Studying and Predicting the Wear Performance of Destabilized High-Cr WIs Using Machine Learning and Artificial Intelligence Techniques

The current work aims to study and prediction of wear performance of destabilized high-Cr WIs using machine learning and artificial intelligence techniques. High Cr-WI alloys with various compositions and Cr/C ratios were tested under various sliding distances of 350, 700, and 1000 m at 20, 40, and 60N loads. Experimental results showed that, high-Cr WI with a higher ratio of Cr/C demonstrated the lower wear resistance at all sliding distances and applied loads in both the as cast and destabilized situations. Moreover, destabilization heat treatment of HCWCI alloys exhibited substantial variations in abrasive wear performance of these iron alloys, these may be due to the changes of the microstructure and the formation of hard-martensitic matrix enclosed by M ₇ C ₃ carbides network of as explained by SEM. The alloys with hard-martensitic matrix structure gave the better abrasive wear performance than the alloys with both austenitic and pearlitic matrices. Subcritical heat treatments at a temperature below 500 ^o C on the destabilized alloys gave little changes of wear resistance without any indications to secondary hardening. The wear resistance deteriorated significantly after tempering above 500 ^o C due to the formation of ferrite/carbide aggregates as a result of martensite decomposition. In addition, the wear behavior of destabilized high-Cr WIs results is used to identify the wear behavior of the destabilized high-Cr WIs using optimal machine learning regression (OMLR) methods. OMLR methods are implemented and performed using MATLAB/software. The OMLR methods utilize the input parameters of the destabilized high-Cr WIs, including sliding distance and load, as well as the weight loss due to abrasive wear, to build their optimal models. OMLR methods particularly predict outcomes with low errors, specifically the Ensemble regression (EN) method. The proposed EN method results were compared to those attained from the other OMLR models with the efficacy of the EN model.