An adaptive monitoring method for SMA wires by integrating dual resistance signals and machine learning

Abstract

The state monitoring of shape memory alloy (SMA) wires is crucial for enhancing their intelligent actuation and sensing capabilities. However, existing methods face challenges such as temperature measurement difficulties, limited accuracy, and poor environmental adaptability. The dual-resistance monitoring approach, which integrates the electrical resistance signals of the SMA wire and the auxiliary temperature-sensing wire through a constitutive model, addresses these issues but remains constrained by strong parameter dependency, complex modeling, and limited predictive accuracy. This study proposes an innovative machine learning-based dual-resistance monitoring method, directly predicting the state of the SMA wire using neural networks without traditional equation-based modeling. Four neural network architectures are designed and compared to address key challenges, including long-term dependency modeling, bidirectional information capture, local feature extraction, and global attention allocation. Experimental results demonstrate that the proposed convolutional neural networks & bidirectional long short-term memory networks & self-attention mechanisms (CBLS) model achieves the best performance, with an average stress prediction error of 1.13% and an average strain prediction error of 0.46%. Moreover, it exhibits excellent robustness and adaptability under low strain rates and complex environmental conditions. This method provides a novel intelligent solution for high-precision, adaptive SMA wire monitoring, potentially accelerating its engineering applications and expanding its use in smart structures and advanced manufacturing.