Predictive Modelling of Erosion Behaviour in Polymeric and Composite Materials Using Machine Learning

Abstract

Accurate prediction of erosion rates in polymeric and composite materials is essential for their effective design and maintenance in diverse industrial environments. This study presents a predictive modelling framework developed using the JMP Pro machine learning integrated system to estimate erosion rates of polymers and polymer composites. For better model generalisation under various conditions, a curated dataset was compiled from peer-reviewed literature, standardised, and subjected to outliers and multivariate exploratory data analysis to identify dominant variables. The model utilises key input parameters, including impact angle, impact velocity, sand content, particle size, material type, and fluid medium, to predict the erosion rate as the target output variable. Six machine learning algorithms were evaluated through a systematic model comparison process, and two were selected. Model performance was assessed using robust error metrics, and the interpretability of erosion behaviour was validated through prediction profilers and variable importance analyses. Artificial Neural Network (ANN) and Extreme Gradient Boosting (XGBoost) demonstrated the best training and validation performance based on the evaluation metrics. While both models yielded high training performance, the ANN model demonstrated superior predictive accuracy and generalisation capability across a broad range of conditions. Beyond prediction, the model outputs also showed a meaningful representation of the influence of input variables on erosion rates.