Study On Motion Recognition Of Bionic Manipulator Based On Surface Electromyography

Surface electromyography(sEMG)is a series of time series collected on the human skin surface through the body surface electrodes and produced by human muscle movements.Different hand movements triggered different muscle tissue,thus producing difference electrical signals.In this paper,the recognition and control of bionic manipulator is based on the surface EMG signal,which is applied to the dangerous places Such as routine maintenance of nuclear power equipment,equipment replacement and emergency accident treatment in the field of nuclear power.The main work is as follows:(1)We firstly analyzed the mechanism and characteristics of surface EMG and determined the best position of EMG electrode,and designed 6 kinds of arm movements.Then We designed a hardware signal acquisition system including 10(25)1000Hz bandpass filter and 50 Hz FIR notch filter.Finally,Using the short-time average amplitude function to process sEMG,and testing activities segment combined with threshold comparison method.(2)Feature extraction of sEMG.We used the independent principal component(ICA)algorithm to eliminate the denoise and redundancy of the surface EMG signals and combined with unsupervised clustering algorithm(SOM)to clustere and analyze the processed surface EMG signals,which indicated the clustering graph of each action is quite different.(3)Pattern recognition of sEMG.We proposed the improved particle swarm optimization(IPSO)algorithm to optimize the support vector machines(SVM).IPSO-SVM introduces a way to simplify the position and velocity formulas of PSO,then proposes ESE state estimation for premature convergence.(4)The experiment of the motion recognition of bionic manipulator.Firstly,we analyzed the human forearm muscles and selected 3 muscles as the signal source of surface EMG signal.Then we adopted 5 test algorithms(the proposed IPSO-SVM,traditional SVM,LIBSVM,PSO-SVM,GA-SVM algorithm)to classify the six hand motion patterns recognition(fist clenching,fist unfolding,internal and external rotation,wrist intorsion and wrist extorsion),the results showed that the average accuracy rate of IPSO-SVM is 93.75%,the hardware and software of the bionic manipulator are connected together,which can accurately track the movement of the hand.Finally,we found that the arm will gradually enter the fatigue state under the condition of continuous movement,so the improved wavelet packet entropy algorithm is proposed to draw the fitting curve of muscle fatigue,and effectively weaken the fatigue effect of muscle.The results showed that the average accuracy rate of IPSO-SVM is also 93.75%.