Research On Muscle Fatigue Identification And Compliance Control For Rehabilitation Exoskeleton Robots

With the intensification of social aging and the increasing incidence of various accidents,the number of motor dysfunction groups such as stroke,spinal cord injury,and limb joint injury has increased year by year,bringing a heavy economic and medical burden to society.Sustained and effective rehabilitation treatment contributes to achieving brain neural plasticity and the ability to recover from daily living activities.However,China is short of rehabilitation medical resources,and there is a shortage of rehabilitation physicians.At the same time,it is limited by the personal level and experience of therapists,and the efficiency of rehabilitation training is low.Therefore,rehabilitation training based on exoskeleton robots has received much attention.By helping patients perform repetitive actions to achieve the purpose of rehabilitation training,it reduces the workload of therapists and improves the treatment efficiency of neuromuscular function recovery.The active rehabilitation training mode of rehabilitation exoskeleton robots is beneficial for inducing patients’ enthusiasm for movement,but continuous limb movements can lead to muscle fatigue.Therefore,accurately identifying the patient’s muscle fatigue status and adjusting the training mode in a timely manner is of great significance for avoiding secondary injuries caused by excessive fatigue.In response to the requirements for flexibility and safety of patients during rehabilitation training,a nonlinear dynamic system analysis method is introduced to conduct research on muscle fatigue state identification models and propose a compliance control strategy framework.The focus is on the collection and preprocessing of electromyographic signals,nonlinear feature analysis of fatigue,and high-precision identification models for muscle fatigue Research has been conducted on joint stiffness estimation and adaptive admittance controller design based on the Hill skeletal muscle model,aiming to provide new methods and research ideas for exoskeleton compliance control based on muscle fatigue identification.The main research work is as follows:(1)In order to obtain data containing muscle fatigue status information,taking knee flexion and extension exercise as an example,a weight bearing training fatigue experiment was first conducted to collect sEMG signals from six lower limb muscles of the subject.Secondly,in view of the problem that EMG signals are susceptible to noise interference during the process of muscle dynamic contraction to fatigue,and existing denoising methods are difficult to effectively filter out noise,a multiple denoising method based on empirical wavelet decomposition and improved wavelet threshold function is proposed.Finally,it is verified through experiments that the proposed method can remove noise to the maximum extent while retaining useful parts of the signal.(2)Due to the complex characteristics of sEMG signals such as non-stationary,nonlinear,and chaotic signal characteristics,the traditional time and frequency characteristics are linear analysis techniques based on the assumption of signal stability.There are limitations in the study of the nonlinear characteristics of complex transients in sEMG signals.Therefore,a method for analyzing muscle fatigue characteristics based on the multifractal descending exponential weighted average method(MFDEWA)of sEMG signals is proposed.Firstly,this method is used to perform nonlinear dynamic analysis on different types of signals to clarify the multifractal characteristics of EMG signals;Secondly,the theory of muscle synergy is introduced to use non negative matrix decomposition for muscle selection.After segmentation of active segments by envelope threshold method,three states,namely,non fatigue state,fatigue transition state,and fatigue state,are labeled to extract multiplicity features under different fatigue states;Finally,a single factor variance and correlation analysis was performed on the proposed features to determine four multifractal features.At the same time,the computational efficiency was compared with time domain,frequency domain,and real-time frequency domain features,verifying the practical feasibility of the proposed features,providing new feature references for muscle fatigue identification models and rehabilitation medicine research.(3)In order to improve the accuracy of fatigue identification,a relationship model between EMG characteristics and exercise fatigue status was constructed using a gated loop unit(GRU)neural network algorithm,and the model parameters were optimized using the Dung Beetle Optimization Algorithm(DBO).The recognition accuracy of the same classification model with single feature and fused feature sets as input and different classification models under the optimal feature set were compared.The experimental results showed that the DBO-GRU model based on fused multifractal combined feature sets can effectively improve the recognition accuracy of muscle fatigue.(4)An adaptive admittance control controller that integrates joint stiffness and fatigue information of the human body is proposed to meet the rehabilitation needs of patients for safe and compliant assistance of skeletal robots during active rehabilitation training.The human joint stiffness estimated using Hill’s skeletal muscle model and the identified fatigue state are used to adjust the parameters of the admittance controller to achieve the compliance and safety of human-computer interaction.Finally,based on Matlab/Simulink joint simulation,the feasibility and effectiveness of the proposed method were verified.