Optimization and Quick Verification of an Electric Vehicle Rocker Design

A validated machine learning workflow for Li-ion battery crashworthiness. The trained predictor was reused after the design changed, with no new training data.

Evaluating a single rocker variant means running a full explicit side impact analysis. Each design revision starts that work over. This white paper documents a Li-ion battery module platform in a 60 km/h side collision against a rigid pillar, optimized using the ANSA Optimization tool with machine learning prediction.

The paper follows the workflow end to end, from dataset generation through to the verified result:

• 100 design experiments generated by Uniform Latin Hypercube sampling, with plate thicknesses and plate location as design variables
• Battery modeling with BatMac, based on the equivalent Randles circuit, with internal shorts triggered on cell stress
• Dataset quality checked before training, returning a predictive power score of 0.98 for mass and 0.64 for damaged cell count
• A Differential Evolution optimization minimizing damaged cells under a 36 kg mass constraint, converging on a 24.5 kg design
• Transfer learning applied to a revised design, ruling out a plate thickness change before any solver time was spent on it

Both the optimization and the verification results are reported in full evaluation tables. Written for crash and safety, EV platform, and body structure engineers.