Research On Soft Upper Limb Exoskeleton And Patient-Dominiated Brain-Robot Cooperative Rehabilitation Control

There are more than two million new stroke patients in China every year,and the trend is increasing year by year.About three-quarters of stroke survivors have different degrees of dysfunction,such as movement control disorder,muscle strength decline,endurance decline,muscle spasm,balance and coordination ability decline,which seriously affect their quality of daily life.According to the principle of brain plasticity,through training of specific tasks,patients can store correct movement patterns in the reconstituted cerebral cortex via sufficient repetitive activities.Rehabilitation robots can realize highly difficult,targeted and repeated rehabilitation training,which is of great significance to improve rehabilitation efficiency,ensure rehabilitation quality and reduce labor intensity.At the same time,the brain computer interface does not depend on the communication system of peripheral nerve and muscle pathways.It can control the external equipments through EEG signals to realize the direct movement of patients’ limbs,and effectively improve their limb motor function and nerve rehabilitation effect.Relying on the relevant national natural science foundation projects,this paper focuses on research on the upper limb exoskeleton robot driven by pneumatic muscle flexibility and its motion control methods.The research starts from the multimodal brain computer interface paradigm design and EEG signal analysis and processing,and in order to obtain the patient’s motion imagination intention and stimulate the robot to assist the patient in training tasks.At the same time,the brain fatigue feedback and body fatigue state during the training of patients are monitored in real time.Based on the fatigue status,the control mode and task of robot assisted rehabilitation training are intelligently adjusted.Through the patient-dominiated brain computer cooperation and optimization control,the patients’ nerve system can actively participate in the training process.The main research work of this paper includes:(1)Pneumatic muscles-driven modular upper limb rehabilitation exoskeleton and motion control.Combined with the physiological movement characteristics of upper limb and the requirements of assistance rehabilitation using flexible pneumatic muscles,and the multi-joint flexible exoskeleton mechanism that meets the assiatance requirements of upper limb joints is studied,and the flexible joint actuator and transmission mechanism are optimized based on the motion mechanism of the upper limbs;for the pneumatic muscle-driven modular upper limb exoskeleton rehabilitation robot mechanism,the exoskeleton rehabilitation robot platform and software and hardware system are established;a multi-input single-output high-order model-free iterative learning control algorithm is proposed through the multi-cycle error signals learning,the exoskeleton is controlled accurately without an accurate model to realize the high-performance and precise tracking for the predetermined training trajectory;experiments carried out on patients show that this method has superior robustness,laying the foundation for patient-led robot-assisted rehabilitation.(2)Patient intent recognition based on a multimodal brain-computer interface.To achieve patient control of the upper extremity exoskeleton,the steady state visual evoked potential(SSVEP)is used to start and terminate the exoskeleton movement,and the motor imagery(MI)is used to control the elbow/wrist joint.The frequency characteristic and phase characteristic of SSVEP signal are considered simultaneously to enhance the efficiency of feature identification.The extracted features are classified and identified through a convolutional neural network.Experiments are carried out on public datasets and laboratory-made datasets,and the classification accuracy of more than 90% can be obtained on multiple subjects,which verifies that this method can improve the generalization ability of SSVEP signal processing.Considering the time domain,spatial domain,and frequency domain features of the MI signal,and a regularization loss function is added in EEGNet to avoid the overfitting and improve the classification accuracy of subjects with poor classification performance.At the same time,the combination of SSVEP and MI signal can effectively solve the problem of low dimension of brain-computer interface control instructions,reduce the difficulty of multi-classification,and achieve efficient upper limb exoskeleton control.(3)EEG/EMG-based brain fatigue and muscle fatigue monitoring methods.The continuous SSVEP / MI experiment leads to the fatigue of the subjects,and the fatigue state is not conducive to the rehabilitation training of the patients.Pearson correlation coefficient is used to calculate the correlation between channels in time domain,and amplitude squared coherence method is used to calculate the coherence between channels in frequency domain.Correlation features are extracted from the two domains respectively.Based on the dual graph method,the edge features of the correlation matrix are transformed into node features,and a unified adjacency matrix is obtained,which makes the graph convolution neural network better learn and characterize these edge features.Before the full connection layer,the time domain and frequency domain features learned by the graph convolution neural network are spliced and fused to achieve higher recognition accuracy than that in a single domain.The proposed method is tested on public data sets and self-made data sets in the laboratory to verify the effectiveness of the method.At the same time,based on muscle synergy theory and particle swarm optimization algorithm,the upper limb muscle activity fatigue evaluation model based on EMG signal is studied.The fatigue of body activity is analyzed based on EMG signal,and the fatigue is distinguished by support vector machine classifer based on whale optimization algorithm.According to the fatigue state of patients,rehabilitation tasks and strategies can be optimized to realize the rehabilitation training adapted to patients.(4)Adaptive control of patient-led upper limb exoskeleton rehabilitation robot.During the rehabilitation training process,the assistance motion of the exoskeleton rehabilitation robot is triggered by the decoding results of the multimodal brain-robot interface to realize the patient-led rehabilitation training.Perceiving the patient’s motion intension and fatigue state,a hierarchical rehabilitation training strategy adjustment method suitable for different patients in different rehabilitation stages is proposed,which can adjust the rehabilitation mode and training task of the current or next training,and realize customized training tasks adapted to the cognitive state of the patient.A compliance control method under muscle fatigue supervision is proposed to adjust the parameters of the impedance model based on limb fatigue information,so that subjects can freely regulate the motion of robot through their voluntary force,which can improve the enthusiasm of rehabilitation training.Brain-controlled robot assiatance rehabilitation system is developed to provide multiple patient-led training modes.Combined with the established central nervous movement process of motor imagery,repeated and progressive real movements help patients to relearn their movements and promote the remodeling of their neural circuits.