3D Generative Automated NVH Optimization of EV Motor Housing

This paper presents an advanced automated optimization framework for the radiated noise performance of electric vehicle (EV) motor housings. The proposed methodology integrates multiple AI-driven techniques to enhance both accuracy and computational efficiency. First, a modal reduced-order model (ROM) enables the rapid generation of training data, allowing real-time updates to the AI prediction model during the optimization process. Second, a 3D shape recognition AI accelerates prediction speed, surpassing conventional CAE-based simulations. Third, a generative design approach utilizing morphing techniques facilitates the creation of diverse candidates for optimized geometries, potentially benefiting from historical designs as well. By seamlessly integrating these components, an efficient optimization workflow is established. The developed framework was applied to an EV motor housing, achieving an NVH prediction accuracy within 2 dB and identifying an optimized design within a single day. This research demonstrates the potential of AI-driven generative optimization combined with reduced-order modeling to accelerate the NVH design process.