Plastic properties estimation of aluminum alloys using machine learning of ultrasonic and eddy current data

Abstract

In this study, a nondestructive testing (NDT) technique was developed to estimate the plastic properties of aluminum (Al) alloys using machine learning (ML) based on ultrasonic and eddy-current data. To validate the performance of the proposed technique, hundreds of Al alloys with a wide spectrum of mechanical properties were fabricated under different compositional and heat treatment conditions. From these specimens, the NDT parameters were measured and used as the input features of the ML model. The outputs estimated from this ML model were yield strength, ultimate tensile strength, and elongation. The estimation results were compared with those obtained from destructive tensile testing, which was performed after the completion of the NDT measurements. The relative error between the estimated and ground-truth values was approximately 8%. In particular, among the various NDT parameters, the ultrasonic nonlinearity and electrical conductivity are sensitive to the plastic properties. The related scientific findings are also discussed.