Advanced Tribological Simulations and Predictive Modeling of Wear Behavior in Al5052/Cenosphere Composites Using Machine Learning
Khursheed Ahmad Sheikh, Mohammad Mohsin Khan, Sukanta Roga, Tabrez Qureshi
Abstract
Abstract The study focused on developing Al5052 composites reinforced with cenosphere particles to improve their wear resistance. The wear-rates of the test materials were measured using a pin-on-disc apparatus at room temperature, utilizing a dataset comprising 27 experimental observations. The results demonstrate that increasing the cenosphere reinforcement content effectively reduced the wear-rates. The microhardness improved from 68.5 Hv to 78.75 Hv by adding 4 wt% cenosphere particles to the Al5052 alloy. Four machine learning models—decision tree (DT), random forest (RF), support vector regression (SVR), and k-nearest neighbors (KNN)—were employed for wear-rate prediction. While the DT model achieved the highest test accuracy (R2 = 0.95), it exhibited signs of overfitting as indicated by its R2 of 1.0 on the training data. In contrast, the RF (R2 = 0.94) model provided a better balance between accuracy and generalizability, making it a more reliable choice for predictive analysis. An analysis of the importance of features was carried out to evaluate the contribution of input parameters to predict wear-rate. The results revealed that the reinforcement wt% had the most significant impact on wear-rate prediction. These findings suggest that data-driven machine learning approaches hold potential as powerful tools in tribological studies, paving the way for the emergence of tribo-informatics.