Machine Learning Model based Prediction of Spall Strength of Metals and Alloys

A machine-learning (ML) regression model, developed using established mechanical properties related to compressibility and strength, is used to predict the spall strength of metals and alloys, reducing the need for extensive experimentation. The ML model trained on a dataset (from plate-impact spall experiments) consisting of over 70 metals and alloys and validated using spall strength data for several recently published high-entropy alloys, demonstrates its ability to more accurately predict their spall strengths, than those from Grady’s energy balance model [Grady, J Mech Phys Solids. 1988, 36(3), 353-384]. The results indicate that tensile yield strength and bulk modulus are critical property features for predicting spall strength, with their higher values correlating with increased spall strength. The study also offers insights on observable trends for spall strengths of specific material categories such as variations for pure metals with FCC versus BCC structures, or consistently low values of spall strengths of Al and its alloys, or spall strength increase with increasing complexity of iron and steel alloys compared to pure iron or plain carbon steels. The predictive guidelines enable efficient screening of metals and alloys, based on commonly available mechanical properties, though further experimental validation and expansion of the dataset is needed to enhance the ML model's robustness and applicability.