Identification of thermo-mechanical behavior for AA6061 on the physical information machine learning method

This work aims to explore the potential of machine learning in conjunction with constitutive models for characterizing the rheological behavior of metallic materials under complex conditions. In this work, the constitutive model of metal material rheological behavior is given as prior knowledge to the neural network, and embedded into the loss function of the neural network in a soft constraint manner. A physical information machine learning (PIML) model is established to map the relationship between temperature, strain, and strain rate in aluminum alloys and rheological behavior. In this study, the PIML model accurately predicts the rheological behavior of AA6061-O at deformation temperatures of 20 degrees C, 100 degrees C, and 160 degrees C, and strain rates of 0.001 s-1 , 0.1 s-1 , and 10 s-1 . Compared with the modified Voce model and BP neural network model, the PIML model has the highest prediction accuracy, with an average relative error (MAPE) and correlation coefficient (R) of 0.807% and 0.9972, respectively. Furthermore, by embedding disparate constitutive models into the loss function of the PIML learning model, the predicted results of the physical information machine model obtained are consistent with the physical trend of the embedded constitutive model.It has been proven that the prediction results of the PIML model are more in line with the fundamental principle of physics. This findings provide fresh insights into accurately describing the rheological behavior of Al alloys during hot deformation.