Predicting Impact Loads on Polymer Materials from Laser-Induced Cavitation Bubble Collapse Using Random Forest Regression

Cavitation bubble collapse generates impact loads that can initiate surface damage and accelerate material degradation in hydraulic and fluid-handling systems. Predicting these loads remains challenging because the collapse response depends on nonlinear interactions among bubble dynamics, stand-off distance, and polymer mechanical, acoustic, and viscoelastic properties. This study developed machine-learning regression models to predict impact loads from laser-induced single-bubble collapse on polyethylene, polytetrafluoroethylene (PTFE), polyamide, and antistatic polyethylene terephthalate (antistatic PET). Experiments were performed using a pulsed Nd:YAG laser at energies of 12.5, 25, and 50mJ and normalized stand-off distances of γ=1–5, producing 180 measurements. Impact loads were measured using a force-calibrated PVDF sensor with material-specific calibration equations. Bubble radius was obtained from high-speed imaging, whereas collapse time was determined from the PVDF voltage-time response. Thirteen input variables were used to train linear, ridge, LASSO, multilayer perceptron, and random forest regressors with Bayesian optimization and grouped five-fold cross-validation. Random forest regression achieved the best performance, with RMSE=0.6820N, MAE=0.5339N, R2=0.9168, and MAPE=12.44%. SHAP analysis identified collapse time, acoustic impedance, and loss modulus as dominant predictors. The model is therefore suitable for trend prediction within the tested experimental domain.