Artificial neural network for performance modelling of shape memory alloy

Abstract In recent years, significant strides in technological advancement have revolutionized our lifestyles, driving a surge in demand for multifunctional and intelligent materials. Among these materials, Shape Memory Alloy (SMA) stands out for its unique ability to memorize and revert to its original shape through phase transformation. Despite its remarkable properties, SMAs exhibit a minor limitation in accurately retaining their original shape or length. Furthermore, there is a notable dearth of information regarding the modelling of SMA behaviour with high precision. This study endeavors to address these challenges by integrating experimental data with neural network modelling to comprehensively examine SMA behaviour for mechanical applications. Leveraging an experimental dataset, this research employs feedforward backpropagation neural network (BPNN) modelling to forecast the strain recovery of SMA Nitinol alloy. The model aims to predict the recovery strain of SMA by utilizing three input parameters: temperature conditional, number of coils, and initial length. Remarkably, the achieved error rates of 0.29%, 0.80%, and 9.20% for various strain measurements significantly undercut the commonly accepted error threshold of 10% for nonlinear data predictions in SMA behaviour. The final results underscore the high prediction accuracy of the Artificial Neural Network (ANN), offering promising prospects for SMA applications involving temperature-strain interactions and enhancing engineering design.