Artificial Neural Network based dynamic modelling of indigenous pneumatic muscle actuators

Robots are increasingly becoming popular medical devices, helping surgeons and practitioners as surgical, rehabilitation or service robots. Robots have been proved commendable in working together with patients and practitioners to achieve the common goal of well-being. Apart from high power to weight ratio and accuracy, robots are expected to be safe and flexible. At The University of Auckland we had earlier developed robots for ankle joint and lower limb rehabilitation using McKibben pneumatic muscle actuators (PMA) which were safe and flexible. However, these actuators had larger response time and hysteresis apart from compromised actuation limits. As a result of our further research we have been able to develop inhouse pleated PMA (PPMA) in our laboratory which show improved response time with low hysteresis. The newly developed actuators have larger actuation as well. In order to cope with the non-linear and transient nature of these actuators, this paper further proposes a new Artificial Neural Network (ANN) based approach. To optimize ANN model parameters a hybrid approach combing back propagation (BP) algorithm with Modified Genetic Algorithm (MGA) is developed. Results show that the hybrid approach is able to model the PPMA behaviour closely.