Date of Award

12-23-2014

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Systems Engineering

First Advisor

Kamran Iqbal

Abstract

Human body modeling (HBM) offers the benefit of replicating human activity into robotics and rehabilitation devices. Gait impairment is a consequence of several diseases like stroke, spinal cord injury, or Parkinson' disease. The modeling techniques developed in robotics have been used to improve gait abnormalities in walking and in building Exoskeletal robots. HBM is resource intensive as it requires setting-up laboratory facilities and using expensive instruments. Further, the data acquired is limited to the number of the participants. In addition, the accuracy of the model is important as it is related to human pathology. In this research, several Matlab based models of multibody systems were developed with symbolic variables for representation of human body mechanics. A simple procedure was proposed to replace VICON® software in determining human body kinematics based on coordinate conversion and calculation of hip joint center (HJC) using multivariable regression (MR) or neural networks (NN). Both methods improved on the calculations of HJC for available subject data (average error 7.9mm and 5.4mm) compared to 17.7mm for the VICON® software. Based on accurate estimation of HJC and a modified reference point, a new multibody planar model was proposed, which provided kinetic results close to those of a 3D model with correlation of 0.9-0.98. A novel approach based on persistently excited data generation was taken to build a repertoire of kinematic data for walking movement. The approach helped to construct a versatile HBM for gait simulation using Time-delay neural networks. A comparison of the model output with the actual subject data resulted in mean squared kinematic error of 10-3m. A NN based control structure driven by muscle activations was developed to provide joint torques to drive the HBM. This structure was successfully applied to planar and 3D models driven, respectively, by 42 and 54 muscles. The output of the 3D model was compared with actual kinematic data and resulted in maximum error for one stride of 1.57 and 2.38 deg./frame for normal and crouch walking, respectively.

Download

Share

COinS