Triboinformatic Machine Learning Model for Frictional Behavior and Wear Volume Loss Prediction of SS316L Alloy Clad with WC/NiCrBSi

This study uses a laser cladding approach to clad NiCrBSi and reinforced WC particles on SS316L. The wear characteristics of a laser cladded material were examined under dry sliding conditions, with sliding velocities ranging from 2.5 to 4.5 m/s, loads varying from 10 to 40 N, and compositions of WC-NiCrBSi ranging from 15 to 60%. The worn surface was analyzed through SEM analysis to reveal the wear mechanisms. At 10N load, the adhesive wear mechanism is dominant for the sample with 15 wt% WC, and an abrasive wear mechanism with a minute delamination area is observed on samples with 30 wt% WC, 45 wt% WC, and 60 wt% WC. Three ML models including RF, XGB, and KNN were trained for the prediction of tribological parameters (volume loss and COF) using experimental data. Their hyperparameters were tuned using Gridsearch CV and Randsearch CV for the optimization of hyperparameters. Upon comparison of three ML models, the XGB model outperforms other models in terms of accuracy of 97.74% for predicting volume loss. Conversely, the KNN model demonstrates a superior accuracy of 93.47% compared to other models for predicting COF. Consequently, the XGB model is selected for predicting volume loss, while the KNN model is chosen for predicting COF.