Application of machine learning algorithms to predict mechanical properties of aluminum alloys

Abstract

Alloys used in aerospace, automotive industries and structural engineering have good strength to specific weight ratio, corrosion resistance and workability. Mechanical properties like tensile and yield strengths, elongation and hardness are vital to ensure integrity of structures. However, conventional techniques in measuring these properties are time consuming, tedious and destructive. This paper provides
a framework in which limited experimental data & Machine Learning algorithms are employed to evaluate the important mechanical properties of aerospace grade aluminum alloys. Data set of alloy compositions as well as processing parameters were obtained through material libraries to train and validate through techniques like Random Forest Regression (RFR), Artificial Neural Networks (ANN) and Support Vector Regression. The tensile strength was predicted with the best results for R – Squared & Root Mean Squared Errors were Coefficient of Determination R²= 0.96, RMSE = 12.4 MPa using ANN and RFR were more efficient in predicting the elongation (R²>0.93). The given approach demonstrated a better performance than the ordinary regression approaches with over 20 percent improvement and less experimental work up to about 60%. This approach provides the boosted & scalable approach to greatly increase the pace of material design and mechanical characterization of applications in the aerospace industry.