The Research On Upper Limb Rehabilitation Robot And Related Control Problem

As China is entering the aging society, the stroke incidence is increasing year by year.The reconstruction and rehabilitation of motion function for the patients after stroke alsoattract the researchers’ attention. With the interdisciplinary development and integration ofthe rehabilitation medicine, mechanical design and control theory, the theory andtechnology about rehabilitation robot have been further studied and explored.Under the support of National863project and National Natural Science Foundation,this paper developed two kinds of upper limb rehabilitation robots, which are driven bymotor and pneumatic muscle (PM), respectively. Based on the designed upper limbrehabilitation robot, with the use of Electromyography (EMG) signal, the human motionintention control based rehabilitation training is realized. In addition, based on PMmodeling, a fuzzy compensation based sliding mode control (FCBSMC) algorithm and anonlinear disturbance observer based dynamic surface control (NDOBDSC) algorithm areproposed for the high-accurate control of PM, respectively. Considering that thecomplexity of PM modeling and the dynamic model of human-upper limb rehabilitationrobot varies from individual to individual, an echo state network (ESN) based PID controlalgorithm and an ESN based single-layer neural network predictive control algorithm withparticle swarm optimization are proposed for the control of unknown model by using ESN.The main contributions focus on the following aspects:A three degrees of freedom (DOF) upper limb rehabilitation robot driven by motor isdesigned. It can achieve shoulder extension/flexion, shoulder abduction, and elbowextension/flexion. Considering that the performance of wear, flexibility, safety andcompliance, a five DOF upper limb rehabilitation robot driven by PM is developed. It canrealize shoulder extension/flexion and rotation, elbow extension/flexion,metacarpophalangeal and proximal interphalangeal joint extension/flexion. In order toreduce the complexity of the mechanical structure and control system, PM-torsion springand PM-pull spring actuators are proposed to realize the bi-directional movement for thejoints of rehabilitation robot. This robot can realize the rehabilitation training of arm andhand at the same time, as well as be used separately. This greatly extents the availablerange for clinical application, and lays a good foundation for further clinical utility.By collecting four-channel EMG from arm, the node energy is selected as the featureof motion intention, and the active motion intention based human-upper limbrehabilitation robot interactive control is realized. During the feature dimension reduction,a companion state tree (CST) algorithm is proposed to eliminate the linear correlation of node energy and avoid the singular problem of the within-class scatter matrix. Thereduced feature vector is fed into a BP neural network to recognize the motion intention,and the recognition result is satisfactory. Considering that virtual reality (VR) can makerehabilitation training entertaining, as a result, a VR game based on the motion intentionrecognition is designed to stimulate the patients’ interactive initiative and enthusiasm inthe task-oriented training. The relevant experiments are carried out by the combination ofthe upper limb rehabilitation robot, EMG recognition and virtual reality game to verify theavailability of the whole system.As an actuator, PM has strong time-varying and nonlinear characteristics, whichmake it difficult to achieve high-accurate control in practical application. Thethree-element model is introduced for PM parameter identification and modeling. On thisbasis, FCBSMC is developed for PM control. This control algorithm can effectivelyimprove the control accuracy and ensure the tracking control performance in the conditionof the existence of modeling error and external uncertain disturbances. In addition,NDOBDSC is also proposed. The simulation and practical experimental results bothdemonstrate the feasibility and effectiveness of the proposed algorithm. However, themodel control based on PM modeling is required to do the PM modeling experiment again,when PM is selected with different parameters, such as length, shape and so on.Meanwhile, the whole dynamic model of human-robot depends on the conditions ofdifferent individual. It is difficult to attain an accurate and uniform mathematical model.For above reasons, to solve the problem on the high-accurate control of PM in thepractical application of rehabilitation robot, under the condition of unknown model, thispaper proposes an echo state network (ESN) based PID control algorithm and an ESNbased single-layer neural network predictive control algorithm with particle swarmoptimization, which can both obtain good control performance.Finally, the summary of this dissertation and the future work are presented.To sum up, the design and related control problem on upper limb rehabilitation robotare investigated deeply in this dissertation. These studies provide some researchexperience on the design methods, training strategies and control algorithms for theclinic-oriented application of the upper limb rehabilitation robot. With the furtherimprovement of the research, upper limb rehabilitation robots are expected to transforminto some practical products, which can provide effective rehabilitation training for morepatients, and improve the quality of life and realize the reconstruction of motion function.