Estimation and optimization of reinforcement parameters for composite material using a machine learning approach

The mechanical properties of Metal Powder Reinforced Polymer Matrix (MPRPM) composite materials are significantly influenced by various reinforcement parameters, such as the particle size of the reinforcement material and the loading weight% of the reinforcement metal powder in the matrix. In this study, composite samples were prepared by reinforcing metal powder into a polyethylene terephthalate (PET) matrix, followed by mechanical testing to evaluate tensile strength, flexural strength, and percentage elongation. A machine learning-based predictive models utilizing the Random Forest (RF) and Extreme Gradient Boosting (XGBoost) algorithm were developed to estimate the effects of the particle size of the reinforcement and metal powder loading by weight% on these mechanical properties. The model demonstrated outstanding accuracy with $$\left( {R^{2} } \right)$$ values of 0.3851, 0.3526, and 0.1639 for tensile strength, flexural strength, and percentage elongation, respectively, for RF model and of 0.8785, 0.6786, and 0.6360 respectively for XGBoost model during testing. Experimental results show good values for MSE for RF model and XGBoost model both, for tensile strength (0.5025 and 0.1144 respectively), flexural strength (0.0918 and 0.0462 respectively) and elongation (0.3572 and 0.1268 respectively). Root Mean Squared Error (RMSE) values were also minimal to the value of 0.7089, 0.3030 and 0.5977 for RF and 0.842, 0.2149 and 0.7731 for XGBoost for tensile strength, flexural strength and percentage elongation respectively. highlighting the reliability of the model. This study establishes XGBoost as a robust and interpretable method for optimizing mechanical properties in MPRPM composites over RF after systematic implementation and study of Random Forest (RF) and Extreme Gradient Boost (XGBoost) algorithms to the obtained dataset. The results provide a foundation for enhancing material performance and optimizing reinforcement parameters, with future potential to extend predictions to other mechanical properties like compressive strength and shear strength.