Novel laser processed shape memory alloy actuator design with an embedded strain gauge sensor using dual resistance measurements. Part II: Recurrent neural network-based position and force estimation

The current paper is a continuation of Part I, performing position and force estimation on a similar monolithic shape memory alloy (SMA) actuator with two distinct phases embedded through laser processing and post-processing. The recurrent neural network-based model proposed in this work outperforms the mathematical model developed in Part I, achieving average position and force estimation accuracy of 97.5% and 95.0%, respectively, using only electrical resistance measurements across the two actuator phases. Furthermore, the model can be applied to SMAs with varying compositions and geometries. The described actuator and sensorless estimation model are widely suitable for robotics, haptics, and various other systems which involve the application of unknown or dynamic load.