Research And Implementation Of Artificial Limb Grasping Force Control System Based On Auto-encoder Neural Network

In recent years,the number of limb disabled people in China is increasing.Due to its high cost,large volume and poor sensory,the traditional prosthetics on the market bring the disabilities inconvenience in their daily life.With the continuous improvement and innovation of biomedical technology and modern engineering technology,many efforts and process have been made on prosthesis systems and have gained more and more attentions.The prosthesis plays an important role in the current muscular rehabilitation medicine and neural engineering.The bionic control of prosthesis,especially using the neural signal(surface Electromyo Graphy,sEMG)to control the upper prosthesis effectively,is one of the most important fields of prosthesis technology.Although the research on the classification of sEMG patterns is relatively mature,for the functional control of prosthetic hand,it is insufficient to obtain only the motion pattern information.As far as practicality is concerned,the control of the prosthetic hand force is indispensable.The application value of prosthetic hand will be greatly improved if the stable grip of prosthetic hand can be achieved.To address this problem,in this study,a system for controlling the grasping force of prosthetic hand with bioelectrical signal is proposed to improve patient's sense of using prosthetic hand and the thus improving the quality of life.The main research contents are as follows:(1)The subject is oriented to a cost-effective prosthetic hand suitable for the needs of amputated patients.A MYO gesture control armband is used to collect sEMG signals from the upper limb.A six-axis force sensor is deployed to measure grasping force signal.Looking for ways to establish the relationship between sEMG and force,so as to achieve the purpose of controlling grip force through patient's own physiological signals.To realize stable grip,the gripping force is divided into 8 grades from 0-40 N in this design.(2)Extracting the characteristics of expressing force size information from sEMG signals.Then,auto-encoder network of deep learning is applied to learn from the unlabeled feature set to get the hidden layer parameters which can characterize the deep features of the data and generate a new feature set to reduce the dimension of the feature set and improve the calculation speed.(3)Softmax regression is used to generate a classifier by learning from the tagged new feature set generated in(2).And then BP network is applied to create a sEMGforce regression model for force prediction at different levels to improve forecast accuracy.(4)FSR force sensor attached on the index finger the prosthetic hand to the actual grip value of the prosthetic hand.Compare this value with the predicted force value to get the error and error change.Then use the error and error change as the input to the fuzzy controller.Finally,the actual control output is get to control the speed and steering of the motor and realize accurate grasping.To test the effectiveness of the scheme,15 able-bodied subjects participated in the experiments.Based on the experimental results from 15 participants,8-channel sEMG applying all four time-domain features,which contains mean absolute value(MAV),root mean square(RMS),standard deviation(SD)and waveform length(WL).With auto-encoder reduction from 32 to 8 dimensions results in the highest classification accuracy.And the average recognition rate is over 95%.On the other hand,from the statistical results of standard deviation,the between-subject variations ranges from 2.85 to 0.53%,proving that the robustness and stability of the proposed approach.The prosthetic hand control system will not only bring the gospel to the upper limb amputation patients,improve their lives,better integrate them into society,but also promote the development of rehabilitation training and better solve the problem of integration of humans and machinery.It has set off a boom in the medical device industry.