Using machine learning to drive faster automotive design cycles

This article discusses how automakers can leverage machine learning models to accelerate the automotive design process, reducing the time required for iterative design cycles from hours or days to seconds.

Specifically, the article covers:

• Background on the typical automotive product design process involving CAD geometry creation and engineering simulations for validation.
• Using AI/ML techniques like MeshGraphNets, U-Nets, and Variational Autoencoders to create surrogate models that can quickly predict aerodynamic metrics like coefficient of drag and flow fields, reducing the need for time-consuming physics-based simulations.
• Building a web application on AWS using services like EC2, Batch, S3, and SageMaker, to allow engineers to interactively explore design variations and get rapid feedback from the ML models.
• Training the ML models on synthetic data generated by morphing baseline car geometries and running CFD simulations on AWS Batch.
• Integrating the rapid ML-based design iterations into the overall product engineering process to explore more designs faster and enable more collaborative reviews.