ICEM24 - IPM machine model

From Geometric Model to System Simulation: Physical Modeling and Control of Electric Motors with Simulink and Simscape Feng He, Mathworks ([email protected]) Simone Ferrari, Politecnico di Torino ([email protected]) ICEM 2024, TORINO SEPT. 1-4, 2024 1 Content ◦Introduction ◦Part 1: Electric Motor Modeling, Feng He (45 min) ◦Part 2: SyR-e environment for eMotor Design and Analysis, Simone Ferrari (40 min) ◦Part 3: Beyond the Electric Motor, Feng He (45 min) ◦Conclusions ◦Q&A (20 min) ICEM 2024, TORINO SEPT. 1-4, 2024 2 About the speakers Feng He is an Application Engineer at MathWorks. He holds a M. Sc in Automation Engineering and Control of Complex Systems from Università degli Studi di Catania. Before joining MathWorks, he worked in the automotive industry: five years at General Motors as Application Software and Algorithm Development Engineer implementing the next generation thermal control for internal combustion engines; two years at Plastic Omnium (Austria) deploying a Model Based Design toolchain for the Fuel Cell control, simulation, and target HW integration. Simone Ferrari (Member, IEEE) receive the Ph.D. degree “cum laude” in 2020 from Politecnico di Torino, where he is currently an Assistant Professor. From July to December 2018, he was a Visiting Scholar at North Carolina State University, Raleigh, NC, USA. He is one of the authors of SyR-e, an open-source design tool for synchronous reluctance and permanent magnet machines. From 2021 he is also one of the responsible of the eDrive testing infrastructure TEST-eDRIVE managed from Energy Department and the Power Electronics Innovation Center of Politecnico di Torino. His research interests include electrical machine design and testing, and multi-physical evaluation of electrical machines, with a focus on synchronous reluctance and permanent magnet machines. ICEM 2024, TORINO SEPT. 1-4, 2024 3 Tutorial Map ICEM 2024, TORINO SEPT. 1-4, 2024 4 System Model (Simulink/SimScape) Battery Inverter Machine LoadControl Unit Commands Feedbacks Tutorial Map ICEM 2024, TORINO SEPT. 1-4, 2024 5 System Model (Simulink/SimScape) Battery Inverter Machine LoadControl Unit Commands Feedbacks Tutorial Map (Part 1) ICEM 2024, TORINO SEPT. 1-4, 2024 6 System Model (Simulink/SimScape) Battery Inverter Machine LoadControl Unit Commands Feedbacks •Model from datasheet •System identification •Model from lab data •Simple model (symbolic) •Import from external software •Introduction to system modeling •Simulink/SimScape Environment Tutorial Map (Part 2) ICEM 2024, TORINO SEPT. 1-4, 2024 7 System Model (Simulink/SimScape) Battery Inverter Machine LoadControl Unit Commands Feedbacks •FEA simulations •Magnetic Model Elaboration •SyR-e •syreDrive Tutorial Map (Part 3) ICEM 2024, TORINO SEPT. 1-4, 2024 8 System Model (Simulink/SimScape) Battery Inverter Machine LoadControl Unit Commands Feedbacks •Control design •Implementation •Calibration •Faults analysis •Sharing simulation models •Thermal management Content ◦Introduction ◦Part 1: Electric Motor Modeling, Feng He (45 min) ◦Part 2: SyR-e environment for eMotor Design and Analysis, Simone Ferrari (40 min) ◦Introduction ◦FEA simulation of PMSM ◦Motor Modeling ◦eDrive Model: syreDrive ◦Examples ◦Conclusions ◦Part 3: Beyond the Electric Motor, Feng He (45 min) ◦Q&A (20 min) ICEM 2024, TORINO SEPT. 1-4, 2024 60 Content ◦Introduction ◦Part 1: Electric Motor Modeling, Feng He (45 min) ◦Part 2: SyR-e environment for eMotor Design and Analysis, Simone Ferrari (40 min) ◦Introduction ◦FEA simulation of PMSM ◦Motor Modeling ◦eDrive Model: syreDrive ◦Examples ◦Conclusions ◦Part 3: Beyond the Electric Motor, Feng He (45 min) ◦Q&A (20 min) ICEM 2024, TORINO SEPT. 1-4, 2024 67 FEA Simulation of PMSM Computational time is a crucial aspect for FEA simulation. To reduce the model complexity: ◦2D model instead of 3D model is adopted ◦Periodic or anti-periodic condition (based on adopted winding) ◦Rotation of less than 1 period, thanks to electric symmetry ◦Time-Step Static FEA simulation →FEMM ◦Current-controlled simulation (already in steadystate) ICEM 2024, TORINO SEPT. 1-4, 2024 68 3D→2D 2p→1 360°→60° Operating Point FEA Simulation ICEM 2024, TORINO SEPT. 1-4, 2024 69 dq→abc abc→dq1→2p 60°→360° 𝜃𝑟 𝑖𝑎𝑏𝑐 𝑖𝑑𝑞 FEA Simulations with SyR-e FEA simulations are launched from the dedicated tab on the main GUI Several evaluation types possible: ◦Single operating point ◦Flux maps ◦Demagnetization analysis ◦Flux density analysis ◦Structural analysis ◦HWC peak short-circuit current ◦… Besides FEMM, some FEA commercial SW are supported ICEM 2024, TORINO SEPT. 1-4, 2024 70 Results: Flux Density Waveform From the same FEA simulations, the airgap flux density waveform can be extracted for each rotor position. ◦Not included in standard eDrive models, but advanced models can be implemented ◦Important for design verification ◦The data can be further manipulated to get the harmonic content ◦Stator yoke and tooth flux density can be extracted, too ICEM 2024, TORINO SEPT. 1-4, 2024 71 Results: Airgap Magnetic Pressure (NVH) The airgap magnetic pressure is one of the sources of eNVH issues. ◦Not included in the standard eMotor model, but mandatory for NVH analysis ◦Symmetry conditions are adopted to get the full motor / full period characteristic ◦Tooth forces are computed by integration of the magnetic pressure over one tooth ICEM 2024, TORINO SEPT. 1-4, 2024 72 Results: Iron and PM Loss Iron and PM loss are computed offline, after the FEA simulation The data are imported in Matlab for each mesh element and rotor position: ◦The flux density for lamination elements ◦The magnetic potential for PM elements The iron loss are computed with iGSE: ◦Hysteresis loss are obtained from main and minor loops and the related loss coefficient ◦Eddy-current loss are obtained from FFT and the related loss coefficient ◦PM loss are obtained by from FFT and the PM conductivity Computational time: ~5min ICEM 2024, TORINO SEPT. 1-4, 2024 73 Demagnetization Analysis Demagnetization is a critical problem for PMSM In general, FEA uses linear PM curves ◦Knee point is identified as limit for correct modeling ◦If non-conventional PMs are adopted, the nonlinear BH curve must be considered The demagnetization analysis is performed at given temperature and current ◦The flux density in each PMs mesh element is considered and compared with the knee point ◦Magnetization direction only is considered ICEM 2024, TORINO SEPT. 1-4, 2024 74 80°C, 1940Apk 80°C, 3100Apk Demagnetization Limit Besides the single point, a useful information is the demagnetization limit: ◦maximum current, injected against the PM flux linkage, that irreversibly demagnetize 1% of the PMs volume ◦Identifiedin SyR-e through an iterative process ◦Function of the PM temperature ◦NB: the shape and slope of the curve is function of PM material andnmotor geometry! ICEM 2024, TORINO SEPT. 1-4, 2024 75 AC Loss Modeling For hairpin winding, loss due to AC effects (skin effect and proximity) are crucial AC loss is computed in SyR-e with timeharmonic FEA over a 𝑓, Θ𝐶𝑢 map ◦Unsaturated iron ◦Active section only The result is the AC factor, defined as: 𝑘𝐴𝐶 =𝑃𝐶𝑢,𝐴𝐶 𝑃𝐶𝑢,𝐷𝐶 Computational times: ◦Less than 1s for a single simulation ◦About 36s for the entire map (parallel computing) ICEM 2024, TORINO SEPT. 1-4, 2024 76 AC Loss Model FEA-computed AC loss factor is applied just on the active section ◦End-windings have almost no AC effect ◦Effective AC factor is computed ◦In some cases, the AC factor of the end-winding can be defined (via 3D FEA or experimental results) Phase resistance map is computed ◦Function of temperature and frequency ◦Account for both AC loss and temperature effects on DC resistance ICEM 2024, TORINO SEPT. 1-4, 2024 83 Inductance Maps Computation Inductance maps can be adopted for eMotor dynamic modeling (e.g. VBR) Two models are typically considered: ◦Apparent inductances (Vs/A ratio) 𝝀𝒎= 𝜆𝑚𝑖𝑞, 𝑳𝒅=Λ𝑑𝑖𝑑, 𝑖𝑞− 𝜆𝑚𝑖𝑞 𝑖𝑑 , 𝑳𝒒=Λ𝑞𝑖𝑑, 𝑖𝑞 𝑖𝑞 ◦Incremental inductances (differentiation) 𝒍𝒅𝒅 =𝜕Λ𝑑𝑖𝑑, 𝑖𝑞 𝜕𝑖𝑑 , 𝒍𝒒𝒒 =𝜕Λ𝑞𝑖𝑑, 𝑖𝑞 𝜕𝑖𝑞 𝒍𝒅𝒒 =𝜕Λ𝑑𝑖𝑑, 𝑖𝑞 𝜕𝑖𝑞 = 𝒍𝒒𝒅 =𝜕Λ𝑞𝑖𝑑, 𝑖𝑞 𝜕𝑖𝑑 Both inductances modeling can be computed through 𝑑𝑞 flux maps elaboration with SyR-e ICEM 2024, TORINO SEPT. 1-4, 2024 84 Inverse Model Computation Inverse flux maps are another useful framework for eMotor modeling (e.g. CCG). ◦Currents and torque are expressed function of flux linkages ◦Both 𝑑𝑞 and 𝑑𝑞𝜃 inverse model The model inversion is performed with scatteredInterpolant Matlab function The maps limits must be carefully considered: ◦If rectangular domain is pursued →data loss ◦Extrapolation should be avoided ◦For loss-less inversion →non-rectangular domain ◦Normalization techniques can be implemented to avoid data loss ICEM 2024, TORINO SEPT. 1-4, 2024 85 Computation of Control Trajectories Control trajectories are directly computed from 𝑑𝑞 flux and torque maps ◦MTPA: Maximum Torque per Ampere ◦MTPV: Maximum Torque per Voltage (~flux) The same evaluation routine is adopted: ◦A vector of currents (or flux linkages) amplitudes is considered ◦For each current (or flux linkage) level, the maximum torque is identified and the corresponding 𝑖𝑑, 𝑖𝑞coordinates are stored ICEM 2024, TORINO SEPT. 1-4, 2024 86 Torque –Speed Limits With the control trajectories and the inverter ratings, the operating limits can be identified, almost istantaneously ◦Different current and voltage limits ◦Loss-less condition ICEM 2024, TORINO SEPT. 1-4, 2024 87 Efficiency Maps Computation Efficiency maps are computed over a torquespeed regular grid Several loss components can be included ◦Copper loss (DC and AC components) ◦Iron and PM loss can be considered (SIN supply) ◦Mechanical loss can be included Inverter limits are considered Post-fault conditions (ASC and OC) are considered and evaluated Computational time: ◦About 13 min for 10201 points (101x101 grid) ◦Lower discretization can be adopted ◦Parallel computing not (yet) implemented ICEM 2024, TORINO SEPT. 1-4, 2024 88 Efficiency Maps Computation For each 𝑇, 𝑛 point, the flux and loss maps are scaled and the 𝑖𝑑, 𝑖𝑞operating point is computed ◦Point that are not compatible with the inverter are dropped ◦The minimum copper loss point is identified ICEM 2024, TORINO SEPT. 1-4, 2024 89 Symmetric Short-Circuit Condition Symmetric short-circuit condition (3-phase short-circuit) can be modeled through flux maps ◦Valid for all the motors modeled in this framework Steady-state short-circuit (SSSC): ◦Torque and current function of speed ◦Voltage maps computed over the 𝑑𝑞 plane and SC point identified Transient short-circuit (TSC): ◦Torque and current function of time ◦Inverse maps needed ◦Differential equation discretized and solved with back-Euler method ◦Pre-fault condition must be set ICEM 2024, TORINO SEPT. 1-4, 2024 90 Content ◦Introduction ◦Part 1: Electric Motor Modeling, Feng He (45 min) ◦Part 2: SyR-e environment for eMotor Design and Analysis, Simone Ferrari (40 min) ◦Introduction ◦FEA simulation of PMSM ◦Motor Modeling ◦eDrive Model: syreDrive ◦Examples ◦Conclusions ◦Part 3: Beyond the Electric Motor, Feng He (45 min) ◦Q&A (20 min) ICEM 2024, TORINO SEPT. 1-4, 2024 91 eDrive Dynamic Model eDrive modeling is implemented in Simulink/SimScape or other similar environment syreDrive: direct interface between eMotor design, FEA simulation and the eDrive dynamic model ◦Inverter modeled with SimScape block (AVG or PWM) ◦Different eMotor models ◦C-based control algorithm, current, torque and speed control available and auto-tuned ◦Sensorless control scheme included ICEM 2024, TORINO SEPT. 1-4, 2024 92 Transient Short-Circuit is then considered: ◦Rated temperature ◦Pre-fault point: max torque at base speed TSC trajectory: ◦Follows a spiral in the 𝜆𝑑, 𝜆𝑞domain ◦Damping due to copper loss ◦In the 𝑖𝑑, 𝑖𝑞domain, the trajectory is ellyptical, centered on the SSSC point ICEM 2024, TORINO SEPT. 1-4, 2024 99 Example #1: Short-Circuit Analysis Example #1: Short-Circuit Analysis Temperature effect is studied in transient conditions ◦Pre-fault at base speed and peak torque The effect is coherent with the SSSC ◦Higher temperture reduce the peak current and torque ◦Hot condition is critical for demagnetization issue (limit current drastically decreased with temperature) ICEM 2024, TORINO SEPT. 1-4, 2024 100 Example #1: Short-Circuit Analysis Pre-fault condition is investigated at rated temperature Four points are considered: ◦Peak torque, rated speed ◦Peak torque half rated speed ◦No load, rated speed ◦No load maximum speed Results ◦Peak current is function of the pre-fault flux linkage ◦At given flux linkage, lower the speed, lower the peak current (more time for damping) ICEM 2024, TORINO SEPT. 1-4, 2024 101 Example #2: Efficiency and Loss Maps The efficiency maps are computed with sinusiudal current ◦Imposed inverter limits ◦Constant Cu and PM temperature ◦MTPA + field weakening is implemented ◦Different loss terms can be considered ICEM 2024, TORINO SEPT. 1-4, 2024 102 Example #2: Efficiency and Loss Maps Loss components can be studied and separated: ◦Copper loss ◦AC effect on copper loss ◦Iron loss ◦PM loss ◦Mechanical loss ◦… ICEM 2024, TORINO SEPT. 1-4, 2024 103 Example #2: Efficiency and Loss Maps The maps can be computed at different temperatures, to understand the effect on: ◦Peak torque ◦Peak power ◦Efficiency ICEM 2024, TORINO SEPT. 1-4, 2024 104 Example #2: Efficiency and Loss Maps The control algorithm is automatically implemented: ◦Control locus are retrieved for each (𝑇, 𝑛) point ◦Different control strategies can be considered, included loss minimization ICEM 2024, TORINO SEPT. 1-4, 2024 105 Example #2: Efficiency and Loss Maps With the 𝑑𝑞 currents, all the quantities from flux and loss maps can be retrieved: ◦Flux linkages ◦Torque ripple ◦Power factor ◦Phase current and voltage ◦… ICEM 2024, TORINO SEPT. 1-4, 2024 106 Example #2: Efficiency and Loss Maps Post-fault safe state condition can be estimated over the plane: ◦OC: the no-load voltage must be lower than the DC link (otherwise, UGO happens) ◦ASC: peak short circuit current (HWC for computational burden) must be lower than the demagnetization limit At rated temperature: ◦ASC is always safe ◦OC is dangerous above about 7000rpm The turn-off modes changes with temperature! ICEM 2024, TORINO SEPT. 1-4, 2024 107 Example #2: Efficiency and Loss Maps The safe states change with temperature. Increasing temperature: ◦PM flux linkage reduces →OC unfeasible area reduces ◦Demag limit reduces →ASC unfeasible area increase ICEM 2024, TORINO SEPT. 1-4, 2024 108 Example #3: Torque Control with syreDrive CCG model is adopted with PWM inverter ◦PWM-induced current ripple is visible ◦Small to no effects on torque ripple ◦The effect change with speed and load! Computational time are increase: ◦AVG inverter: 18s ◦PWM inverter: 36s ◦Computational effort is increased also with SimScape PMSM block NB: additional loss MUST be re-computed with dedicated FEA simulations! ICEM 2024, TORINO SEPT. 1-4, 2024 115 Example #4: PWM Effect and Loss Efficiency maps was computed with sinusoidal supply, however, PWM can affect losses eDrive model cannot properly estimate the iron loss (loss model based on sinusoidal supply) Solution: PWMfix procedure ◦Some points of the 𝑇, 𝑛 domain are evaluated with eDrive model and currents with PWM ripple are retrieved ◦Dedicated FEA simulations are performed with custom currents to compute the PWM-induced loss ◦Correction factors are computed and extended to the whole efficiency map ICEM 2024, TORINO SEPT. 1-4, 2024 116 Content ◦Introduction ◦Part 1: Electric Motor Modeling, Feng He (45 min) ◦Part 2: SyR-e environment for eMotor Design and Analysis, Simone Ferrari (40 min) ◦Introduction ◦FEA simulation of PMSM ◦Motor Modeling ◦eDrive Model: syreDrive ◦Examples ◦Conclusions ◦Part 3: Beyond the Electric Motor, Feng He (45 min) ◦Q&A (20 min) ICEM 2024, TORINO SEPT. 1-4, 2024 117 Recap and Conclusions The SyR-e approach in the motor modeling is presented The motor modeling is based on flux and loss maps, in the 𝑑𝑞 reference frame. Detailed FEA simulations can be performed for specific motor behaviors Magnetic model elaboration based on flux and loss maps are implemented for several motor metrics evaluation syreDrive is the interface with eDrive model in Simulink. It is automatic and based on the same flux and maps previously computed. The user can select several modeling apporach, from the simplest (average inverter and motor, lossless) to the most complex (PWM inverter, motor with spatial harmonics, with loss). A model of traction PMSM is added to the SyR-e repository as a demo from today’s tutorial SyR-e is constantly evolving under the push of industry and new research challenges We invite you to try SyR-e and cooperate with SyR-e Team! ICEM 2024, TORINO SEPT. 1-4, 2024 118 Galileo Ferraris’ Contest Comparing Data-Driven Methodologies for the Multi-Physics Simulation of Traction Electrical Machines ICEM 2024, TORINO SEPT. 1-4, 2024 119 sponsored by https://cadema-polito.github.io/GalFer_contest/ under patronage of International Compumag Society (ICS)