Research On Emg-based Grasp Force Control Of The Anthropomorphic Prosthetic Hand

Multi-DOF myoelectric prosthetic hand is a kind of bionic device which ismainly for the use of amputees, it consists many subjects, such as biomedicine,robotics, computer science and control theory. Its development tends to beanthropomorphic, intelligent and intuitional. However, both the prosthesis andcontrol methods are far away from human hand, thus was postponed to practicalusage. This paper focuses on the technical that are relative to the design of multi-DOF prosthetic hand, aiming at improving the cosmetics, the kinematics andfunctionality, hoping to find an initiate way to improve the controllability. Themain research subjects are: design of a five-DOF anthropomorphic hand, analysisof the kinematics, statics and dynamics of the coupling linkages mechanism, multi-channel EMG-based EMG-force prediction, and force tracking impedance control.This paper made detailed comparison about the status in quo of prosthetichand from overseas and domestic, the comparison is dedicated to the key problemsof prosthetic hand design, such as driving, transmitting, sensing, and EMG-basedcontrol methods. With this condition, considering the daily requirements ofhandicapped, a novel5-dof prosthetic hand was designed using biomechatronicsapproach. The hand is a five-fingered independent driven one; it hasanthropometric cosmetics and can mimic the motion of human fingers. It integratesdriving, transmitting, control system and torque/position sensing abilities into aspace that’s the same size as human hand. Specifically, in light of human palm, anovel arched palm was designed through special displacement of each finger; inorder to mimic the motion of human finger, planar coupling linkages mechanismwas employed, especially for the spatial cone-surface-like movement of thumb, anextraordinary configuration was designed through simulation and maximizeintersection volumn methods, and a spatial linkages mechanism was applied. Thesensing and control module was integrated into the hand seamlessly. Finally ahuman-like glove was designed to improve the functionality and the cosmetics ofthe prosthesis.The mathematic model of the prosthesis is essential to control algorithmsdesign. Therefore, this paper first analyzed the kinetics, statics and dynamics ofthe hand. Unlike the methods used in the series robotics, the Assur group theorywas applied to analysis the kinetics of the coupling linkages mechanism. In orderto analyze the statics of the hand, the kinetostatics method was applied. Based on the above analysis, the virtue spring approach combining Lagrange method wasapplied to the dynamic analysis, thus a competly decoupled statical model can beachieved. The above analyses were carried out in order. The desired objectiveswere validated. All these analyses were the fundamentals of control algorithmdesign.The functionality of the hand depends on the control of single finger to acertain extent, thus the force-tracking impedance control method was applied byconsidering the contact flexibility and the force-tracking precision. In order toimprove the force-tracking precision, the indirect adaptive algorithm was appliedto estimate the parameters of the environment. Based on the sensors of the finger,a fuzzy PD algorithm was utilized to improve the accuracy of the position control.The generalized momentum based disturbance observer was applied to estimate thecontact force from the torque sensor. By doing so the quality of the force signalwas improved. According to the experiments of contacting different stiffness andvariable location of the environment, the impedance algorithm can track thedesired contact force with a relativly high accuray.As a substitution of human hand, the control methods should be designed inaccordance to human cognitive. This paper proposed a multi-channel EMG-basedgrasp mode and force extraction method, so more optional signals for prosthesiscontrol can be provided. The force prediction method was studied mainly. Thesupport vector regression method was applied to establish the force predictionmodel. The multi-variable parameters optimization algorithm was utilized tofurther improve the force prediction accuracy and speed, a decentralized Kalmanfilter was utilized to decrease the effect of signal distortion and improve the safetyof the system. Based on the above research, the static method was utilized torealize force distribution. In order to find the grasp point, a skelecton algoritm wasutilized. Experiment results of three fingers grasping validated that the grasp forcecan be regulated dynamically according to the predicted force.