Modelling of functional properties of shape-memory alloys by machine learning methods

Abstract

Shape-memory alloys are used in various areas of science and industry due to their unique shape memory effect and superelasticity, caused by martensite and reverse transformations. In this study, it is proposed to model the functional properties of shape memory alloys, namely, the dissipated energy range, strain range and stress range using the methods of machine learning. The modeling is carried ou in the specialized data mining software environment called Orange. There were built five models for each dataset by means of method of neural networks, random forest, gradient boosting, AdaBoost and kNN. The respective regression dependencies are obtained and K fold cross-validation with K=5 is performed. The errors and coefficient for R2 determination are calculated as the results of modeling by means of the above mentioned machine learning methods for the range of dissipated energy, stresses and strains on the number of loading cycles. For each physical quantity, the best results in terms of method error are obtained for k-nearest neighbors method.