Neural Network Based Inverse Modelling for Pneumatic Artificial Muscles

Pneumatic Artificial Muscles (PAM) are soft actuators with advantages such as high force to weight ratio, flexible structure and low cost. have inherent compliance that makes them feasible for exoskeletons rehabilitation robots. However, their nonlinear characteristics yield difficulties in modelling control actions, which is an important factor restricting use of PAM. The PAM associated nonlinearity, hysteresis, time varying characteristics, it more difficult model the dynamics operation based high-performance controllers. Although there many studies overcome issue virtual work , empirical phenomenological models, they either much complicated or very approximate ones a variable stiffness spring input-output relationship. Based on analysis well known previous modeling works our test bed, has been observed efficacy those methods limited representing physical behaviour thus still requirement simple effective models . In this work, apart from approaches, foreseen integrated response pressure input, results simultaneous muscle length change. Therefore, standard direct identification not suitable behaviour. An inverse approach proposed order utilize applications. black box implemented by Neural Network (ANN) using experimental data collected bed. According implementation results, ANN yielded satisfactory performance deducing could be solution