Study On Human-machine Interaction Control Based On Forcemyography Signal

With the economic development in China,the living condition of most people have improved dramatically in the recent years.Such development has brought about improved livelihood,especially in respect of the living quality of the elderly and the people with disability.Rehabilitation robots have the capability to restore movement impairments in patients with disorders.Therefore,the development of human-machine interaction technology driven rehabilitation robots is particularly critical.There are two main types of methods for detecting the human motion intention in the current humanmachine interaction control method.The first is based on detecting the motion intention from electrophysiological signals such as electroencephalograph(EEG)and electromyogram(EMG)while the second uses of methods based on pressure,sound wave,and light among others to measure the changes in certain parts of human body in the course of movement which mirror the motion intention.This thesis mainly focus on upper limb motion intention detection method based on Forcemyography(FMG).Unlike the most commonly used EMG signals,FMG signals are not interfered by factors such as signal degeneration,environmental noise,and impedance changes,which makes FMG an ideal signal for human-machine interaction control.A four-channel FMG signal acquisition system is developed in this thesis.In the pre-experiments,factors such as signal hysteresis,calibration and digital filtering were considered to improve the quality of FMG signals.During the experiments,six healthy subjects performed seven different upper limb movements(hand close,hand open,wrist pronation,wrist supination,wrist extension,wrist flexion and no movement)by following the instructions on a video prompt.Accurate and timely prediction of upper limb movement intentions is the basis of human-machine interaction.In order to verify the feasibility of motion classification based on the acquired FMG signals,this thesis adopts a pattern recognition classifier based on linear discriminant analysis(LDA)algorithm to decode the FMG signals.The study shows that using four channels of FMG data and extracting only one feature of mean absolute value(MAV)in the data,the trained classifier can achieve an average classification accuracy of 91.62%.The classification accuracy of hand close,hand open,wrist pronation,wrist supination,wrist extension,wrist flexion and no movement are 89.93%,89.10%,90.85%,90.21%,89.44%,90.82% and 100%,respectively.In addition,this thesis also discusses a motion onset detection method based on FMG signals.The study finds that by extracting special time-domain features or transforms,the motion onset points have a significant amplitude difference with the motion stationary phase.Therefore,by setting the threshold value,we can approximately evaluate the motion onset points.Sampling data after the motion onset points will potentially improve the classification accuracy of the classifier of the control system.Moreover,further study was conducted to determine the influence of channels and features on the classification accuracy.And the obtained result shows that the number of FMG signal channels has a significant impact on the accuracy of upper extremity movement recognition.With the increase in the number of channels,the accuracy of the motion classification is significantly improved.In addition,the result of different channel combinations shows that the channel combinations have influence on motion classification.An investigation on optimal feature set with respect to classification accuracy was also carried out.And as the number of features increases,the accuracy also increases corresponding.However,the result of ANOVA shows that the effect is not significant.This thesis also discusses the effect of different feature combinations on the motion classification accuracy.It is found that the MAV feature has the highest motion classification accuracy of 91.62% and the summation(SUM)feature has the highest frequency in the three optimal feature combinations of each group.In summary,this thesis studied the application of FMG signal on human-machine interaction control.The research results indicate that the human-machine interaction control system based on FMG signal is promising.This technology may help improve the robustness of the control system,and optimize the interactive control performance of the current rehabilitation robots.