Elbow joint exoskeleton control system based on myoelectricity angle prediction and use method thereof

Abstract

The invention provides an elbow joint exoskeleton control system based on myoelectricity angle prediction and a use method thereof, the elbow joint exoskeleton control system comprises a dual-channel surface myoelectricity signal acquisition device and an advanced one-step prediction model OSA: the dual-channel surface myoelectricity signal acquisition device acquires myoelectricity signals and provides input data for the system; establishing an advanced one-step prediction model OSA based on a non-linear autoregressive model NARX with exogenous input, performing offline training, and determining a mapping relation between the electromyographic signals and the actions; and realizing online intention estimation and real-time control of the upper limb elbow joint exoskeleton system by using an advanced one-step prediction model OSA. By means of the technical scheme, the electromyographic signals can be detected in real time, the exoskeleton is driven to assist in completing elbow joint flexion and extension movement, neural function reconstruction and motor ability recovery are promoted, and the risk of muscle disuse atrophy is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of robot exoskeleton auxiliary equipment, and in particular to an elbow joint exoskeleton control system based on myoelectric angle prediction and a use method thereof. Background Art

[0002] Robotic exoskeleton-assisted devices have been widely used in the functional movement recovery of hemiplegic patients, providing external force support and quantitative, objective assessment methods for the rehabilitation process. By guiding patients to complete limb movements, these devices help rebuild motor skills and promote the recovery of neurological function. However, most current exoskeleton devices still use preset, fixed training modes, which often force patients to passively receive movement guidance during the rehabilitation process and lack feedback on their voluntary movement intentions. This passive training method limits the patient's subjective initiative and is not conducive to the remodeling of the nervous system and functional recovery.

[0003] Surface electromyography (surface electromyography) is a bioelectric signal that can reflect the human body's muscle activity and movement intention in real time. It has the characteristics of non-invasiveness, high sensitivity, and good time resolution. Through the real-time acquisition and decoding of surface electromyography signals, the patient's autonomous movement intention can be accurately identified and converted into control instructions to drive the exoskeleton's movements, thus realizing an "intention-driven" movement control method. This method not only enhances the patient's active participation in the rehabilitation process, but also effectively activates the brain-muscle pathway, promotes the reconstruction of the sensory-motor loop, and promotes the plastic adjustment of the neural network. Therefore, the motion control strategy based on surface electromyography is considered to be a key way to achieve collaborative interaction between rehabilitation robots and users.

[0004] Currently, a variety of exoskeleton-assisted devices have been published or put into use to address the rehabilitation needs of upper limb elbow dysfunction. These devices primarily utilize a rigid structure combined with motors or pneumatic drives to achieve elbow flexion and extension training. They utilize preset trajectory control, impedance control, or surface electromyography-based intention recognition control methods to assist patients in completing rehabilitation movements. However, existing technologies still suffer from numerous problems and drawbacks: First, most training modes are fixed, making dynamic adjustments difficult based on individual patient conditions and lacking personalization and intelligence. Second, electromyography signal recognition accuracy and stability are insufficient, making them susceptible to interference, resulting in delayed feedback on movement intent and affecting the naturalness and effectiveness of human-machine collaboration. Third, the exoskeleton's structural design is rigid and bulky, not fully aligned with the human elbow's axis of motion. This results in poor wear comfort and can easily cause discomfort or even potential injury with long-term use. Fourth, the system's operation is complex, making it difficult for patients to control independently, limiting the widespread adoption of home or remote rehabilitation. Fifth, the system's adaptive capabilities are limited when faced with abnormalities such as sudden nonlinear movements or muscle spasms, posing control instability and safety risks. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide an elbow joint exoskeleton control system based on electromyographic angle prediction and its use method, to detect its electromyographic signals in real time, drive the exoskeleton to assist in completing elbow joint flexion and extension movements, promote neural function reconstruction and motor ability recovery, and reduce the risk of muscle disuse atrophy.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an elbow exoskeleton control system based on myoelectric angle prediction, comprising a dual-channel surface myoelectric signal acquisition device and an OSA one-step-ahead prediction model:

[0007] The dual-channel surface electromyography signal acquisition device acquires electromyography signals to provide input data for the system;

[0008] Based on the nonlinear autoregressive model NARX with exogenous input, a one-step-ahead prediction model OSA is established and trained offline to determine the mapping relationship between electromyographic signals and movements.

[0009] The one-step-ahead prediction model OSA is used to achieve online intention estimation and real-time control of the upper limb elbow exoskeleton system.

[0010] In a preferred embodiment, the dual-channel surface electromyography signal acquisition device uses the ADS1298 chip and communicates with the main controller through a standard SPI four-wire communication interface. The communication interface includes an SCLK clock pin, a MISO data output pin, a MOSI data input pin, and a CS chip select pin, and cooperates with the START, PWDN, RESET, and DRDY auxiliary control pins to complete the timing control of the entire process from initialization, configuration to data acquisition.

[0011] In a preferred embodiment, during the startup process of the dual-channel surface electromyography signal acquisition device, the power-on operation is first completed by pulling the PWDN pin high, and then the RESET pin is pulled high to complete the chip reset, so that all registers are restored to the default state; then, the main controller continuously sends three-byte WREG instructions to the ADS1298 chip through the SPI bus according to specific application requirements, corresponding to the register write command and starting address, the register write quantity minus one, and the register configuration data; after the above process is completed, the START pin is pulled high to start the continuous data acquisition mode.

[0012] In a preferred embodiment, during the data acquisition phase, the chip indicates that the data is ready through a low-level pulse on the DRDY pin at a fixed rate. At this time, the main controller reads the data through the SPI interface. The data collected for each channel is a 24-bit value in the complement format. According to the data format of ADS1298, the original value range is ±(2 23-1), where 0x000000 represents the minimum value 0, 0x7FFFFF is the positive full scale, and 0x800000 to 0xFFFFFF is the negative range; in the dual-channel surface electromyography signal acquisition device, all values ​​are finally calculated through the minimum resolution and mapped to the corresponding analog signal amplitude to achieve electromyography measurement.

[0013] In a preferred embodiment, the one-step-ahead prediction model OSA in step 2 is combined with a shallow feedforward neural network, and the actual values ​​of the previous time steps are used to predict the output of the next time step; the proposed one-step-ahead prediction model OSA includes an input layer, a hidden layer, and an output layer, and the hidden layer and the output layer use tansig and purelin activation functions respectively; by transferring the prediction target from the current output to the output of the next time step; the mathematical formula of the one-step-ahead prediction model OSA is shown in formula (1):

[0014]

[0015] Among them, y k+1 represents the output of the system at time k+1, u k represents the input, f(·) is the nonlinear model function, y k is the sample value output at time k, n a and n b denote the order of the model input and model delay sequences, respectively.

[0016] In a preferred embodiment, the offline training phase in step 2 is performed using pre-recorded surface electromyography signal data; a 10-400 Hz fourth-order Butterworth bandpass filter is used to remove high-frequency and low-frequency interference, a 50 Hz notch filter is used to remove power frequency interference, and finally a full-wave rectifier is used to obtain pre-processed data; after obtaining the pre-processed data, a conjugate gradient algorithm is used to train the prediction model; a sliding window of 100 ms is used to extract the mean absolute deviation feature between the current moment and the previous moment from the dual-channel surface electromyography signal, and the actual joint angles at the current moment and the previous moment are combined as the input of the model; the model output is the predicted joint angle for the next time step.

[0017] In a preferred embodiment, in step 3, online intention estimation and real-time control are realized by the main controller; the desired joint angle is obtained by the one-step-ahead prediction model OSA, and the joint angle control is realized by the dual closed-loop servo control of the position loop and the speed loop.

[0018] In a preferred embodiment, the position mode is used to implement closed-loop control of the motor, a maximum speed value is set, and the control system uses the deviation between the desired angle and the exoskeleton motor angle for closed-loop control, converting the position error into the desired angular velocity, and then controlling the motor current through a magnetic field orientation algorithm.

[0019] The present invention also provides a method for using an elbow joint exoskeleton control system based on electromyographic angle prediction, which is characterized in that the elbow joint exoskeleton control system based on electromyographic angle prediction is adopted; step 1, using a dual-channel surface electromyographic signal acquisition device to obtain biceps and triceps signals; step 2, introducing a one-step-ahead prediction model OSA to map the dynamic relationship between electromyographic signals and elbow joint angles; step 3, through the integrated dual-channel surface electromyographic signal acquisition device and the one-step-ahead prediction model OSA, online estimation and real-time control of elbow joint movement intention are realized; actual measurement results show that the system average response time is significantly shorter than the time window in which the electromyographic signal leads the muscle contraction.

[0020] Compared with the existing technology, the present invention has the following beneficial effects: the present invention detects its electromyographic signals in real time, drives the exoskeleton to assist in completing the flexion and extension movement of the elbow joint, promotes the reconstruction of nerve function and the recovery of motor ability, and reduces the risk of muscle disuse atrophy. For the elderly or patients with muscle weakness, the equipment can be used as a functional compensatory tool to effectively compensate for insufficient muscle strength; in addition, for assembly line workers engaged in repetitive operations, the system can help share the load on the elbow joint and prevent strain injuries. In summary, this product can not only serve the patients' home rehabilitation needs, but also provide support for healthy people. It is suitable for many fields such as medical rehabilitation, elderly care assistive devices and industrial assistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A flow chart of the method of the present invention according to a preferred embodiment of the present invention;

[0022] Figure 2 This is a flow chart of myoelectric signal acquisition in a preferred embodiment of the present invention;

[0023] Figure 3 Schematic diagram of the OSA model structure of a preferred embodiment of the present invention;

[0024] Figure 4 This is a flowchart of offline training in a preferred embodiment of the present invention;

[0025] Figure 5 This is a flow chart of motor position mode control according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0027] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present application belongs.

[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form, and it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations.

[0029] refer to Figure 1-5 The purpose of the present invention is to provide an upper limb elbow joint exoskeleton rehabilitation system based on electromyographic signals, which introduces surface electromyographic signals as a control signal source and uses motors to drive the upper limb exoskeleton to assist in training.

[0030] The proposed system can be implemented through three steps: First, a dual-channel surface electromyography (EMG) signal acquisition device is designed and implemented to provide input data for the system. Second, a one-step-ahead nonlinear autoregressive model is built and trained offline to determine the mapping between EMG signals and movements. Third, the one-step-ahead model is used to implement online intention estimation and real-time control of the upper limb elbow exoskeleton system.

[0031] The flow chart of the present invention is as follows Figure 1 As shown:

[0032] Part 1:

[0033] This section designed and implemented a multi-channel bioelectrical signal acquisition device based on SPI communication, using the ADS1298 chip to achieve high-precision and timely data acquisition. The system communicates with the main controller via a standard SPI four-wire communication interface (including the SCLK clock pin, MISO data output pin, MOSI data input pin, and CS chip select pin), implementing functions such as register configuration and data reading. Auxiliary control pins such as START, PWDN, RESET, and DRDY complete the timing control of the entire process from initialization and configuration to data acquisition.

[0034] During system startup, the system first powers on by pulling the PWDN pin high. Then, pulling the RESET pin high resets the chip, restoring all registers to their default states. Subsequently, based on specific application requirements, the host controller continuously sends three-byte WREG instructions to the ADS1298 chip via the SPI bus. These instructions correspond to the register write command and start address, the register write count minus one, and the register configuration data. After this process is complete, the START pin is pulled high to initiate continuous data acquisition mode.

[0035] During the data acquisition phase, the chip indicates data readiness via a low-level pulse on the DRDY pin at a fixed rate. The host controller then reads the data via the SPI interface. Each channel's acquired data is a 24-bit value in two's complement format. According to the ADS1298 data format, the raw value range is ±(2²³-1), where 0x000000 represents the minimum value of 0, 0x7FFFFF represents the positive full-scale range, and 0x800000 to 0xFFFFFF represents the negative range. In this device, all values ​​are ultimately calculated at the minimum resolution and mapped to the corresponding analog signal amplitude, enabling accurate myoelectric measurement.

[0036] The flow chart of EMG signal acquisition is as follows Figure 2 As shown:

[0037] Part II:

[0038] In order to construct the mapping relationship between electromyographic signals and joint angles, the present invention establishes a one-step-ahead (OSA) prediction model based on a nonlinear autoregressive (NARX) model with exogenous input to output the expected joint angles.

[0039] The nonlinear autoregressive moving average model with exogenous input (NARMAX) is a generalized nonlinear system identification model that combines the advantages of the nonlinear autoregressive (NAR) and autoregressive moving average (ARMAX) models with exogenous input. As a subset of the NARMAX model, the nonlinear autoregressive (NARX) model with exogenous input is more widely used in practice due to its strong noise resistance. This model combines the advantages of autoregressive and exogenous input, can effectively model the nonlinear relationship between input and output, and exhibits high prediction accuracy and good adaptability. The OSA model is a type of NARX model. In the present invention, the OSA model is combined with a shallow feedforward neural network to use the actual values ​​of the previous time steps to predict the output of the next time step. The proposed OSA model includes an input layer, a hidden layer and an output layer, and the hidden layer and the output layer use tansig and purelin activation functions, respectively. By transferring the prediction target from the current output to the output of the next time step, the OSA model realizes forward-looking control and can effectively compensate for system delays such as signal processing time and actuator response delay. This improvement helps to map features from historical data and exogenous inputs more directly and makes the model better adaptable to the time-varying dynamics of the system. The mathematical formula of the OSA model is shown in Equation (1):

[0040]

[0041] Among them, y k+1 represents the output of the system at time k+1, u k represents the input, f(·) is the nonlinear model function, y kis the sample value output at time k, n a and n b Respectively represent the order of the model input and model delay sequence. In the present invention, n a and n b The structure diagram of the model is as follows Figure 3 As shown:

[0042] The offline training phase is carried out using pre-recorded surface electromyography signal data. The original electromyography signal contains interference and noise, such as power line interference, artifacts, and high-frequency noise. The signal is mainly concentrated in the frequency range of 10-400Hz, so we use a 10-400Hz fourth-order Butterworth bandpass filter to remove high-frequency and low-frequency interference, and then use a 50Hz notch filter to remove power frequency interference, and finally use full-wave rectification to obtain the preprocessed data. After obtaining the preprocessed data, the conjugate gradient algorithm is used to train the prediction model. A sliding window of 100ms is used to extract the mean absolute deviation features of the current moment and the previous moment from the dual-channel surface electromyography signal, and combined with the actual joint angles of the current moment and the previous moment as the input of the model. The model output is the predicted joint angle for the next time step. The flowchart of the offline training of the model is as follows. Figure 4 As shown:

[0043] Part III:

[0044] Based on the designed signal acquisition device and the trained prediction model, the present invention also realizes online estimation and real-time control. STM32F4 was selected as the controller of the present invention. All calculations, including signal acquisition, preprocessing, angle prediction and motor control, were implemented on the microcontroller. In the signal acquisition module, ADS1298 samples the surface electromyography signals of the biceps and triceps as digital signals with a sampling frequency of 1000Hz. The motor feedback angle is used as the actual angle as the input of the prediction model. Due to the noise interference in the original signal, a series connection of an IIR notch filter (50Hz) is first used to remove the power interference, and then an IIR Butterworth bandpass filter (10-400Hz) is used to further filter the surface electromyography signal. A sliding window with a window length of 100ms and a step size of 10ms is used to extract the mean absolute deviation feature of the preprocessed electromyography signal, and a one-step advance model of the surface electromyography drive is implemented in the microcontroller to obtain the desired joint angle in accordance with the autonomous will. Joint angle control is achieved through dual closed-loop servo control of the position loop and the velocity loop. The present invention uses position mode to achieve closed-loop control of the motor, sets the maximum speed value, and the control system uses the deviation between the desired angle and the exoskeleton motor angle for closed-loop control, converting the position error into the desired angular velocity, and then controlling the motor current through the magnetic field orientation algorithm. The flow chart of the motor position mode control of the present invention is shown in the figure. Figure 5 As shown:

[0045] This product is an upper limb elbow exoskeleton assistance system based on surface electromyography (sEMG). It is designed to provide intelligent assistance and rehabilitation support for patients with limited motor function or upper limb strength. The usage process is as follows: First, high-sensitivity surface electrode patches are attached to the corresponding positions of the biceps and triceps to collect real-time bioelectrical signals from the user's muscles. The system uses a built-in signal acquisition module to filter, amplify, and extract features from the EMG signals to ensure data accuracy and stability. Next, the exoskeleton is worn on the upper limb requiring assistance. After the device is powered on, the control system dynamically analyzes the user's movement intentions (such as elbow flexion or extension) based on the collected EMG signals and drives the servo motor or power-assisting mechanism to perform the corresponding movements, achieving natural and coordinated limb assistance. To ensure safety during use, the system is equipped with a one-button emergency stop mechanism. In the event of an accident or emergency, the user or caregiver can quickly press the emergency stop button to immediately disconnect the power supply and stop all device operations, preventing potential injury. The system is easy to operate and responds quickly. It is suitable for various application scenarios such as rehabilitation training, functional assistance, and auxiliary wear for the elderly.

[0046] This product has wide applicability and promotion value. For patients with stroke or spinal cord injury, the system can detect their electromyographic signals in real time, drive the exoskeleton to assist in completing elbow flexion and extension movements, promote nerve function reconstruction and motor recovery, and reduce the risk of muscle disuse atrophy. For the elderly or patients with muscle weakness, the device can be used as a functional compensatory tool to effectively compensate for insufficient muscle strength; in addition, for assembly line workers engaged in repetitive operations, the system can help share the load on the elbow joint and prevent strain injuries. In summary, this product can not only serve the patients' home rehabilitation needs, but also provide support for healthy people. It is suitable for many fields such as medical rehabilitation, elderly care assistive devices and industrial assistance.

Claims

1. An elbow joint exoskeleton control system based on myoelectric angle prediction, characterized in that: Includes a dual-channel surface electromyography signal acquisition device and a one-step-ahead prediction model for OSA: The dual-channel surface electromyography signal acquisition device acquires electromyography signals to provide input data for the system; Based on the nonlinear autoregressive model NARX with exogenous input, a one-step-ahead prediction model OSA is established and trained offline to determine the mapping relationship between electromyographic signals and movements. The one-step-ahead prediction model OSA is used to achieve online intention estimation and real-time control of the upper limb elbow exoskeleton system.

2. The elbow joint exoskeleton control system based on myoelectric angle prediction according to claim 1, characterized in that: The dual-channel surface electromyography signal acquisition device uses the ADS1298 chip and communicates with the main controller through a standard SPI four-wire communication interface. The communication interface includes the SCLK clock pin, MISO data output pin, MOSI data input pin, and CS chip select pin, and cooperates with the START, PWDN, RESET, and DRDY auxiliary control pins to complete the timing control of the entire process from initialization, configuration to data acquisition.

3. The elbow joint exoskeleton control system based on myoelectric angle prediction according to claim 2, characterized in that: During the startup process of the dual-channel surface electromyography signal acquisition device, the power-on operation is first completed by pulling the PWDN pin high, and then the RESET pin is pulled high to complete the chip reset, so that all registers are restored to the default state; then, according to the specific application requirements, the main controller continuously sends three-byte WREG instructions to the ADS1298 chip through the SPI bus, corresponding to the register write command and starting address, the register write quantity minus one, and the register configuration data; after the above process is completed, the START pin is pulled high to start the continuous data acquisition mode.

4. The elbow joint exoskeleton control system based on myoelectric angle prediction according to claim 3, characterized in that: During the data acquisition phase, the chip indicates that the data is ready through a low-level pulse on the DRDY pin at a fixed rate. At this time, the main controller reads the data through the SPI interface. The data collected for each channel is a 24-bit value in the complement format. According to the data format of ADS1298, the original value range is ±(2 23 -1), where 0x000000 represents the minimum value 0, 0x7FFFFF is the positive full scale, and 0x800000 to 0xFFFFFF is the negative range; in the dual-channel surface electromyography signal acquisition device, all values ​​are finally calculated through the minimum resolution and mapped to the corresponding analog signal amplitude to achieve electromyography measurement.

5. The elbow joint exoskeleton control system based on myoelectric angle prediction according to claim 1, characterized in that: In step 2, the one-step-ahead prediction model OSA is combined with a shallow feedforward neural network, and the actual values ​​of the previous time steps are used to predict the output of the next time step. The proposed one-step-ahead prediction model OSA includes an input layer, a hidden layer, and an output layer. The hidden layer and the output layer use the tansig and purelin activation functions respectively. The prediction target is transferred from the current output to the output of the next time step. The mathematical formula of the one-step-ahead prediction model OSA is shown in formula (1): Among them, y k+1 represents the output of the system at time k+1, u k represents the input, f(·) is the nonlinear model function, y k is the sample value output at time k, n a and n b denote the order of the model input and model delay sequences, respectively.

6. The elbow joint exoskeleton control system based on myoelectric angle prediction according to claim 1, characterized in that: In step 2, the offline training phase is performed using pre-recorded surface electromyography signal data; a 10-400 Hz fourth-order Butterworth bandpass filter is used to remove high-frequency and low-frequency interference, a 50 Hz notch filter is used to remove power frequency interference, and finally a full-wave rectifier is used to obtain pre-processed data; After obtaining the preprocessed data, the conjugate gradient algorithm is used to train the prediction model; The mean absolute deviation feature between the current moment and the previous moment is extracted from the dual-channel surface electromyography signal using a sliding window of 100ms, and combined with the actual joint angles of the current moment and the previous moment as the input of the model; the model output is the predicted joint angle of the next time step.

7. The elbow joint exoskeleton control system based on myoelectric angle prediction according to claim 1, characterized in that: In step 3, online intention estimation and real-time control are realized through the main controller; the desired joint angle is obtained through the one-step-ahead prediction model OSA, and the joint angle control is realized through dual closed-loop servo control of the position loop and the speed loop.

8. The elbow joint exoskeleton control system based on myoelectric angle prediction according to claim 7, characterized in that: The position mode is used to implement closed-loop control of the motor, setting the maximum speed value. The control system uses the deviation between the desired angle and the exoskeleton motor angle for closed-loop control, converting the position error into the desired angular velocity, and then controlling the motor current through the magnetic field orientation algorithm.

9. A method for using an elbow joint exoskeleton control system based on myoelectric angle prediction, characterized in that: An elbow exoskeleton control system based on electromyographic angle prediction as described in any one of claims 1 to 8 above is adopted; step 1, a dual-channel surface electromyographic signal acquisition device is used to obtain biceps and triceps signals; step 2, an OSA one-step-ahead prediction model is introduced to map the dynamic relationship between the electromyographic signal and the elbow joint angle; step 3, online estimation and real-time control of the elbow joint movement intention are realized through the integrated dual-channel surface electromyographic signal acquisition device and the OSA one-step-ahead prediction model; actual measurement results show that the average response time of the system is significantly shorter than the time window in which the electromyographic signal leads the muscle contraction.

Images