Please use this identifier to cite or link to this item:
Title: Myoelectric Control of Exoskeleton Knee
Authors: Dhindsa, Inderjeet Singh
Supervisor: Agarwal, Ravinder
Keywords: Exoskeleton;SEMG;Knee angle prediction;Muscle selection;Pattern recognition;SVM Classifier;Classification
Issue Date: 26-Aug-2020
Abstract: Exoskeleton is primarily designed to increase the physical performance of the wearer. Exoskeleton system are utilized for power-assist device, human-amplifier and rehabilitative device etc. Last three decades have witnessed significant development in the field of myoelectric control of rehabilitative and assistive devices. The major focus of the myoelectric prediction has been on predicting the motion modes for different locomotion terrain, gesture control and prediction of kinetic parameters like joint torque/force. The present research work is focused in the direction of myoelectric control of exoskeleton knee. The research work is divided in four parts namely, development of multichannel SEMG acquisition system to acquire SEMG signal from lower limb muscle while carrying out daily life activities, selection of optimal group of muscles to control the exoskeleton, predicting the knee angle information from SEMG signal of selected muscles and development of a simple prototype of knee exoskeleton. An eight-channel bioinstrumentation system based on ADS1298 was developed to record the SEMG signals from the selected lower limb muscles. Human lower limb consist of numerous muscles. One of the main considerations of the myoelectric controlled exoskeleton is the selection of muscles, which can control the exoskeleton system. Biological musculoskeletal model of human is highly redundant in nature. Apart from few exceptions, there exits more muscles than necessary to produce desired joint movement. The human lower limb has more than fifty muscles, but the human motion intent can be predicted by considering only some of these muscles. The research work proposes a principal variable based method for muscle selection. The selected group of muscles are sufficient to describe the knee movements. The force developed in the eighteen lower limb muscles while performing a sit to stand movement was considered as a base for muscle selection. The principal variable method is based on the criterion of minimum conditional covariance of the rejected variables. For a two-channel system Vastus lateralis and Rectus femoris are sufficient in describing knee flexion/extension. For a three-channel system Gluteus medius and for a four-channel system Semi- tendinosus joined the set of optimum muscles. The results were further substantiated statistically by performing a t-test on the muscle force obtained from SEMG signal of the selected muscles and that obtained from Inverse dynamic analysis of the MMHB. A new approach of predicting the joint angles discretely is developed. The complete range of motion of the knee joint was quantized into levels or classes. The instantaneous level/class of the knee angle was predicted using a classifier with features of the SEMG signals as predictor variables. The parameters of the classifiers were evaluated for different activities: Sitting down on and standing up from chair, stair assent, stair descent and level plain walking was carried out. The SEMG signal from the lower limb muscles were acquired using the developed bioinstrumentation system while performing daily life activities. The complete range of motion of the knee angle was divided into different levels. The performance of LDA, k-NN, Naive bayes and SVM classifiers for predicting the instantaneous knee angle level were evaluated. SVM with a quadratic kernel outshined others with the average classification accuracy of 92.25 ± 2.24 % for sitting down/standing up from a chair, 87.58 ± 2.30 % for walking on plain ground, 89.25 ± 2.32 % for ascending stairs and 89.91 ± 2.37 % for descending stairs. A prototype of the knee exoskeleton was designed in SolidWorks. It has a shank support segment, a thigh support segment and an actuator. The predicted knee angle was used to control the prototype offline. This work presents a successful development of a myoelectric knee angle prediction capable of controlling exoskeleton.
Appears in Collections:Doctoral Theses@EIED
Doctoral Theses@EIED

Files in This Item:
File Description SizeFormat 

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.