Wear Prediction of Functionally Graded Composites Using Machine Learning

Abstract

This study focuses on the production of functionally graded composites by utilizing magnesium matrix waste chips and cost-effective eggshell reinforcements through centrifugal casting. The wear behavior of the produced samples was thoroughly examined, considering a range of loads (5 N to 35 N), sliding speeds (0.5 m/s to 3.5 m/s), and sliding distances (500 m to 3500 m). The worn surfaces were carefully analyzed to gain insights into the underlying wear mechanisms. The results indicated successful eggshell particle integration in graded levels within the composite, enhancing hardness and wear resistance. In the outer zone, there was a 25.26% increase in hardness over the inner zone due to the particle gradient, with wear resistance improving by 19.8% compared to the inner zone. To predict the wear behavior, four distinct machine learning algorithms were employed, and their performance was compared using a limited dataset obtained from various test operations. The tree-based machine learning model surpassed the deep neural-based models in predicting the wear rate among the developed models. These models provide a fast and effective way to evaluate functionally graded magnesium composites reinforced with eggshell particles for specific applications, potentially decreasing the need for extensive additional tests. Notably, the LightGBM model exhibited the highest accuracy in predicting the testing set across the three zones. Finally, the study findings highlighted the viability of employing magnesium waste chips and eggshell particles in crafting functionally graded composites. This approach not only minimizes environmental impact through material repurposing but also offers a cost-effective means of utilizing these resources in creating functionally graded composites for automotive components that demand varying hardness and wear resistance properties across their surfaces, from outer to inner regions.