Dynamic modeling of pneumatic muscles using modified fuzzy inference mechanism

Pneumatic muscle actuators (PMA), owing to their obvious advantages over conventional linear actuators and pneumatic cylinders, have been recently used in the medical and industrial robotic applications. However, their potential has not been fully exploited due to their highly nonlinear and time dependent behavior. An attempt is being made in the proposed work to accurately predict the uncertain and ambiguous characteristics of PMA. It was revealed from a scrupulous review of the previous work that conventional tools such as analytical and numerical methods can model a nonlinear system but the time dependent behavior cannot be accurately modeled. In the present research, Artificial Intelligence (AI) based techniques such as Neural Network (NN) and Fuzzy Inference System (FIS) have been used and their results are analyzed. It was found that FIS based on Takagi-Sugeno-Kang inference mechanism provides better accuracy and can model the time dependency of PMA. However, to achieve higher accuracy from the Fuzzy model, its parameters are required to be optimized. Three different approaches, namely, gradient descent method (GD), genetic algorithms (GA) and Modified Genetic Algorithm (MGA) have been used to identify the fuzzy parameters. Results clearly illustrate the improved prediction performance of the MGA based fuzzy inference system. Compared to the previous research in dynamic modeling of PMA, the proposed fuzzy inference system is found to provide better prediction accuracy.