the latter case, calculations can be based on the volumes o f the thermal runa way gases measured in module o r larger scale UL 9540A t esting. Usually, thi s wou ld entail assuming the measured gas volumes from worse-case near-stoichiometric mixtures with enclosure air. If N FPA 69 (2019) or another explosion pre venti on standard is to be used, the applic able explosibility parameters are either the gas mixture LFL to a chieve prevention by combustible concentration reduction or the Limiting Oxygen Con centra tion (LOC) to achieve prevention by inerting. LFL data are obt ained with gas samples from UL 9540A thermal runawa y tests. An LFL value of 8.9 v% was obtained for the ga s shown in Table 2, as determined us ing ASTM E918 with an electric arc ignition. Baird et al. (2020) report LFL values in th e range 6.1% to 11.8% for g as compositions obtained from diffe rent cell chemistries and SOCs. Thermal run away g as L OC data are not routinely measured because they depend on the inert gas used and, so f ar, the re have been f ew, if any, ESS inerting applications. C alculation methods to estimate fuel gas mixture LOC values from the gas composition and LOC va lues of the constituent gases a re described by Britton et al. (2016) and summarized in Ann ex B of NFPA 69 -2019. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 150 13th International Symposium on Hazards, Prevention, and Mitigation of Industrial Explosions Braunschweig, GERMANY – July 27-31, 2020 A Eulerian m odel for dust defla grations, i ncluding inne r particle effec ts Christoph Spijker a , & Harald Raupe nstrauch a a Ch air of Ther mal Pr oces si ng Tec hno logy ( Mon tanun iv ersi ta e t Le obe n, Leobe n, A ust ria ) E-mail: [email protected] Abstract Previous studies (Spijker 2017) h ave show n that t he oil evaporation rate of Lycopodium particles is dependent on the resistance for diffusive and conv ective transport through the pores. Based on these studies a simplified inner particle transport model w as created. The model assumes that the evapora tion of the oil oc curs in a layer. The species and energy transport between this layer and the particle surface are mod elled by algebraic e quations assuming a momentary ste ady state condition. The evapo ration rates calculated by this approach achieve similar results to the 3D fully resolve d particles. In the dust d eflagration of Lycopodium , pyrolysis of the li gnin structure is of se condary importance, therefore a simplified one e quation pyrolysis mechanism wa s developed. The results are compared with the re sults from the model e xcluding inner tra nsport eff ects, the Eule rian flame speed model developed by Spijker ( 2014). Fu r th er more the influenc e of he at transfer by radiation was considered. Keywords: inner particle effects, dust deflagration, simulation, Eulerian approach, Lycopodium 1. Introduction Usually, the transport e ffects in the particles are neglected du e to the high demand on computing resources. To model particle inner effects with discretized particles would i ntroduce new differential equations to be solved fo r each particle. A radial 1- D model from Spij ker (2 015) wi th 25 layers n eeds 680 seconds to model 0. 25 seconds real time on on e co re of an Intel i7 -970. Using a Eul erian approac h su ch a particle must be solved in ev ery cell and additional transport equations have to be introduce d on the fluid grid, to account the movement of the particles. When a L agrangian a pproach is used, the differentia l equations must be solved for e very parc el. The grid used to m odel the duc t experime nt from Kern (2013) ha s 6 ,3 Mill ion cells. Pollhammer (2015 ) used additionally 4 mi llion pa rcels, repre senting approx. 16 mil lion particles for his Lagrangian approa ch . As shown in Fig. 1 the dominate process on the thermal conversion of Lycopodium in a dust deflagration is the evaporation of the contained oil. The 0 D model, negl ecting inner particle transpo rt effects, sho ws higher evaporation rates th an the fully resolved 3D p article model. This will result in a faster flame propagation. Wi th an analytical model describing the transport resistance of the oil vapour in th e pores of the particle, thi s e ffect can be modelled with low additional computing resources. The pyrolysis of the lignin st ructure is of secondary relevance and sho ws no diffe rence betw een the 0 D an d 3D model. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 253 Fig. 1. Change of mass for a Lycopodium particle in a dust deflagration (Spijker 2015) 2. Model description 2.1 Global Eulerian Mod el The global Eulerian model uses an Eulerian approach to describe multi component reacting mixtures in the gas phase, includin g turbulence and radiation. An additional Eulerian approach is used for the particle phase, with additional transport equations fo r the composition of the particles. 2.1.1 Momentum and continuity equations Equation 1 represents the momentum e quation for the gas phase. Here is 𝑣 → the ve locity vector, 𝜀 the gas phase volume f rac tion calculated from the p article concertation 𝐷 and de nsity of the m ixture 𝜌 , calculated using the ideal gas equation and the molar mass of the mixture. Turbulenc e is consider ed by the effective viscosity 𝜇 𝐸𝑓 𝑓 which is the sum of the molecular viscos ity , given by the Sutherland approac h (Sutherland 1893) and the turbulent vi scosity from the turbulence model. To consider gravitational effec ts, the volume force s caused by the gravitational acceleration vector 𝑔 → are included in the equation. The int erac tion forces betw een particle and gas phase i s considered by Wen -Yu (1966 ) copping coefficient 𝐾 (equation 2) a nd the relative velocity between the phases 𝑢 → − 𝑣 → . The continu ity equation for t he gas phas e (equation 3) considers the gas phase volume fr action and the mass transfer source term 𝑄 𝐷 , calcu lated using the inner particle model. ∂ ( 𝜀𝜌 𝑣 → ) ∂𝑡 + 𝑑𝑖𝑣 ( 𝜀𝜌 𝑣 → 𝑣 → ) − 𝑑𝑖𝑣 [ 𝜀 𝜇 𝐸𝑓𝑓 𝑔𝑟𝑎𝑑 ( 𝑣  ) ] − 𝜀𝜌 𝑔 → − 𝐾 ( 𝑢 → − 𝑣 → ) = −𝑔𝑟𝑎𝑑 ( 𝑝 ) 𝜀 (1) 𝐾 = 3 4 𝐶 𝑤 𝜀 ( 1−𝜀 ) 𝜌 𝑑 𝑝 𝜀 2, 65 | 𝑣  − 𝑢 󰇍  | (2 ) ∂ ( 𝜀𝜌 ) ∂𝑡 + 𝑑𝑖𝑣 ( 𝜀𝜌 𝑣 → ) = 𝑆 𝐷 (3) 0% 10% 20% 30% 40% 50% 60% 0 0,05 0,1 0,15 0,2 0,25 mass/initi al mass time [s] Lignin 3D-Model Oil 3D-Modell Oil 0D-Model Lignin 0D-Model Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 254 The particle phase conservation equation ( equation 4) describes the momentum of the dust, wh ere 𝐷 stands for dust mass concentration and 𝑢 → is velocity vector of the particle phase. Diff usive effec ts on the particle phase are considered by the kinetic particle viscosity 𝜇 𝑘𝑖𝑛 and the particle collision viscosity 𝜇 𝑐𝑜𝑙𝑙 . Granular pre ssure 𝑝𝑠 as well as 𝜇 𝑘𝑖𝑛 and 𝜇 𝑐𝑜𝑙𝑙 are calculated by an a nalytical approac h (Syamlal 1993). The continuit y equation (equation 5) is based on the mass concentra tion 𝐷 and takes the mass transfer source term 𝑄 𝐷 into account. ∂(𝐷 𝑢 → ) ∂𝑡 + 𝑑𝑖𝑣 (𝐷 𝑢 → 𝑢 → ) − 𝑑𝑖𝑣 [ 𝜀 ( 𝜇 𝑘𝑖𝑛 + 𝜇 𝑐𝑜𝑙𝑙 ) 𝑔𝑟𝑎𝑑 ( 𝑢 󰇍  )] − 𝐷 𝑔 → − 𝐾 ( 𝑣 → − 𝑢 → ) = −𝑔𝑟𝑎𝑑 ( 𝑝 )( 1 − 𝜀 ) − 𝑔𝑟𝑎𝑑 ( 𝑝𝑠 ) (4) ∂ ( 𝐷 ) ∂𝑡 + 𝑑𝑖𝑣 (𝐷 𝑢 → ) = −𝑆 𝐷 (5) 2.1.2 Energy equations An e nergy equation is solved for each phase. The gas phase ene rgy equation (equation 6) is based on the e nthalpy ℎ , using in O penF OAM 2.4 implemented rea ction a nd ther mophysical models. The heat conduction is calculated using effective heat conductivity 𝜆 𝐸𝑓𝑓 which is the sum of the molecular heat conductivity and the turbulent heat conductivity. The turbulent heat conductivity is calcula te d from the turbulent viscosity and turbulent Prandtl number. This equation c onsiders the h eat source from homogenous reactions 𝑄 𝑅 ,𝑔 , the heat transfer with the partic le phas e and the heat source term for radiation 𝑄 𝑅𝑎𝑑 . This source term considers only gas absorption and wall interac tions. F or the he at transfer coefficient to the partic le phase 𝛼 𝑔 ,𝑝 th e Nusselt corre lation by Ranz-Marshall (Ansys 2009) is used. 𝐴 𝑉 describes the volume specific surface a rea of the particles, based on the dust concentration 𝐷 . 𝑇 𝑝 is the particle temperature and 𝑇 𝑔 the gas tempe rature from the thermophysical model used in OpenFOAM. ∂ ( 𝜀 ℎ ) ∂𝑡 + 𝑑𝑖 𝑣 (𝜀 𝑢 → ℎ ) − 𝑑𝑖𝑣 [ 𝜀 𝜆 𝐸𝑓𝑓 𝑔𝑟𝑎𝑑 (𝑇 𝑔 )] = 𝑄 𝑅 ,𝑔 + 𝛼 𝑔, 𝑝 𝐴 𝑉 (𝑇 𝑝 − 𝑇 𝑔 ) + 𝑄 𝑅𝑎𝑑 (6) In the energy equation for the pa rticle ph ase (equ ation 7) enthalpy is formulated using dus t concentration 𝐷 , temperature of the particle phase 𝑇 𝑝 and heat capac ity of the particles 𝑐 𝑝 𝑝 . Heat capacity is calculated con sidering particle composition and temperature, bas ed on the work of Spijker (2015). Additionally to transient and source te rms the convective heat tran sport is considered. Heat source term for the particle reactions 𝑄 𝑅,𝑝 is mo delle d with inner particle model. Absorption an d emission from the pa rticles are tak en in account in the radiation model with a separate source term 𝑄 𝑅𝑎𝑑 ,𝑝 for the particle phase. ∂ ( 𝐷𝑇 𝑝 𝑐 𝑝 𝑝 ) ∂𝑡 + 𝑑𝑖𝑣 (𝐷 𝑣 → 𝑇 𝑝 𝑐 𝑝 𝑝 ) = 𝑄 𝑅, 𝑝 + 𝛼 𝑔 ,𝑝 𝐴 𝑉 (𝑇 𝑔 − 𝑇 𝑝 ) + 𝑄 𝑅𝑎𝑑, 𝑝 (7) Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 255 2.1.3 Species equations and reactions Species transport in the gas phase is given by equation 8. This equa tion is based on the gas phase mass fraction 𝑌 𝑖 . Diffusive transport is modelled by an effec tive diffusivity 𝑑 𝑒𝑓𝑓 which is the sum of the molecular diffusivity ca lculated from the viscosity , constant Schmidt number and th e turbulen t diffusivity from the turbulence model. The source term 𝑆 𝐷 ,𝑖 represents the heterogenous reaction rate fo r the speci es i in the in ner particle model. 𝑆 𝑖 is the homogenous reaction r ate c alcula ted by using the Partially Steered Reactor (PaSR) model implemented in OpenFOAM 2.4. Fo r the homogenous reac tions, two combined reaction mechanisms were used. For the combustion of the pyr olysis gases the Jones-Linstedt (1988) was used, whereas for the combustion of the oil vapor in the gas phase the mechanism by Pollhammer (2015) was implemented. 𝜕 ( 𝜀𝜌 𝑌 𝑖 ) 𝜕𝑡 + 𝑑𝑖𝑣 ( 𝜀𝜌 𝑣 → 𝑌 𝑖 ) − 𝑑𝑖𝑣 [ 𝑑 𝑒𝑓𝑓 𝑔𝑟𝑎 𝑑 ( 𝑌 𝑖 ) ] = 𝑆 𝐷 ,𝑖 + 𝑆 𝑖 (8) In the particle phase two transport equations are sol ved, one for the m ass frac tion of the liquid oil 𝑋 𝑜𝑖𝑙 and one fo r th e li gnin mass fra ction 𝑋 𝑙𝑖𝑔𝑛𝑖𝑛 . Only convective transport a nd the py rolysis re action rate fro m the inner particle model 𝑆 𝐷 ,𝑗 were considered. 𝜕 ( 𝐷𝑋 𝑗 ) 𝜕𝑡 + 𝑑𝑖𝑣 (𝐷 𝑢 → 𝑋 𝑗 ) = −𝑆 𝐷,𝑗 (9) 2.1.4 Turbulence modelling For turbulence modelling a modi fied standard k- ε model was used. Bas ed on the studies of turbulence behaviour in dust deflagrations by Spijker (2014), an additional sour ce term, based on the consistent approac h of Crowe (Crowe, 2000) was implemented in the equation for the turbulent kinetic energy 𝑘 (equation 10 ). This term models the g eneration of turbulent ki netic en ergy due to the relative velocity of the phases. ∂ ( 𝑘𝜌 ) ∂𝑡 + 𝑣  𝑖 ∂ ( 𝑘𝜌 ) ∂𝑥 𝑖 = ∂ ∂𝑥 𝑖 ( 𝜇 𝑡 𝑃𝑟 𝑘 ∂𝑘 ∂𝑥 𝑖 ) + 𝜇 𝑡 ( ∂𝑣 𝑖 ∂𝑥 𝑗 + ∂𝑣 𝑗 ∂𝑥 𝑖 ) ∂𝑣 𝑖 ∂𝑥 𝑗 − 𝜀𝑝𝜌 + 𝜌 𝐷 𝜏 𝑝 | u 󰇍  − 𝑣  | 2 (10) 2.1.5 Radiation The ra diation model use d is a P1-model, consid ering absorption and emissi on of gas phase a nd absorption, emission an d uniform scattering o f the particle pha se. Radi ation int ensity 𝐺 is using equation 11. Here is 𝑎 the absorption coefficient of the gas phase. This coefficient will b e modelled by a Weighted Sum of Gray Gases Model, based on the TNF6 Proceedings (Grosshandler 1993) . This model considers te mperature de pended absorption of CO 2 , CO and H 2 O. The absorption coeff icient of the particles 𝑎 𝑝 is calculated by formula 12, where 𝜀 𝑝 is a constant emission coefficient , 𝐷 the mass concertation of dust, 𝜌 𝑝 the density of the particle and 𝑑 𝑝 the particle di am eter. I n sim ilar manner the scattering co eff icient 𝜓 𝑝 for uniform scatt ering modelled (formula 13). The source t er m for the energy equation o f each ph ase ( formula 14) is calculated by the diffe renc e between absorption by the phase 𝐺𝑎 and emission by the phase 4𝑎𝜎 𝑇 𝑔 4 . 𝑑 𝑖𝑣 [ 1 3 ( 𝑎+𝑎 𝑝 +𝜓 𝑝 ) 𝑔𝑟𝑎𝑑 ( 𝐺 ) ] + 4𝑎𝜎 𝑇 𝑔 4 + 4𝑎 𝑝 𝜎 𝑇 𝑝 4 = ( 𝑎 + 𝑎 𝑝 )𝐺 (11) 𝑎 𝑝 = 𝜀 𝑝 3 𝐷 𝜌 𝑝 𝑑 𝑝 (12) 𝜓 𝑝 = (1 − 𝜀 𝑝 ) 3 𝐷 𝜌 𝑝 𝑑 𝑝 (1 3) 𝑄 𝑅𝑎𝑑 = 𝐺𝑎 − 4𝑎𝜎 𝑇 𝑔 4 an d 𝑄 𝑅𝑎𝑑,𝑝 = 𝐺 𝑎 𝑝 − 4𝑎 𝑝 𝜎 𝑇 𝑔𝑝 4 (1 4) Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 256 2.2 Inner particle Mode ls 2.2.2 Oil evaporation The numeric al studies by Spijker (2017) on different for ms of resolved Lycopodium particles have sh own that the maximum occurring thickness of the oil e vaporation region is less than 3,4 µ m . So, the analytical model for the oil evaporation uses a c ore approach shown in figure 2. This approa ch is based on t wo radii, 𝑟 0 is the constant particle radius equal to 𝑑 𝑝 2 , whereas the second 𝑟 𝑣 is the radius of the o il saturated particle, ca lculated as a func tion of the oil mass fr action 𝑋 𝑜𝑖𝑙 . To a ccount the d ens core, an a veraged core radius 𝑟 𝑐 is introduc ed. The calculation of the radius 𝑟 𝑣 is shown in e quation 15, where 𝑋 𝑜𝑖 𝑙, 𝑠𝑡 𝑎𝑟𝑡 is the oil mass fraction before evaporation. Fig. 2. Concept of the core model for the oil ev aporation 𝑟 𝑣 = [ 𝑋 𝑜𝑖𝑙 𝑋 𝑜𝑖𝑙,𝑠𝑡𝑎𝑟𝑡 (𝑟 0 3 − 𝑟 𝑐 3 ) + 𝑟 𝑐 3 ] 1 3 (1 5) The main transport m echanism for the oil vapor from the evaporating layer to the particle surface is the convection through the po re structure. To de scribe the pr essure gr adient, the Carman Kozeny (1956) equation for a sp herica l coordinate system is us ed (equation 16). Here is 𝑢 𝑟 radial velocity, 𝜇 𝑜𝑖𝑙 viscosit y of the oil vapor a t evapora ting temperature, 𝜀 𝑒𝑓𝑓 e ffective pore volum e fraction, 𝑑 𝑝𝑜𝑟𝑒 pore diameter and 𝐹 stands for an empirical factor for the form of the pore structure. 𝜕𝑝 𝜕𝑟 = −𝜇 𝑜𝑖𝑙 𝑢 𝑟 ( 1−𝜀 𝑒𝑓𝑓 ) 2 𝐹 𝜀 𝑒𝑓𝑓 𝑑 𝑝𝑜𝑟𝑒 2 (1 6) To obtain the radial vel ocity continuity equation for a spheri cal coordinate system is de fined b y equation 17 and used as the source term for the g lobal species equations. Here is 𝜌 𝑜𝑖𝑙 the density o f the oil vapor, 𝑝 ∗ the vapor pressure of the oil ca lculated by the Antoine equation a nd 𝑝 ∞ the pre ssure of the computing cell. 𝑆 𝐷 ,𝑜𝑖𝑙 = 𝐷 𝜌 𝑜𝑖 𝑙 𝜋𝐹 𝜀 𝑒𝑓𝑓 𝑑 ℎ 𝑝𝑜𝑟 𝑒 2 4𝜇 𝑜𝑖𝑙 ( 1−𝜀 𝑒𝑓𝑓 ) 2 𝑝 ∗ −𝑝 ∞ 1 𝑟 𝑣 − 1 𝑟 0 (1 7) This equation fails nume rically when 𝑟 𝑣 ≈ 𝑟 0 . For that reason, if 𝑟 𝑣 > 0.9 𝑟 0 simple enthalpy differe nce eva poration model is used. To avoid unphysical conditions, the s ource them 𝑆 𝐷 ,𝑜𝑖𝑙 is set to zero i f 𝑝 ∗ ≤ 𝑝 ∞ or 𝑋 𝑜𝑖𝑙 𝐷 < 𝑆 𝐷,𝑜 𝑖𝑙 ∆𝑡 2 . To eva luate the oil evaporation model, evaporation rates of the single particle models by Spijker (2015) were use d. The 0D model n eglects transport effec ts and as a consequence overestimates evapora tion rate of the oil. The 3D model represents a fully resolved pa rticle with a core structur e, diffusive and convective transport though the pores . The boundary c onditions are mappe d fr om the Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 257 two Lagrangian models (Particle 1, Particle 2), used for a dust deflagration in a vertical duct by Spijker (2015). As can been see n fr om figure 3 evaporation rates calculated using eva poration model 3D model are almos t ide ntical. The sli ght diffe rence is c aused by th e sim plified round core in the evapora tion model. Fi g. 3. Compression of oil evaporation model to the 0D and 3D results from Spijker (2015) 2.2.3 Pyrolysis of the Lignine The pyrolysis rates between the 0D, 1D and 3D studies of Spij ker (2015, 2017) show basically no differe nce. The implementation of 8 reaction pyrolysis model would re quire 7 additional variables in the particle phase of th e global model. To simpli fy the pyrolysis only one averaged reaction is implemented (equation 18 ). The frequency factor 𝐴 , the activation energy 𝐸𝑎 and the reaction order 𝑛 are a veraged from the result s of the studies by Sp ijker (2015, 2017). 𝑆 𝐷 ,𝐿𝑖𝑛𝑔𝑖𝑛𝑒 = D 𝐴 𝑋 𝐿𝑖𝑔𝑖𝑛𝑖𝑛 ,𝑠𝑡𝑎𝑟𝑡 𝑛 𝑒 − 𝐸𝑎 𝑅𝑇 𝑝 (𝑋 𝐿𝑖𝑔𝑖𝑛𝑖𝑛, 𝑚𝑎𝑥 − 𝑋 𝐿𝑖𝑔𝑖𝑛𝑖𝑛 ) 𝑛 (18) To provide different sou rce terms for the gas pha se, a spli tting gas factors for each spe cies 𝜙 𝑖 were introduced. Th ese provide an average composition of the pyrolysis gases, based on the studies by Spijker (2015, 2017). 𝑆 𝐷 ,𝑖 = 𝑆 𝐷 ,𝐿𝑖𝑛𝑔𝑖𝑛𝑒 𝜙 𝑖 (1 9) 0 1E-13 2E-13 3E-13 4E-13 5E-13 6E-13 0 0,05 0,1 0,15 0,2 evapora tion rate per particle [kg/s] time [s] Particle 1 3D (Spijker 2015) Particle 2 3D (Spijker 2015) Particle 1, 0D (Spijker 2015) Particle 2, 0D (Spijker 2015) Particle 1, evaporation model Particle 2, evaporation model Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 258 3. Model evaluation and results 3.1 Evaluation The model was evaluated by the flame propa gation from du ct experiment s by Kern (201 3) and th e flame temperatur e meas urements by Wieser (2015). For both experiments the sa me apparatus was used. A vertical duct wit h an inner diamet er of 140 mm and a length of 20 00 mm. At the top of the duct pa rticles we re dispersed by a scre w conveyer and ignited a t the bottom by a spark. T he fir st step in the modelling of dust deflagration in the duct is to a chieve the ini tial conditions befor e ignition. For that purpose, the dust dispersion was modelle d, until there was no significant change in the dust concertation and turbulence fi eld. Subsequently the mi xture w as igni ted by setting one ce ll to a constant gas phase temperature of 2000 K for 20 m s. The top flame posit ion is the highest point i n the duct wh ere the flame re aches a temperature ab ove 1000 K. In figure 4 the top flame position from the evaporation model is comp are d to the flame sp eed model by Spijker (201 4) and experiments by Kern (2013) w here the flame position is trac ked by video ana lysis. For the fir st simulations the same grid was used as by Spij ker (2014), called normal grid. This grid has a hexahedral cell length of 2.5 mm . Here is the flame v elocity using the evaporation model much slower than th e 0D model and adequate accordance to t he measurements. The fl ame speed m odel by Spijker (201 4) has the best agree ment with the experiment unti l 200 ms. Flame velocities of the evaporation model are in good accordance to the mea surements after 120 ms but overe stimated at the beginni ng. A grid study for this model shows grid independency for a hexahedral cell length of 1 mm (fine grid). The coarser grid overestimates the flame ve locity at the begi nning, caused by the l arger volume of the ignition cell and therefore higher ignition energy. Fig. 4. Comparison of the top flame position between different models and experimental results from Kern (2013) 0 0,25 0,5 0,75 1 1,25 1,5 0 40 80 120 160 200 240 top flame position [m] time after ignition [ms] Experiment (Kern 2013) 0D model Evaporation model, normal grid Evaporation model, fine gr id Flame speed model (Spijker 2012) 4 Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 259 The measurements by Wieser (2015) wh ere carried out by the an alysis of the transient h eat up of thermocouples. The measurements have a large spread, still all simulation results are within given range (T able 1). The temperatures of the 0D and evaporation model are close due to the li mitation of oxygen. The flame speed model has a different th ermophysical model, ba sed on p rogress v ariables and calculates slightly lower te mperatures. Table 1 : Comparison of the Flame temperatures Hight in t he duct Measured tempera te (Wieser 201 5) Temperat ure flame speed model (Spijker 2 014) Temperat ure 0D m odel Temperat ure evapora tion model 300 mm 1 186 K – 1642 K (3  ) 1205 K 1238 K 1232 K 600 mm 1222 K - 1299 K (3  ) 1219 K 1251 K 1246 K 900 mm 1 149 K – 1307 K (3  ) 1 153 K 1225 K 1213 K 3.2 Evaporation and flame propagation Till approx. 120 ms th e evaporation take s place in a thin laye r and the flame grow s in an elliptical shape, as shown in figure 5 at 100 ms. After 120 ms the flame gets influenced by the w all and sta ts to accelerate, where the top of the flame de forms and the thi ckness of the ev aporation layer in creases , as can be seen from figure 5 at 200 ms. Th is behaviour continues and the top of the fla me deforms further, the volume of t he evaporation zone increases, as well as the total evaporation rate in th e system. At 245 ms the flame reaches the high est point . Due to the rich mixture and the lack of oxygen the flame starts to collapse, as shown at 250 ms. 100 ms 200 ms 225 ms 245 ms 250 ms Fig. 5. Representation of the flame position and evaporation zone. The orange iso- su rface represents the flame position at the temperature of 1000 K, the green represents particles with 1 % of the initial o il content and the brown particles with 99 % of the initial o il content. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 260 4. Conclusions and outlook Imple mentation of the ev apora tion mode l using an analytical solution for the transport resistance in the particle gives a bette r prediction of the flame in a Lycopodium deflagration then a 0 D model , without increasing calculation time significant ly . In the curr ent form the model is applicable to large industrial geometries with reasonable calculation t ime and can expl ain the physics on a d etailed level. Th e approach in this form is only applicable for the evaporation of f luids in the po re structures o f small particles. The focus of further research is to d eve lop simplified models for other particles, using highly resolved single particle models. A model for coal dusts i s in progress. Reference s Ansys Inc. (2009). Ansys Fluent 12 Theory Guide Carman, P.C. (1956). Flow of gases through porous media, Butterworths Scientific Publi cations , London Crowe, C. T. (2 000). On models for turbulence modul ation in fuid -particle fows, International Journal of Multiphase Flow , 26: 719-727 Grosshandler, W. L., R ADCAL (1993). A Na rrow-Band Model for Ra diation Calculations in a Combustion Environment, NIST technical note 1402 Jones, W. P., Lindstedt, R. P . (1988). Global reaction schemes for hydroca rbon combustio n, Combustion and Flame 73: 222 – 233 Kern, H. (2013). Brennbare Staub/Luft Gemische - Untersuchungen zur Flammenfortpflanzung unter nichtatmosphärischen Bedingungen, PhD Thesis , Montanuniversit ae t Leoben, Leoben Austria Pollhammer, W. (2015). Numerische Untersuchung der Mindestzünde nergi e sowie der Flammenfortpf lanzung i n Staub/Luft – Gemischen mittels eines Euler – Lagrange – Modells in OpenFOAM, Master Thesis , Montanuniversit ae t Leoben, Leoben Austria Spijker, C. (2014). A CFD Approace for dust explosions based on OpenFOAM, X ISHPMIE Conference Procee dings Spijker, C. (2015 ). Numerische Untersuchung der partikelinneren Effekte bei Lykopodiumstaubdeflagrationen, PhD Thesis , Montanuniversit ae t Leoben, Leoben Austria Spijker, C. (2017). Numerical investigation on inner particle effects in Lycopodium/Air dust deflagrations, Journal of Loss Prevention in the Process Industries , 49: 870-879. Sutherland W. (1893). The viscosity of gases and molecular force , Philosophical Magazine , 5: 507 – 531 Wen, C.- Y. , Yu, Y. H. (1966). Mechanics of Fluidization, Chem. Eng. Prog. Symp . :100 – 111 Wieser, G . (2015). Einflü sse der T emperatur und der Turbulenz auf die Flammenausbreitung in Staub/Luft-Gemischen unter reduzierten Druckbedingungen, 4. Magdeburger Brand- und Explosionsschutztag , Magdeburg Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 261 CFD Simulation of an Unconfined V apor Cloud Explosion thr ough obstacles using OpenF oam R  C. Langrée a , b , E. Franquet c , J. Re veillon a , G. Lecocq b & F .X. Demoulin a a Comple x e de Recherche Interprofessionnel en Aérothermochimie (CORIA), Rouen, France b Institut National de l’En vironnement Industriel et des Risques (INERIS), V erneuil en halatte, France c Laboratoire de Thermique, Ener gétique et Procédés (LaTEP), P au, France E-mail: langr [email protected] Abstract In this work, numerical simulations are carried out to characterize the impact of obstacles on flame propagation v elocity to properly assess the consequences of Unconfined V apor Cloud Explo- sions (UVCE). W ithin this scope, an UVCE within a congested en vironment is simulated using RANS modeling of the turb ulence. Comb ustion phenomena are characterized through the progress v ariable transport equation. The simulation is done with the open source library OpenFoam R  . Numerical results are compared with data from the INERIS and GDF-SUEZ e xperimental campaign EXJET (hydrogen-air and methane-air e xplosions). K eywords: hazar ds, industrial explosions, CFD, pr emixed comb ustion 1 Intr oduction Unconfined V apor Cloud Explosion (UVCE) is one of the most dev astating industrial accidents possible (J. Daubech (2016)). The loss of containment of a flammable product (e.g. methane or hydrogen) can lead to the formation of a flammable cloud all across the plant. The resulting mixture can be lit by an y ener gy source within the cloud (e.g. sparkles, hot surfaces). The propagation of the flame front within the cloud can cause damaging pressure ef fects in close and long range. In order to assess these damages, CFD can be used, in parallel with widely used phenomenological tools. The open source code OpenF oam R  has already been used for such purpose (Bauwens et al. (2011), Ghasemi et al. (2014), Rao et al. (2018)). The global aim of this study is to improv e numerical methods for CFD simulation of industrial UVCE. Hazardous consequences of UVCE are mainly caused by the o v er -pressure generated by the propagation of the flame front. In deflagration cases, this ov er-pressure is directly link ed to the flame front propagation speed (J. Daubech (2016)). In an industrial configuration, the flame front is impacted by the v arious obstacles in its way (e.g. pipes, walls). One of the consequences of the presence of obstacles is an acceleration of the flame front (Moen et al. (1980), Moen et al. (1982)) and, consequently , man y e xperiments has been conducted to e v aluate this ef fect. The EXJET cam- paign (INERIS, GDF-SUEZ) has been set to study experimentally the ef fect of an obstacle module on the propagation of a flame front in man y configurations (steady flammable mix, methane-air jet, hydrogen-air jet). 2 Description of the experimental work The e xperimental work considered here has been carried out by INERIS and GDF-SUEZ (Sail et al. (2014)). The aim of the experiment w as to study methane jet e xplosion in a congested en viron- ment. The congestion module was a medium size (3m × 1m × 0.45m) obstacle module, consisting of 296 steel tubes of 2cm diameter . In the framew ork of the EXJET experimental setup, se veral e xperi- ments ha v e been carried out : methane jet dispersion within the module (without ignition), methane jet ignition within the module, and ignition of a steady methane-air mix. The steady methane-air mix 13 t h International Symposium on Hazar ds, Pr evention and Mitigation of Industrial Explosions Braunsc hweig, GERMANY - J uly 27-31, 2020 ISHPMIE 2020 Braunschweig, Germany Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 262 5 Conclusions In this study , a methane-air explosion in a confined en vironment has been simulated. The e xplosion "at rest" case of the EXJET (INERIS GDF-SUEZ) as been taken to assess for the fidelity of the simulation with OpenF oam R  . So far , se v eral aspects of the simulation ha ve sho wn good agreement with the e xperiment : The accelerating ef fect of the obstacles can be reproduced, and magnitude of both flame speed and ov er-pressure peaks are conserv ed. But the study fails to reproduce the accelerating beha vior of the flame front and hence the ov er-pressure signals obtained numerically are not accurately reproducing the e xperimental ones. The ke y to reproduce with fidelity the ov er -pressure signals (and more specifically , the ov er pressure peak v alues) is to capture with fidelity the flame propagation speed. Current works on the transport equation for Ξ are carried out to take care of that aspects. Results will be updated accord- ingly . Refer ences Bauwens, C., Chaf fee, J., Dorofee v , S. (2011). V ented e xplosion o verpr essur es fr om comb ustion of hydr ogen and hydr ocarbon mixtur es . International Journal of Hydrogen Energy , 36(3):2329 – 2336. ISSN 0360-3199. doi:https://doi.or g/10.1016/j.ijhydene.2010.04.005. The Third Annual International Conference on Hydrogen Safety . Ghasemi, E., Soleimani, S., Lin, C. (2014). Rans simulation of methane-air b urner us- ing local e xtinction appr oach within eddy dissipation concept by openfoam . International Communications in Heat and Mass T ransfer , 54:96 – 102. ISSN 0735-1933. doi: https://doi.or g/10.1016/j.icheatmasstransfer .2014.03.006. Gulder , O. (1984). Corr elations of laminar comb ustion data for alternative si engine fuels . Holzmann, T . (2019). Mathematics, Numerics, Derivations and OpenFO AM R  . J. Daubech, E. Leprette, C. P . (2016). F ormalisation du savoir et des outils dans le domaine des risques majeurs (eat-dr a-76) les explosions non confinées de gaz et de vapeur s - ω uvce . T echnical report, Institut National de l’En vironnement Industriel et des Risques. Moen, I., Donato, M., Kn ystautas, R., Lee, J. (1980). Flame acceler ation due to turb ulence pr oduced by obstacles . Comb ustion and Flame, 39(1):21 – 32. ISSN 0010-2180. doi: https://doi.or g/10.1016/0010-2180(80)90003-6. Moen, I., Lee, J., Hjertager , B., Fuhre, K., Eckhoff, R. (1982). Pr essur e development due to turb ulent flame pr opagation in lar ge-scale methaneair e xplosions . Comb ustion and Flame, 47:31 – 52. ISSN 0010-2180. doi:https://doi.or g/10.1016/0010-2180(82)90087-6. Moukalled, F ., Mangani, L., Darwish, M. (2015). The F inite V olume Method in Computational Fluid Dynamics: An Advanced Intr oduction with OpenFO AM R  and Matlab R  , v olume 113. ISBN 978- 3-319-16873-9. doi:10.1007/978-3-319-16874-6. Rao, V . C. M., Sathiah, P ., W en, J. X. (2018). Ef fects of congestion and confining walls on turb ulent defla grations in a hydr og en stor ag e facility-part 2: Numerical study . In- ternational Journal of Hydrogen Ener gy , 43(32):15593 – 15621. ISSN 0360-3199. doi: https://doi.or g/10.1016/j.ijhydene.2018.06.100. Sail, J., Blanchetiere, V ., Geniaut, B., Osman, K., Daubech, J., Jamois, D., Hebrard, J. (2014). Revie w of knowledge and r ecent works on the influence of initial turb ulence in methane e xplosion . In SKJOLD, T ., ECKHOFF , R. K., V AN WINGERDEN, K., editors, 10. International symposium on hazar ds, pr evention, and mitigation of industrial explosions (X ISHPMIE) , pages 401–432. Ge xCon AS, Ber gen, Norway . W eller , H., T abor , G., Gosman, A., Fureby , C. (1998). Application of a flame-wrinkling les com- b ustion model to a turb ulent mixing layer . Symposium (International) on Combustion, 27(1):899 – 907. ISSN 0082-0784. doi:https://doi.org/10.1016/S0082-0784(98)80487-6. T wenty-Sev enth Sysposium (International) on Comb ustion V olume One. Proceedings of the 13th Symposium International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions DOI: 10.7795/810.20200724 269 [Document text truncated for crawler view.]