A Comparative Analysis of Physics-Based and Machine Learning Models for Bouncing Ball Energy Dissipation

This study investigates energy loss in bouncing-ball collisions on different surfaces using both a physics-based exponential model and a machine-learning (linear regression) model. A golf ball was dropped from a height of 1.0 m onto wooden, carpeted, and cardboard surfaces, and rebound heights were measured experimentally. The coefficient of restitution and corresponding energy losses were calculated for each surface. A physics-based exponential decay model, assuming a constant coefficient of restitution, was used to predict the rebound heights, while a linear regression model was trained on experimental data which was collected. The performance of both models was evaluated using Mean Absolute Error (MAE) on a shared test dataset. The physics-based model achieved a lower MAE (5.38 cm) than the machine-learning model (7.08 cm), indicating a higher predictive accuracy for the physics model. However, increasing deviations were observed for softer surfaces and at higher bounce numbers, where real behaviour deviates from ideal conditions. These findings demonstrate that physics-based models perform best under ideal assumptions, while machine-learning models can adapt better to non-ideal situations. Both approaches have their own strengths and limitations, but when combined, they can provide a more robust framework for modelling complex real-world physical systems.