Predicting specific wear rate of laser powder bed fusion AlSi10Mg parts at elevated temperatures using machine learning regression algorithm: Unveiling of microstructural morphology analysis

• Published in: Journal of materials research and technology 2024-11, Vol.33, p.3684-3695
• Main Authors: Jatti, Vijaykumar S., Murali Krishnan, R., Saiyathibrahim, A., Preethi, V., Priyadharshini G, Suganya, Kumar, Abhinav, Sharma, Shubham, Islam, Saiful, Kozak, Dražan, Lozanovic, Jasmina
• Format: Article
• Language: English
• Subjects: Gaussian process regression algorithm; Laser powder bed fusion; Machine learning regression algorithm; MAE; Prediction; RMSE; Specific wear rate
• ISSN: 2238-7854
• DOI: 10.1016/j.jmrt.2024.09.244

Precisely predicting the Specific Wear Rate (SWR) of AlSi10Mg components produced using Laser Powder Bed Fusion (LPBF) at high temperatures, which is an essential concern in additive manufacturing. This study aims to address the gap in literature by developing accurate predictive models for SWR via machine learning regression techniques. Experiments using a dry sliding wear rig shown that wear loss in AlSi10Mg print specimens increased with temperature and load, reaching a maximum wear rate of 1.5444E-06 mm3/Nm at 200 °C, 10 N, and 1.2 m/s, compared to the minimum wear rate of 6.4672E-07 mm3/Nm seen at 100 °C, 20 N, and 1.4 m/s. However, to accurately predict the wear rate at high temperatures, six different machine learning regression algorithms were used, namely Support Vector Machine (SVM), Linear Regression (LR), Random Forest Regression (RFR), Gaussian Process Regression (GPR), XGBoost regression (XGB) and Decision Tree (DT). R-squared values and various error functions were employed to validate these strategies against the anticipated outcomes. Within this set of models, GPR model has a lower Mean Absolute Error of 0.3177, Root Mean Square Error of 0.6704 and higher R2 value of 0.9686, resulting a prediction accuracy of 96.86%. From these findings, it is suggested that GPR is a very useful model for predicting the rate at which LPBFed AlSi10Mg printed parts wear under high temperature conditions in comparison with other developed models. These machine learning methods are anticipated to be beneficial for additive manufacturing enterprises.