Investigation on regression model for the force of small punch test using machine learning

Abstract

The analytical solution for the force-deflection curve of the small punch test (SPT) is still a challenge. Machine learning is employed in the present study to establish a model for estimating SPT forces by using the strength of a material. The yield strength and ultimate tensile strength of materials and corresponding SPT force-deflection curves generated by finite element simulations are the training and testing data for machine learning. Pearson correlation analysis was performed first to measure the statistical association between the SPT force at a given deflection and the strength of materials. It was shown that a binary linear regression model was capable of correlating SPT forces to the yield strength and ultimate tensile strength of a material. The model parameters were determined after training, and then this model was tested by testing data. The accuracy of the regression model was verified by hypothetical materials, X70 steel and Incoloy 800H alloy. This study gives a new insight into the relationship between force responses of SPT and material properties, and can promote the development of novel approaches for determining material tensile properties using SPT.