Remaining Useful Life Prediction Using Bayesian Additive Regression Trees

Accurate prediction of the Remaining Useful Life (RUL) is crucial for avoiding unscheduled downtime, enhancing safety, and reducing maintenance costs. Traditional methods face challenges with high-dimensional, nonlinear, and uncertain data. This paper presents a framework based on Bayesian Additive Regression Trees (BART), integrating RUL prediction with feature selection. The model is trained and tested on the NASA CMAPSS dataset, identifying key sensor features through SHAP analysis. Results show that BART can achieve accurate predictions, reasonable uncertainty estimates, and effectively identify critical variables.