Design Analysis and Control of Wearable Ankle Rehabilitation Robot

Abstract

While rehabilitation robots are not uncommon in the literature, they are undesirably inspired by industrial robot designs. Some of the shortcomings which are common to all these contemporary robots are, kinematic incompatibility, stiff actuation, non-backdrivability, high cost, unfriendly or intimidating appearance due to use of heavy and bulky electromagnetic actuators. Wearable robots, owing to their biologically inspired design, compliant actuation, backdrivability and safe use, are better candidates for rehabilitation robots compared to industrial robots. In recent years, wearable robots have received considerable attention and several instances such as exoskeletons, orthotics, and prosthetics have been proposed by researchers. However, there are certain challenges from the design and control perspective of wearable robots, which limit their wider implementation. Bio-inspired or biological design, kinematic compliance and holistic design optimization are the chief design issues, whereas, suitable actuation, development of appropriate physical and cognitive human-robot interaction are the essential control related concerns. Most of the skeletal joints in the human body are actuated by parallel action of a group of muscles and hence a bio-inspired wearable robot design is likely to be based on parallel mechanisms. Impending research issues associated with the use of parallel mechanism are small workspace, abundance of singularities and unavailability of forward kinematics solution. Ambulatory requirement of the wearable robots also calls for compact, light weight, and energy proficient technologies for actuators, sensors, and controllers. This thesis explores the wide-ranging potential of wearable robots in rehabilitation in the pretext of a wearable ankle rehabilitation robot. In this research, a parallel mechanism based wearable robot for ankle rehabilitation was developed to study design and control related aspects of wearable robots in general. Arrangement of actuators, in the kinematically compliant design, had been carefully selected to allow natural foot-ankle motions while keeping the ankle joint position stationary. A fuzzy based computational model was developed in this research to provide a unique solution for the forward kinematics of parallel robots. The proposed method is accurate and time efficient compared to previous methods proposed in the literature. The fast computation of forward kinematics has facilitated its online use in the controller replacing use of heavy inclinometers. A complete design analysis had been carried out by mathematically formulating important performance indices affecting robot performance in three major aspects such as, kinematic, actuation and structural aspects. Initially, a single objective optimisation approach was adopted following past practice, wherein a performance index called global condition number was optimized. Analysis of the results shows that some of the objectives were of conflicting nature and hence the single objective approach could not optimize all the performance criteria simultaneously. Subsequently, robot design optimization was carried out using existing multiobjective optimization methods, namely, preference based optimization and the evolutionary algorithm (EA) based optimization. Interestingly, these existing optimization methods were also found to be unsuccessful due to the incompatible and contradictory nature of objectives, their large number and continuous solution space. Further investigation in the EA methodology revealed fundamental shortcomings in the existing NSGA II approach. As a result of subsequent research efforts, a major breakthrough was achieved through the development of a fuzzy dominance based evolutionary optimization method to address the inadequacies of existing EA approach. Finally, the robot design optimization was carried out using newly developed fuzzy sorting genetic algorithm (FSGA) and the wearable robot was constructed using the optimized design. To improve the compliance of the wearable robot, light weight yet powerful actuators called Pneumatic muscle actuators (PMA) were used which exhibit skeletal muscle like behaviour. Construction of a dynamic model of the PMA was a difficult task owing to their non-linear and time dependent behaviour. Therefore, a Mamdani based fuzzy model was developed and optimized to accurately predict the PMA behaviour in the presence of an external force. The forward kinematics model of the robot and the dynamic model of PMA were finally incorporated in an overall fuzzy controller designed for the position control of the wearable robot. Apart from the conceptualization of a wearable ankle robot design, optimization of two variants of fuzzy inference systems namely, Takagi-Sugeno fuzzy system and Mamdani fuzzy system as well as their distinctive uses in this thesis are important contributions of the present research. The major contribution of this research lies in the development of a fuzzy dominance based evolutionary optimization method which is a strong alternate to the predominantly used evolutionary algorithm NSGA II, which has been used in diverse optimization applications over the last two decades.