Lower limb motion recognition based on surface electromyography signals and its experimental verification on a novel multi-posture lower limb rehabilitation robots

https://doi.org/10.1016/j.compeleceng.2022.108067Get rights and content

Highlights

  • Different eigenvectors are tested respectively. Combined with different kinds of BP neural network and ELM classifiers, two motion decoders are constructed for lying and sitting postures, respectively.

  • The experimental platform of this study is a novel multi-posture lower limb rehabilitation robot, which has three postures: lying, reclining and sitting.

  • A simple and effective connection method based on label number retrieval is designed, and two different motion decoders based on lying and sitting are constructed.

  • Through real-time experiments, the effectiveness of active control based on sEMG signals is verified.

Abstract

The construction of motion decoder based on surface Electromyography (sEMG) signals is an important part in the clinical trial of rehabilitation robot. Its performance directly determines the success of clinical trial. However, feature extraction is essential in motion decoder. A single feature such as time domain and frequency domain can achieve good classification results, but it is only suitable for a single posture. Hence, for the lying and sitting postures, different feature analysis and their combination are used in this study to improve the sEMG-based lower limb motion classification performance. Nine participants in the clinical trial performed four different movements respectively. Through feature extraction and pattern recognition of sEMG, the trained motion decoders were obtained. The control commands are sent to the robot through labels retrieval to drive the lower limbs for corresponding rehabilitation training. The effectiveness of the based on sEMG signals control method is verified through real-time experimental analysis.

Introduction

The rehabilitation of patients with lower limb hemiplegia and spinal cord injury is a difficult and slow process. In addition, insufficient medical resources, combined with heavy personal financial burden and failure to undergo proper rehabilitation training and treatment, are usually the reasons why the golden period of rehabilitation is lost, resulting in slow recovery. In the past, the experience level of physical therapists was different, potentially resulting in secondary injury [1]. Rehabilitation robots have high control precision and fast response speed. Biological information, multi-sensor feedback information, virtual reality technology and human body state monitoring system are all integrated. The safety of the patient is guaranteed, the rehabilitation effect is ensured, while the interest and efforts of the patients' training are also improved. In addition, the relevant information and data, during the training process, are recorded, providing a sound basis for later queries and improvement of training schemes for faster recovery [2].

Since the 1960s, the rehabilitation robots began to develop rapidly, and have achieved good results and been received gradually in clinic. The types of a lower limb rehabilitation robot can be mainly divided into the following categories: (1) Wearable exoskeleton rehabilitation robots, such as the Hybrid Assistive Leg (HAL) series of lower limb exoskeleton rehabilitation robots [3] developed by University of Tsukuba, assisting patients in completing walking, squatting, standing, climbing stairs and other motions. The lower limb powered exoskeleton robot developed by Harbin Institute of Technology in China [4] has as many as 14 degrees of freedom, and can walk for a long time under heavy load. The overall structures are light in weight, have excellent fit property and comfortable to wear. (2) Sitting and lying lower limb rehabilitation robots, such as the Virtual Gait Rehabilitation Robot (ViGRR) [5] developed by Carlton University in Canada, with four degrees of freedom in the sagittal plane, it can be combined with virtual reality technology to complete various routine movements of the lower limbs. A new foot-plate based sitting type robot [6] developed by Indian Institute of Technology, it is equipped with passive series orthosis, which can carry out targeted rehabilitation training for lower limb joints. (3) Suspension type lower limb rehabilitation robots, such as the Lokomat robot [7], developed in Switzerland, which can correct the patient's gait, exercise the muscles of the lower limbs and gradually help recover the nervous system. However, most of the above-mentioned rehabilitation robots are about passive training. The training mode is mostly single, does not support multiple postures and cannot switch the patient's postures in a variety of ways. Moreover, the hanging robot is prone to make the patient feel uncomfortable and can't carry out rehabilitation training for a long time. Therefore, these aspects need to be further improved and strengthened.

The sEMG signals of human body can feedback the activity state and movement information of muscles. Through processing and analyzing sEMG signals, the human movement intention can be obtained [8]. The main steps and sequences of the process of pattern recognition and rehabilitation training based on sEMG signals include: Original sEMG signals acquisition and preprocessing, feature extraction, pattern recognition, experimental analysis of real-time active training control for lower limb rehabilitation robots, so as to complete a whole set of main control system based on sEMG signals excitation drive exercise training. Many researchers have done a lot of research on the active training of sEMG signals rehabilitation robot, and achieved some results. Sun [9] used the fuzzy Approximate Entropy (fApEn) of the patient's sEMG signals of triceps and biceps to evaluate the rehabilitation effect of the rehabilitation training system. The complexity of sEMG signals, quantitatively analyzed by fApEn, while the maximal autonomic contraction, during elbow flexion and extension, was used as a response to muscle strength. Based on the data of eight patients, it was concluded that the mean value of sEMG signals was significantly improved, so the fApEn shows a better correlation with maximum voluntary contraction, which offered more understanding of the improvement of motor evoked function. Kim [10] used four channels to collect the human body sEMG signals, extracted absolute value of time domain (TD) feature integral as eigenvectors, and utilized the fuzzy maximum neural network classifier to recognize six kinds of human wrist actions. The pattern recognition rate of each action is more than 90%, and the average recognition rate can reach 97%. Naik [11] extracted the sEMG signals of human hand movements, and used Support Vector Machine (SVM) method to recognize human hand action modes, and obtained a high recognition rate. However, the recognition speed of SVM classifier is slow, which is not conducive to multi sample training, so further research is needed.

Many researchers have done a lot of research on the active training project of robots, based on sEMG signals stimulation. Zhang [12] used sEMG signals' torque estimation and two three layer BP neural network for modeling, to obtain the motion intention of the human lower limb, realized the estimation of the muscle torque of the hip joint and knee joint of the patient's lower limb, while the lower limb exoskeleton robot was instructed to drive the lower limb for active training. Qiu [13] synchronously collected the sEMG signals of six muscles of lower limbs and the motion angles of hip and knee joints, introduced the coherence analysis method, to quantitatively describe the coupling relationship between the sEMG signals and each joint angle, while the ELM algorithm was used to analyze the motion of lower limbs. The experimental results show that the motion accuracy and real-time performance of the seven participants in the clinical trial can meet the control requirements of the rehabilitation robot. Peng [14] proposed two kinds of EMG driving models. By collecting and analyzing sEMG signals of lower limbs, the driving control of lower limb rehabilitation robot is realized. The admittance controller is used to integrate the subjective wishes of patients into the control system, which can improve the path tracking effect in the training process. Based on sEMG signals-driven active training, combined with the actual conditions of patients with hemiplegia and spinal cord injury, for patients with severe hemiplegia with very weak sEMG signals, the healthy leg can be used to control the weak leg for corresponding rehabilitation training. Using the residual sEMG signals in the lower limbs of the weak leg to design an interactive control strategy can encourage the patient to actively contract the patient's muscles and improve the effectiveness of rehabilitation training. After the acquired sEMG signals are processed, the motion intention is obtained as an instruction to drive the lower limb rehabilitation robot to perform rehabilitation training. A critical control system, the lower limb rehabilitation robot responds to instructions combined with a preset control strategy to complete rehabilitation training tasks [15,16].

In view of the research status that the adjustable postures of the existing lower limb rehabilitation robots are not enough, patients haven't more training during rehabilitation training stage. In this study, it mainly includes the following three parts of works and contributions: (1) A new type of multi-posture lower limb rehabilitation robot with three arbitrarily switchable rehabilitation training postures of lying, reclining and sitting postures is proposed, which improves the flexibility of mechanisms (Figs. 1 and 2). The length of the mechanical joint connecting rods is adjustable, which help to provide personalized rehabilitation training programs for patients with different needs and training stage, and improve rehabilitation efficiency. It improves a good experimental platform for the prototype experiment of pattern recognition based on sEMG. (2) A rehabilitation training based on sEMG signals excitation is proposed. The time domain, frequency domain, nonlinear features and their linear combination are adopted. Compared with the traditional research methods [10,12], this combination method fully excavates the results of the internal interaction between different data and improves the accuracy of pattern recognition. It can identify the patient's movement intention in real-time and accurately. (3) Prototype test design and result analysis: In this study, a simple and effective connection method based on label number retrieval is designed, and two different motion decoders based on lying and sitting are constructed. Compared with the admittance control in Ref. [14], the operation is more convenient and effective and the motion intention of patients' lower limbs can be obtained accurately and quickly. The real-time response result of the rehabilitation robot based on sEMG pattern recognition result is improved. This work effectively applied human sEMG signals from theoretical research to engineering practice, which is of great significance.

The rest of this paper is organized as follows: In Section 2, the main mechanisms of the multi-posture lower limb rehabilitation robot and the functions and movement modes of each main part are presented. In Section 3 methods of feature extraction and pattern recognition are demonstrated. Section 4 are the content of the construction of motion decoder, the experimental principles and processes, while analyzes the control effect and experimental results in detail. Conclusions of this study and future works are drawn in Section 5.

Section snippets

Mechanisms of multi-posture lower limb rehabilitation robot

The multi-posture lower limb rehabilitation robot is mainly composed of a bed support mechanism and an exoskeleton mechanism for the hip, knee and ankle joints. The three dimensional mechanism model is shown in Fig. 1. The main application field of the robot is the early and mid-stage rehabilitation training of patients in this study. For the early stage of training, it's all passive training, including hip joints training in lying posture, knee and ankle joints training in sitting posture. The

Research on control strategy based on lower limb sEMG signals excitation

During rehabilitation training, mode diversity and system stability are essential for the safety of patients. In this study, the schematic diagram of software and hardware control system and the active control strategy, based on lower limbs sEMG signals, are developed, as shown in Fig. 3.

How to extract useful features from sEMG signals is the crucial process of pattern recognition. Whether the extracted eigenvalues are representative and reflect the internal relationship between sEMG signals

Experimental analysis of rehabilitation training based on sEMG signals excitation

The original sEMG signals of the lower limb were preprocessed, using signal amplification, filtering and analog-to-digital conversion, in order to derive the useful initial sEMG signals. The combined (CB) eigenvectors are obtained by combining and arranging TD, FD and NL in order, namely, [CB]=[TD  FD  NL]. The eigenvectors of different actions are divided into combinations and randomly mixed and scrambled. Next, the training and testing sets are formed and input into the BP neural network and

Conclusions and furure works

In the field of rehabilitation robots, the control strategy based on sEMG lower limb pattern recognition has attracted more and more attention. The experimental platform of this study is a novel multi-posture lower limb rehabilitation robot, which has three postures: lying, reclining and sitting. It can verify the lying and sitting posture experiments respectively, and realize multi-joint linkage at the same time. To improve the classification performance of lower limb, four kinds of

CRediT authorship contribution statement

Bingzhu Wang: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Visualization, Writing – original draft, Writing – review & editing. Changwei Ou: Data curation. Nenggang Xie: Supervision, Funding acquisition, Project administration, Writing – review & editing. Lu Wang: Supervision, Project administration. Tiantang Yu: Supervision. Guanghui Fan: Data curation. Jifa Chu: Data curation.

Declaration of Competing Interest

We would like to submit the enclosed manuscript entitled “Lower limb motion recognition based on surface Electromyography signals and its experimental verification on a novel multi-posture lower limb rehabilitation robot”, which we wish to be considered for publication in “Computers and Electrical Engineering”. No conflict of interest exits in the submission of this manuscript, and manuscript is approved by all authors for publication. I would like to declare on behalf of my co-authors that the

Acknowledgement

The authors would like to thank the volunteers who helped in participating in the experiment and recording the data. This work were supported by the Science and Technology Major Project of Anhui Province, China (Grant No. 17030901037) and the University Synergy Innovation Program of Anhui Province, China (Grant No. GXXT-2021–044), and Scientific Research Foundation of Education Department of Anhui Province, China (Grant No. KJ2019ZD09).

Bingzhu Wang is a Ph.D. candidate at the Mechanics at College of Mechanics and Materials, Hohai University. His-research interests include mechanical modeling of soft robot and structural design and electrical control of rehabilitation robot.

References (23)

  • H. Xu et al.

    Advances and disturbances in sEMG-based intentions and movements recognition: a review

    IEEE Sens J

    (2021)
  • Cited by (0)

    Bingzhu Wang is a Ph.D. candidate at the Mechanics at College of Mechanics and Materials, Hohai University. His-research interests include mechanical modeling of soft robot and structural design and electrical control of rehabilitation robot.

    Changwei Ou is currently pursuing the M.S. degree in College of Mechanical EngineeringAnhui University of Technology. Her research interests include wireless biomedical signal processing, and sEMG signals control.

    Nenggang Xie received the Ph.D. degree in engineering from Hohai University (1999). He is currently a professor and doctoral supervisor of the College of Management Science and Engineering, Anhui University of Technology. His-research interests include engineering mechanics, game theory and rehabilitation robot.

    Lu Wang graduated from Anhui University of Technology. He is currently an Professor of the Department of Asset Management, Anhui University of Technology. His-research interests include modern design methods and simulations.

    Tiantang Yu received the Ph.D. degree in engineering from Hohai University (2000). He is currently a professor and doctoral supervisor of the College of Mechanics and Materials, Hohai University. Her research is focused on advanced numerical methods and damage and fracture mechanics.

    Guanghui Fan has a M.S. degree in College of Mechanical Engineering, Anhui University of Technology (2020). His-main research interests are pattern recognition and intelligent algorithm.

    Jifa Chu has a M.S. degree in College of Mechanical Engineering, Anhui University of Technology (2020). His-research interest is robot structure design.

    This paper is accepted by Associate Editor Dr. Mohammad Mehedi Hassan.

    View full text