Machine learning–based wear prediction of carbon fiber–reinforced polypropylene composites using an L25 orthogonal array

Applications of lightweight, wear-resistant polymer composites in automobile and engineering fields require an effective method for predicting and optimizing their wear performance, which will also curtail the time for designing new materials. Therefore, the present work employs L25 orthogonal array, machine learning (ML) techniques and wear mapping to study the dry sliding wear behavior of short carbon fiber–reinforced polypropylene (CF/PP) composites. Wear loss of the CF content (5–20 wt. %), applied load, sliding velocity and sliding distance were systematically investigated. The governing wear mechanisms were explained by using microstructural and worn surface analyses. The results show that the use of CF reinforcement changes the prevailing wear-mode from severe adhesive wear and plastic deformation in neat PP to stable micro-abrasive wear due to heat softening and crack initiation and propagation being prevented. Composites with 10–15 wt. % CF had the best wear resistance because the interfacial bond, matrix continuity, and reinforcing efficiency complemented the best. Of the other factors examined, Analysis of Variance revealed that wear loss was strongly impacted by all four parameters ). The most significant contribution was from load (32.88%), followed by sliding distance (15.20%), sliding velocity (21.89%), and fiber content (26.17%). Transition between mild and severe wear regimes, under different operating conditions, was identified with the use of wear maps. The developed and compared four ML algorithms are: K-Nearest Neighbors, Random Forest, Decision Tree, and Support Vector Machine (SVM). The SVM model showed the best accuracy of prediction (R² = 0.9362), which can be used to predict the wear and is used as a strong tool for the design and optimization of wear resistant CF/PP composite.