A Multiphysics Dataset Generation Procedure for the Data-Driven Modeling of Traction Electric Motors

This paper presents the work done to address two main challenges in the simulation and design of electric machines for traction applications. On one hand, the modeling process is becoming increasingly complex as the demand for higher efficiency, high power density, and low cost pushes the speed and compactness of the motor to high levels. As a result, the interactions between multiple physical domains (e.g., electromagnetic, thermal, structural, etc.) can no longer be neglected, even in preliminary designs. Consequently, research into new modeling solutions in this area is currently active and widespread. On the other hand, new computational methodologies based on data-driven machine learning are becoming increasingly widespread as the computational power available for this task increases. However, to assess their performance and realize their potential in surrogate and meta-modeling electrical machines, a standardized benchmark for comparing these new approaches is needed. To address these challenges, the paper presents an open-source dataset that provides a reliable foundation for the multi-physical analysis of electric motors used in traction applications. One of the main novelties of this approach is that geometrical and physical data of the motor configuration are shared among different analysis codes. Attention is focused on tailoring the numerical discretization so that the same mesh can be used in different domains, avoiding data conversions and possible numerical inaccuracies. The paper thoroughly explains the workflow developed to create the database, detailing the methodological aspects. Ultimately, the resulting database is made available as an open resource for other researchers in the field. The resulting dataset represents a tool for benchmarking advanced computational methodologies and promoting reproducibility in research.