Prediction of Electrochemical Surface COF of Titanium Alloy Using an Enhanced LightGBM with Lag and Rolling Features

To achieve accurate prediction of the surface COF (coefficient of friction) of titanium alloys under electrochemical corrosion conditions, this study investigates the tribological behavior of titanium alloys across various solutions, concentrations, voltages, and sliding velocities to construct a systematic dataset. Two machine learning models are developed and optimized: a standard Light Gradient Boosting Machine (LightGBM) and an enhanced LightGBM model incorporating lag and rolling features. These models are employed to predict the friction coefficient and feature importance analysis. The results indicate that solution concentration is the primary factor influencing the friction coefficient of the titanium alloy, followed by test duration, while sliding velocity exerts the least influence. Following experimental validation and iterative optimization, the enhanced LightGBM model, integrated with lag and rolling features, demonstrates superior predictive accuracy, achieving a coefficient of determination (R2) of 0.979 on the training set and 0.951 on the test set. This research establishes a data-driven predictive framework that demonstrates superior accuracy and interpretability compared to models using only raw features, showcasing the potential of feature-engineered machine learning in optimizing electrochemical machining parameters.