Dynamics-based NARMAX/ANFIS Muscle Modeling And Fuzzy PID Control Of Functional Electrical Stimulation

In recent years, the techniques of standing up without supported tools and artificial walking-assistive rehabilitation based on functional electrical stimulation (FES) were received wide attentions which has been a hot research topic in the field of neural engineering , rehabilitation engineering and sports medicine. FES is a cross frontier technology which refers to using the surface electrode or implanted electrode to stimulate the muscles losing the neural control, cause the muscle contraction, and thus restore the function of specific parts of the human body. The precision of FES stimulus effect is mainly dependent on the reliability of muscle-stiumulated model and the effectiveness of control algorithm.In this thesis the human kinematics information were used to deduce the knee joint moment through inverse dynamics method. A muscle model was established by integrating the knee joint moment with FES stimulation level. Together with the closed-loop controller designed subsequently, this model could adjust the stimulation pattern dynamically and obtain precise control effect. In the research, 15 subjects participated in this muscle-modeling experiment, and their data were used for modeling by two kinds of nonlinear dynamic modeling methods, NARMAX (Non-linear auto-regressive moving average with exogenous input) and ANFIS (Adaptive neural-network fuzzy inference system). NARMAX model used least squares identification method based on MGS(Modified Gram-Schmidt) for the integrated identification of process model and parameters which results showed that the averaged error of model output was less than 0.2(N*M) and the error variance was less than 0.1, ANFIS model combined all the advantages of fuzzy inference and neural-network. The original model structure and parameters of fuzzy inference were adjusted through data training to achieve the best output, which results showed that the averaged error of model output was less than 0.12(N*M) and the error variance was less than 0.06. The two models of the joint moments and the stimulus current had high stability and accuracy, and ANFIS could obtain better results.In the research the fuzzy-PID controller was designed, which modulate the current intensity for FES with a higher accuracy by parameters optimization, according to the real-time error of desired knee moment and the actual output. The fuzzy-PID controller was combined with NARMAX and ANFIS muscle model separately for the feedback-control test. The absolute mean error of fuzzy PID controller tracking based on the NARMAX muscle model and the ANFIS muscle model was less than 0.3 and 0.05(N*M)separately. So the ANFIS control tracking results were much better than those of NARMAX.This research implied that a reliable muscle model would guarantee an accurate feedback control for FES system. The ANFIS muscle based on knee joint moment with stimulation level and combined with the fuzzy-PID controller obtained a perfect output through modulating the FES current intensity. So it is proved a promise approach for future practical application in the automatic FES control system and benefits for the design of updated lower limb artificial motor prosthesis system.