Modelling the properties of shape memory alloys using machine learning methods

In this paper, the properties of shape memory alloys (SMA), in particular nickel-titanium alloy (Nitinol), were modelled using machine learning methods. The strain of the material ε was predicted depending on the applied stress σ and the number of loading-unloading cycles N by boosted trees, random forests, support vector machines (SVM), k-nearest neighbors (KNN), and artificial neural networks (ANN) algorithms. Experimental data were used to train the models. The highest accuracy was achieved with the ANN, for which the mean absolute percentage error (MAPE) was 0.29% for the loading period and 0.38% for the unloading period. Additional model validation at 127 cycles showed an error of 0.75% for the loading period and 0.92% for the unloading period. These results confirm the high efficiency of ANNs for predicting complex nonlinear material behavior, which can significantly reduce the number of experiments required to study SMA properties.