Machine learning-based prediction of wear rate in TiB ₂ and ZrC reinforced LM13 hybrid metal matrix composites

Abstract

Accurate wear rate prediction is crucial for designing aluminium composite components for automotive applications. Most studies rely on response surface methodology (RSM), whose polynomial structure cannot adequately capture nonlinear tribological behaviour. To address this limitation, random forest (RF), artificial neural network (ANN), and support vector machine (SVM) with a radial basis function (RBF) kernel were applied to an experimental dataset of 20 trials and compared with the established RSM model. All three machine learning (ML) algorithms outperformed RSM. The ANN agreed most closely with experimental data, with a coefficient of determination ( R ² ) of 0.9861, root mean square error (RMSE) of 0.000231 g mm ⁻¹ , and mean absolute error (MAE) of 0.000187 g mm ⁻¹ , followed by SVM ( R ²  = 0.9803) and RF ( R ²  = 0.9724). RF feature importance scores confirmed the analysis of variance (ANOVA) parameter ranking from previous work, with sliding speed the most influential factor. ANN prediction errors against confirmation experiments remained around 2.5 %. ML-based wear prediction offers a more accurate and viable alternative to RSM for hybrid metal matrix composites in automotive design and service life estimation. However, the 20-run dataset limits the results to a proof of concept for this composite system rather than general-purpose wear prediction.