Predicting the tensile properties of heat treated and non-heat treated LPBFed AlSi10Mg alloy using machine learning regression algorithms

• Published in: PloS one Vol. 20; no. 6; p. e0324049
• Main Authors: Jatti, Vijaykumar S., Saiyathibrahim, A., Yadav, Arvind, R., Murali Krishnan, Jayaprakash, B., Kaushal, Sumit, Jatti, Vinaykumar S., Jatti, Ashwini V., Jatti, Savita V., Kumar, Abhinav, Gouadria, Soumaya, Bonyah, Ebenezer
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 02.06.2025; Public Library of Science (PLoS)
• Subjects: 3-D printers; Additive manufacturing; Algorithms; Alloys; Alloys - chemistry; Aluminum alloys; Aluminum base alloys; Analysis; Aviation; Biology and Life Sciences; CAD; Composition; Computer aided design; Computer and Information Sciences; Decision trees; Engineering and Technology; Forecasts and trends; Gaussian process; Heat; Heat treatment; Hot Temperature; Lasers; Learning algorithms; Machine Learning; Magnesium - chemistry; Materials Testing; Mechanical properties; Methods; Neural networks; Physical Sciences; Product development; Regression; Regression analysis; Research and Analysis Methods; Root-mean-square errors; Solidification; Stainless steel; Strength of materials; Tensile properties; Tensile Strength; Thermal properties; India
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0324049

In this study, the ability of machine learning algorithms to predict tensile properties of both heat-treated and non-heat treated LPBFed AlSi10Mg alloy is investigated. The data was analyzed using various Machine Learning Regression (MLR) models such as Linear Regression (LR), Gaussian Process Regression (GPR), Random Forest Regression (RFR), and Decision Tree (DT). The AlSi10Mg alloys, heat-treated and non heat-treated, had different tensile characteristics. The tensile characteristics were forecasted using trained and evaluated MLR models. Because the performance of various MLR models has been verified by several performance indicators, such as Root Mean Square Error (RMSE), R 2 (coefficient of determination), Mean Square Error (MSE), and Mean Absolute Error (MAE). Moreover, scatter plots were made for checking the accuracy of the forecast. The GPR model demonstrated better prediction performance than the other three models, i.e., higher R 2 values and lower error values for the heat-treated samples. For predicting the UTS value of non-heat treated samples, the LR model performs very well with R 2 of 1.000. In that case, GPR has the better predictive performance for the other tensile features in non-heat treated samples. Summing up, it is obvious that GPR is well capable of predicting tensile properties of AlSi10Mg alloy with high precision. This indicates how important GPR is to additive manufacturing to achieve great quality.