Sensorless Position Tracking Control of Shape Memory Alloy Actuator based on LSTM-Model Predictive Control

This paper presents a sensorless position tracking control strategy for shape memory alloy actuators (SMAA) based on long short-term memory model predictive control (LSTM - MPC). A low-complexity resistance-based self-sensing model is developed using a second-order polynomial to enable accurate position estimation without external sensors. An LSTM neural network is trained to capture the nonlinear and hysteretic dynamics of the SMAA and is embedded as the prediction model within an MPC framework. Particle swarm optimization is employed for online rolling optimization. Experimental results show that, compared with a conventional PID controller, the proposed LSTM - MPC approach significantly reduces overshoot and steady-state tracking error in both step and sinusoidal tracking tasks, while maintaining acceptable response speed.