AI Accelerated Multi-Objective Optimization of High-Performance Axial Flux Powertrains Using A Holistic Data Driven Workflow
Electromagnetics
PRESENTER:
Maximilian Clauer, Porsche
EXPERTISE:
Electromagnetics
Software:
JMAG
The development of high-performance electric powertrains requires a precise understanding of the interactions between the electric machine, gearbox, inverter, battery pack, and vehicle dynamics, especially under race track conditions. Conventional finite element based design approaches provide the required accuracy, but they rapidly become impractical when large design spaces and system level optimization targets must be considered simultaneously. This presentation introduces a data driven optimization workflow for a high-performance axial flux electric drive, combining high fidelity JMAG finite element simulations with AI based surrogate and analytical models for efficient powertrain optimization.
The presented workflow starts with a fully parametric three dimensional finite element model of an axial flux machine, from which a simulation dataset is generated using transient three dimensional finite element simulations. Based on the generated data, AI models are used to represent the electric machine behavior within the multi-objective optimization fitness function. This enables fast evaluation of different designs while preserving the essential physical dependencies obtained from the high fidelity simulations. The presentation also discusses the required Multiphysics analysis of the axial flux machine, including electromagnetic performance evaluation, mechanical overspeed assessment, and thermal analysis. The machine model is embedded into a holistic powertrain optimization environment, in which all powertrain components and vehicle dynamics are evaluated combined. The approach therefore moves beyond isolated e-machine optimization and allows system level targets such as acceleration time, vehicle top speed, vehicle range, total vehicle mass, and race track lap time.
The presentation demonstrates how high fidelity JMAG simulation data can be leveraged for AI enabled optimization, accelerating design space exploration while maintaining high calculation accuracy. This approach supports therefore data driven decision making in powertrain development.
