Smashing computational barriers: data-driven ball-impact modeling on AWS

This article discusses a groundbreaking approach to predicting ball-impact simulations using machine learning techniques on AWS, offering significant computational advantages over traditional finite element methods (FEM).

• Developed machine learning models (U-Nets and Fourier Neural Operators) to predict transient structural responses
• Achieved prediction speeds up to 10,000 times faster than traditional FEM simulations
• Created a multi-resolution approach balancing accuracy and long-term stability
• Utilized AWS services like AWS Batch and Amazon EFS for large-scale data generation and training
• Generated a synthetic dataset of 6,500 unique ball-impact scenarios for model training

The research demonstrates the potential of machine learning in accelerating complex engineering simulations, with promising applications in automotive, aerospace, and consumer electronics industries.