Surface Electromyography-based Muscle Fatigue Prediction In Human-robot Cooperation

Human-robot cooperation is one of the development topic of the new generation of robot system.Its purpose is to reduce the burden of human and improve the efficiency of humanrobot cooperative work.How humans and robots recognize each other's state is the key to realize human-robot cooperation.In the process of human-robot cooperation,muscle fatigue of human arm affects the quality and efficiency of human-robot cooperative work,and increase the risk of skeletal muscle disease for human.Therefore,it is a great theoretical significance and application value to predict muscle fatigue in human-robot cooperation.Surface Electromyogram(SEMG),obtained from the electrodes on the surface of the muscle,is a bioelectrical signal that is recorded during the neuromuscular system activity.It is considered to be one of the most effective tools for studying muscle fatigue.However,the intrinsic relationship between muscle fatigue and SEMG is not clear.For this reason,this thesis deals with SEMG signal processing,SEMG-handgrip force model,SEMG-muscle fatigue model and human-robot cooperation.The main contents are listed as follows:(1)A frequency-band selection method based on wavelet transform is proposed to enhances the performance of grip strength estimation by SEMG collected from dynamic muscle contraction.First,Monte Carlo simulation is used to analyze the sensitivity to grip strength at different wavelet scales.Then,a suitable wavelet scale combination(WSC)is selected based on the sequence combination analysis method to determine the effective SEMG frequency-band for predicting handgrip force.Finally,the effectiveness of the proposed method is verified by the validation experiment.(2)In order to solve the low adaptability of SEMG-force model under different types of muscle contractions.A cross model selection(CMS)technology is proposed to build SEMGhandgrip force model.First,the SEMG signals under dynamic and static muscle contraction are obtained by frequency-band selection method.Next,the nine-order polynomial-based cross model is built according to the signals through regularization and cross validation.Then,a coefficient term selection method is proposed to select a suitable combination of coefficient terms for the CMS model.Finally,Finally,the effectiveness and accuracy of the proposed method are verified by the arm grip prediction experiment.(3)To solve the problem in quantification of muscle fatigue,a SEMG-muscle fatigue model based on force loss is built.First,the features of SEMG are extracted from fatigue muscle contraction.Then,the sensitivities of SEMG features in different frequency-bands to muscle fatigue are studied by Monte Carlo simulation and Pearson correlation coefficient.Finally,SEMG-muscle fatigue model based on force loss is established by sequence combination analysis method and Pearson correlation coefficient.(4)An experimental study is conducted on muscle fatigue prediction in human dynamic exercise.First,through the dynamic motion experiment of the arm,the effectiveness of the proposed SEMG-muscle fatigue model under multi-muscle coordination is verified.Then,a robot end-effector force adjustment method combined with the proposed SEMG-force and SEMG-fatigue model is proposed.The human-robot cooperation experiment system is built to verify the effectiveness of the proposed method.