Data-Based Modeling of Twisted Coiled Actuators

Twisted Coiled Actuators (TCAs) have emerged as a promising option for developing artificial muscles with high power-to-weight ratios and inexpensive fabrication. However, their nonlinear time-variant dynamics, hysteresis, and unmodeled parameters pose challenges in their control, which limits their use in designing wearable exoskeleton devices. To overcome these challenges, this study developed and validated a data-based model, the Nonlinear AutoRegressive eXogenous (NARX) neural network, for predicting the nonlinear dynamic behaviours of TCAs cooled by forced air. The model considers the characterization of both heating and cooling systems on the actuation of TCAs. The NARX model was trained and validated using an experimental setup that adopts a novel cooling apparatus for implementation in soft wearable devices. The precision of the NARX model resulted in R2 values of 0.99 and 0.81 for training and testing, respectively. The results demonstrate the potential of the developed model for improving the control systems of wearable soft robotic devices using TCAs.