Sensor integrated bipolar plate for mitigation of critical conditions during PEM fuel cell system operation

EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476271 B0711 Page 1/9 B0711 Sensor integrated bipolar plate for mitigation of critical conditions during PEM fuel cell system operation Bhanu Seth*, Jamil Kharrat, Jan Haußmann IPEK – Institute of Product Engineering, Karlsruhe Institute of Technology (KIT); *Contact corresponding authors: www.EFCF.com/ContactRequest Abstract The operation of PEM fuel cell systems in mobile applications requires a flexible operation due to the dynamic power demand of the drive system. The dynamic load changes in the fuel cell system are limited due to the gas supply components such as the air compressor and the hydrogen recirculation unit. Insufficient mass flow changes during the fuel cell operation may lead to reduced stoichiometries at the fuel cell electrodes. An undersupply of hydrogen and oxygen can lead to degradation of the catalyst layer materials which need to be prevented to enhance the fuel cell system lifetime. As application case a fuel cell system of a fuel cell-battery hybrid truck is operated considering three different power levels. The experiments revealed a significant delay of air mass flow and its influence on a reduced stoichiometry at the fuel cell stack. Besides, the influence of the dynamic operation on temperature, pressure and relative humidity is measured. A simulative study considering the operating conditions of the experiment revealed the influence of the stoichiometry on the local gas concentration distribution and its effect on the local temperature and relative humidity. To overcome the critical states during dynamic operation of the fuel cell, a concept is proposed where in addition to sensors at the inlet and outlet of the fuel cell stack, micro sensors are integrated into the bipolar plate to measure the distribution of temperature and relative humidity in real time. Furthermore, the concept is extended by a soft sensor approach to increase the resolution of the temperature and humidity distribution without increasing the design complexity and costs of the fuel cell stack. The results show that soft sensors based on deep neural networks can predict the local gas conditions on the bipolar plate with high precision. Thus, a combined approach of micro sensors and soft sensors has the potential to detect critical states within the fuel cell and can therefore improve real-time control of the fuel cell system and can help to further increase the fuel cell lifetime. EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476271 B0711 Page 2/9 Introduction Polymer electrolyte membrane fuel cells (PEMFCs) are developing into a key technology for sustainable energy conversion. While fuel cells in long-distance applications typically operate under quasi-stationary conditions, dynamic operation poses a major challenge in short-range logistics and city transport. Due to their limited ability to respond instantly to rapidly changing power demands, PEM fuel cells are unable to follow dynamic load profiles directly. This is primarily due to the gas supply delays. To address these limitations, hybrid powertrains combining fuel cells with batteries are commonly employed. In these systems, the battery provides immediate power during transient phases, allowing the fuel cell to adjust gradually toward its optimal operating point [1]. A simulation environment can be configured to the real-world application data to systematically analyze the impact of dynamic operation on the fuel cell efficiency, the thermal management, and the degradation behavior, which are mainly influenced by the load changes [2]. In this study, a 10-ton battery-fuel cell hybrid truck, designed for various transport scenarios with three power levels (low, medium and high), was used as the test vehicle under realistic operating conditions. As demonstrated in the study by Liu et al. [3] delayed air supply as well as hydrogen supply during dynamic load changes has significant negative effects on the operation of PEM fuel cells. In particular, there is a delay to reach the required air mass flow and pressure. The caused delay of the required air mass flow may result in an imbalance of the reactants within the fuel cell, a condition called starvation where not enough reactants are supplied for the electrochemical reaction. Additionally, the delayed supply may also result in a condition called ‘flooding’ where the liquid water is not removed by the flowing gas. This blockage may enhance in turn the starvation conditions as the reactants cannot reach the reaction sites freely [4,5]. Prolonged operation under starvation conditions may lead to material degradation in the membrane electrode assembly (MEA). This includes reduction in electrochemical active surface area (ECSA), erosion of carbon support material in gas diffusion layers (GDL) and increase in catalyst particle size [6]. To avoid such critical states, the state-of-the-art fuel cell systems incorporate sensors at the inlet and the outlet of the fuel cell to monitor the gas conditions [7]. Based on the conditions of the reactant gases entering and leaving the system, the overall state of the fuel cell can be determined. Conditions such as flooding or drying can be accessed based on these parameters and suitable adjustment in the ‘Balance-of-Plant’ (BoP) parameters can be made to ensure that the fuel cell operates in an optimal state. Since the fuel cell operation is complex involving physical, thermal and chemical interactions, there exist gradients across the active area of the fuel cell, and hence it is not sufficient to measure the inlet and outlet conditions. Dynamically changing loads and the associated electrochemical reactions can lead to conditions where the inlet and outlet conditions remain in the optimal window, while small criticalities such as membrane drying, flooding, or hotspot creation is taking place at a localized level [8]. Hence, our solution is to integrate sensors directly on the bipolar plate to spatially measure the conditions inside the fuel cell during dynamic operation. Sensors such as temperature and humidity sensors can help in forming a spatially resolved image of the fuel cell in real time. Preventive measures can then be taken by the system controller in resolving not only system level but also localized critical states by adjusting the BoP component operation, thus reducing the degradation even further and enhancing the fuel cell life. EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476271 B0711 Page 3/9 Integrating sensors in a fuel cell bipolar plate (BPP) is both an economic as well as a technical challenge. A BPP typically is only a few millimeters thick and incorporating the sensors with their signal cables is itself technically challenging keeping in mind the sealing requirements of the fuel cell. Parallelly, integration of sensors comes at an additional cost. In theory to enhance the spatial detection resolution, the matrix of sensors in length and width of the BPP should be increased, but that will increase the cost and technical feasibility on stack level which consists of hundreds of cells. To solve this issue, a machine learning based soft sensor approach was also incorporated to enhance the spatial resolution of the measurements without the increase in additional cost [9]. The approach starts with gathering the real time measurements from real sensors in a controlled test environment where the fuel cell stack is tested under its normal operation limits. Using machine learning modelling approaches like deep neural networks, a model is trained to capture the relationship between the input operating conditions and the localized temperature and humidity value. The trained model can then be used to predict the localized temperature and humidity value, hence limiting the need of physical sensors to be placed inside the BPP to a minimum. Furthermore, the fuel cell undergoes aging processes and hence the relationships between operating conditions and the localized state also changes with time. Hence, a combined matrix of real and soft sensors is an ideal configuration to enhance the spatial resolution without increasing the fuel cell cost. Here the real sensors are used for real time data gathering and soft sensor calibration as the fuel cell stack experiences aging. 1. Scientific Approach The air compressor characteristics directly influence the reactant supply dynamics and hence, also directly affect the starvation possibilities during dynamic fuel cell operation. These characteristics define how fast the compressor can respond in delivering the required air mass flow at the required pressure. To see the effect of load changes on the compressor performance, a system simulation was conducted. A fuel cell system model developed by MathWorks® in Simscape™ was adjusted to match the characteristics of a fuel cell hybrid truck and then three load cases, low, medium and high-power demands were simulated [10]. Previous simulations showed that at every load change, there is a slight delay in delivering the required air flow, after which the air compressor adjusts its flow rate and maintains the required air flow. This temporary reduction in stoichiometry was analyzed using spatially resolved simulations to understand the gradients across the bipolar plate. Here a spatially resolved fuel cell model developed by Feierabend [11] was used to simulate and analyze the local conditions in terms of reactant concentration, relative humidity, current density, and temperatures etc. Spatially resolved models pointed out the patterns of the gradients along the flow direction and hence also helped in determining potential sensor locations for optimal converge of data points. 2. Experiments/Calculations/Simulations In this study, the analysis of the dynamic operation of a PEM fuel cell system is based on an electrified truck equipped with a fuel cell–battery hybrid powertrain. The truck features a PEM fuel cell system with a maximum power output of 67 kW, integrated with a battery system capable of delivering 100 kW of continuous power and up to 240 kW peak power. Together, they supply an electric motor rated at 170 kW. The technical specifications of the hybrid drivetrain are summarized in Table 1. Table 1: Technical specifications of the fuel cell truck EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476271 B0711 Page 4/9 System Specification Vehicle Iveco Eurocargo 10t Fuel Cell electric power 67 kW Hydrogen tank 14 kg, 700 bar Battery electric power 100 kW (max. 240 kW) Battery capacity 72 kWh Electric motor power 170 kW The fuel cell system is configured to operate under predefined low, medium, and high-power conditions. Each power level requires distinct air and hydrogen mass flow rates, along with corresponding gas pressures on both the anode and cathode side of the fuel cell stack. Throughout the tests, the relative humidity at the gas inlets and the coolant inlet temperature to the stack are maintained constant. The fuel cell system model is representing the different paths such as cathode path, anode path, coolant path and the electrical circuit. The different components in each path are individually parametrized to simulate the behavior of the defined fuel cell system. Table 2 contains the system simulation boundary conditions: Table 2: Boundary conditions system simulation Parameter Anode Cathode Stoichiometry [λ] 2.1 1.5 Inlet Relative Humidity (RH) [%] 50 Electrical Load [A] 185-233-291 (Low-Medium-High) Active Area (length x width) [cm²] 280 (22,4 x 12,5) Similar boundary conditions were used for the spatially resolved simulations in the spatially resolved simulation tool. The Boundary conditions were selected to simulate conditions representing the high load case with dynamic load changes where the cathode stoichiometry is reduced to 1 instead of the normal stoichiometry value 1.5. Lastly, to establish the soft sensor approach, system level data was generated from the Simscape™ model to provide the database for machine learning modelling. This generated data consisted of inlet conditions used as inputs and include parameters such as current, voltage, pressure, mass flow rates etc. It also contains the predicted value, in this case the relative humidity value at the cathode outlet. The machine learning model was then trained on this dataset to train, validate and test the approach. 3. Results The results of the real-system tests, as illustrated in Figure 1, demonstrate that a rapid increase in current demand (load change) is followed by a corresponding rise in air mass flow. However, this increase occurs with a low gradient, indicating a delay in the air supply response. EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476271 B0711 Page 5/9 This delay affects the cathode operating pressure and the system temperature as depicted in Figure 1. Due to the thermal inertia the increase in temperature is delayed and shows a rise of 2 °C in the displayed time range, reaching a stable temperature at 42 °C. The relative humidity is rising from 94 % to 99 % whereas the air pressure is only slightly increased by 5 mbar. By comparing the increase in electric current and air mass flow, the impact on the cathode stoichiometry instead is clearly observable. During the load change, this parameter exhibits a significant drop of up to 36%. Figure 1: Measurement data from the operation of the fuel cell system under an electric load change. Figure 2 depicts the system simulation results. It shows that at the instance of the load change, due to the lack of the available air, the stoichiometry at the cathode of the fuel cell sinks before regaining its normal level. This change in stoichiometry can be detrimental under extreme load variations and cause damage to the fuel cell by causing temporary reactant starvation. EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476271 B0711 Page 6/9 Figure 2: Stoichiometry fluctuation for anode and cathode at load changes Figure 3 shows the resulting spatially resolved gradients at the active area of the fuel cell stack for the reactant concentration drop due to the change in stoichiometry. The reactant stoichiometry at the lower level, i.e., lambda 1 shows a faster concentration drop, while at the normal level of lambda 1.5, the concentration drop is considerably lower. Figure 3: Concentration gradient on the cathode at different stoichiometries Figure 4 shows the resulting effect of a reduction in stoichiometry on relative humidity and temperature. Here it is seen that the stoichiometry has a direct influence on the relative humidity and temperature gradients of the fuel cell. Although the simulations show a relationship between the parameters, the quantitative effect needs to be measured in the real experiments. Hence the integration of temperature and relative humidity sensors will enable us to verify and quantify the above observed effect. EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476271 B0711 Page 7/9 Figure 4: Relative humidity and temperature gradients on the cathode at different stoichiometries With the help of real time information from the sensors such as relative humidity and temperature, it will be possible to not only observe the gradients, but also to include them in the fuel cell control system to directly influence the critical states. Factors such as stoichiometry can be used to control the relative humidity gradients and avoid critical conditions such as flooding or drying of the fuel cell stack thereby enhancing the lifetime of the fuel cell. Additionally, the soft sensor approach, as explained above, is tested successfully in predicting the relative humidity values based on given operating conditions. Thus, enhancing the spatial resolution without increasing the system costs. Figure 5 shows the ‘Soft Sensor’ approach. A Deep Neural Network (DNN) is trained on real sensor data to create a virtual sensor. This soft sensor is incorporated to predict the relative humidity at the cathode based on the operating conditions of the fuel cell. The integrated sensors can be used in a matrix of real and soft sensors to enhance the spatial resolution while reducing the complexity and cost of the system. Figure 5: Soft Sensor approach and Deep Neural Network Figure 6 shows the simulation run of a soft sensor prediction in comparison to the simulated values via the model. EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476271 B0711 Page 8/9 Figure 6: Relative humidity prediction of soft sensor in comparison to simulated data The deep neural network based soft sensor was able to predict the relative humidity conditions based on the inlet operating conditions extremely well with a low Root Mean Squared Error (RMSE) value of 0.019 averaged across three driving cycles i.e., WLTP, FTP75 and NEDC, hence proving its ability to provide soft sensing capabilities under dynamic loading conditions [12]. References [1] Y. Tang, W. Yuan, M. Pan, Z. Li, G. Chen und Y. Li, "Experimental investigation of dynamic performance and transient responses of a kW-class PEM fuel cell stack under various load changes," Applied Energy, Jg. 87, Nr. 4, S. 1410–1417, doi: 10.1016/j.apenergy.2009.08.047., 2010. [2] H. Chen, X. Zhao, T. Zhang und P. Pei, "The reactant starvation of the proton exchange membrane fuel cells for vehicular applications: A review," Energy Conversion and Management, Jg. 182, S. 282–298, doi: 10.1016/j.enconman.2018.12.049., 2019. [3] Z. Liu, B. Zhang und S. Xu, "Research on air mass flow-pressure combined control and dynamic performance of fuel cell system for vehicles application," Applied Energy, Jg. 309, S. 118446, doi: 10.1016/j.apenergy.2021.118446, 2022. [4] Chen, H. et al., ‘Mechanism analysis of starvation in PEMFC based on external characteristics’, International Journal of Hydrogen Energy, 44(11), pp. 5437–5446. doi:10.1016/j.ijhydene.2018.11.135., 2019. [5] Nguyen, H.L. et al., ‘Dynamic flooding management strategy for automotive proton exchange membrane fuel cell system using cathode membrane water content’, Applied Thermal Engineering, 273, p. 126485. doi:10.1016/j.applthermaleng.2025.126485., 2025 [6] Ren, P. et al., ‘Degradation mechanisms of proton exchange membrane fuel cell under typical automotive operating conditions’, Progress in Energy and Combustion Science, 80, p. 100859. doi:10.1016/j.pecs.2020.100859., 2020. [7] Wang, H. et al., ‘Identification of essential sensors for a PEMFC system in automotive applications’. Available at: https://doi.org/10.1109/vppc49601.2020.9330822, 2020. [8] Miao, T. et al.,‘Current density and temperature distribution measurement and homogeneity analysis for a large-area proton exchange membrane fuel cell’, Energy, 239, p. 121922. Available at: https://doi.org/10.1016/j.energy.2021.121922., 2022. [9] Sun, K. et al., ‘Soft sensor development with nonlinear variable selection using Nonnegative Garrote and Artificial Neural Network’, Computer Aided Chemical Engineering, pp. 883–888. doi:10.1016/b978-0-444-63456-6.50148-4, 2014. [10] The MathWorks Inc., MATLAB version: 9.14.0 (R2023a), The MathWorks Inc, MA, USA, can be found under https://www.mathworks.com, 2023. [11] Lukas Feierabend, PEM Fuel Cell Stack Model (Version v1.0.0) [Computer software]. https://doi.org/10.5281/zenodo.7611662, 2023. 0 200 400 600 00 1000 1200 1400 1600 1 00 Time (s) 0 50 100 150 km h 0 200 400 600 00 1000 1200 1400 1600 1 00 Time (s) 0 100 200 A 0 200 400 600 00 1000 1200 1400 1600 1 00 Time (s) 0.4 0.6 0. 1 H Simulation Prediction 0 500 1000 1500 2000 2500 Time (s) 0 50 100 150 km h 0 500 1000 1500 2000 2500 Time (s) 0 100 200 A 0 500 1000 1500 2000 2500 Time (s) 0.4 0.6 0. 1 H Simulation Prediction EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476271 B0711 Page 9/9 [12] Seth, B., & Haußmann, J., Machine Learning Based Soft Sensor Development for automotive fuel cell systems. https://doi.org/10.60687/2025-0066, 2025 Keywords: EFCF2025, H2, Low-Temp. Fuel Cells & Electrolysers, PEMFC, Automotive, Soft-Sensor, Sensor integration, degradation Remark: This work is licensed under Creative Commons Attribution 4.0 International