The Galileo Ferraris contest: A benchmark initiative for data-driven multi-physics modeling of traction electric motors

The Galileo Ferraris Contest is a benchmark initiative aimed at evaluating data-driven surrogate modeling methodologies for multi-physics simulation of traction electric motors. The design of such motors traditionally relies on high-fidelity finite-element simulations, which are accurate but introduce severe computational bottlenecks for large-scale design exploration. The use of emerging surrogate modeling strategies offers a promising path to overcome these limitations. Yet, a systematic and fair comparison of their capabilities for realistic electric motor design is still lacking. Built upon an open-access dataset including electromagnetic, thermal, and structural results for three families of V-shaped interior permanent magnet motors, the contest provided a standardized testbed to assess interpolation, extrapolation, and innovation capabilities of surrogate models. A uniform multi-objective optimization and FEM validation pipeline was applied to all participant models, ensuring fair comparison across different machine learning strategies. The results show that different approaches successfully reproduced the input–output relationships of the reference motors, achieving accurate predictions even in unseen design regions. Moreover, when integrated into the optimization loop, surrogate models identified Pareto-optimal configurations beyond the resolution of the original dataset, enabling physically-consistent design exploration at negligible computational cost compared to direct finite element analysis. By establishing a reproducible framework, the Galileo Ferraris Contest preliminarily validated surrogate models as essential tools for electric motor optimization and advanced the standardization of data-driven multi-physics design workflows for future research.

• View at Publisher

Full Text ( Final Version , 11mb )