A control method for variable stiffness knee joint rehabilitation exoskeleton

Abstract

The present invention discloses a control method suitable for a variable-stiffness knee rehabilitation exoskeleton. By establishing an inverse kinematics model of the human lower limb, and using a multi-sensor acquisition system including surface electromyography sensors, angle sensors, and plantar pressure sensors, and constructing a long short-term memory neural network (LSTM) model, and constructing a training set and test set data to train the neural network model, the method can quickly and accurately estimate knee joint torque and stiffness, identify the wearer's movement intention, and use the estimated knee joint torque and stiffness values to perform admittance control on the exoskeleton position motor to achieve compliant control of the exoskeleton. Closed-loop PID control is performed on the stiffness motor to adjust the knee joint stiffness in real time. The control strategy proposed in the present invention can output continuous control signals, and after completing model training, it can identify the wearer's movement intention. The exoskeleton has good control compliance and is suitable for assisted walking for the elderly and postoperative rehabilitation training for patients with knee injuries.

Description

Technical Field

[0001] The present invention relates to the technical field of knee joint rehabilitation exoskeleton robots, and in particular to a variable stiffness knee joint rehabilitation exoskeleton and a control method and a rehabilitation training method suitable for the variable stiffness knee joint rehabilitation exoskeleton. Background Art

[0002] The knee rehabilitation exoskeleton is a wearable human-machine collaborative system that can generate a controllable assistive torque at the wearer's knee joint. Compared to traditional manual rehabilitation treatments, the knee rehabilitation exoskeleton has the characteristics of high rehabilitation efficiency and a short rehabilitation cycle. It has a kinematic structure similar to that of the human lower limbs and is usually designed with a driver, sensor, controller, and power supply. It can provide controllable assistive force / torque for the patient's knee joint, helping the patient regain muscle strength and motor control function. After wearing the exoskeleton, patients with knee injuries can achieve the same standing / sitting, walking, and going up and down stairs as normal people under the support and power drive of the exoskeleton, which can greatly improve the quality and enjoyment of their lives. The most important goal of the knee rehabilitation exoskeleton is to assist patients with knee motor dysfunction to stand and walk, restore the patient's physical and mental health, and build their confidence in life.

[0003] Knee injuries caused by neurological diseases, aging, and surgical injuries have long caused muscle weakness, gait disturbances, and even long-term motor disability, severely impacting patients' quality of life and confidence. Knee rehabilitation exoskeletons can provide more scientifically effective, simpler, and faster rehabilitation training. For patients with knee injuries, many exoskeleton controls simply address position control without considering the patient's movement intent and the external environment, resulting in a lack of safety and comfort. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: In response to the above problems, the present invention provides a variable stiffness knee joint rehabilitation exoskeleton and its control method, which not only takes into account the patient's movement intention, but also takes into account the stiffness change of the knee joint, estimates the knee joint torque and stiffness from the human body's knee joint angle signal and lower limb electromyographic signal, quickly and in real time obtains the patient's movement intention, and provides exoskeleton stiffness control and flexible position control respectively, making the rehabilitation process more stable, comfortable and safe.

[0005] In order to solve the above technical problems, a technical solution adopted by the present invention is:

[0006] Provided is a variable stiffness knee joint rehabilitation exoskeleton, comprising a variable stiffness knee joint rehabilitation exoskeleton body, a power module, a control system, a sensor acquisition module, and a control strategy module based on motion intention recognition;

[0007] The variable stiffness knee joint rehabilitation exoskeleton body includes a knee joint formed by hinged connection of a thigh rod and a calf rod at the bottom end of the thigh rod, and an ankle joint formed by hinged connection of the calf rod and a sole support at the bottom end of the calf rod. A screw-nut mechanism driven by a stiffness adjustment motor is fixedly installed on the side of the thigh rod, and a slider is fixedly connected to the power output end of the screw-nut mechanism. The power module includes a mounting bracket fixedly connected to the side of the calf rod, a position control motor fixedly installed on the mounting bracket, and a disc torsion spring rotatably installed on the mounting bracket. The output shaft end of the position control motor is connected to the torsion spring torque transmission shaft of the disc torsion spring through a gear pair. A groove rod is fixedly provided on the outer circumferential surface of the disc torsion spring, and the slider is slidably provided in the groove rod.

[0008] The sensor acquisition module includes a surface electromyography signal acquisition instrument, a potentiometer, a plantar pressure sensor, and a three-dimensional motion capture system. The surface electromyography signal acquisition instrument is used to collect electromyographic signals of the lower limbs. The potentiometer is installed at the hinge of the thigh rod and the calf rod to collect the rotation angle of the knee joint. The plantar pressure sensor is arranged on the bottom surface of the sole support to collect the ground reaction force. The three-dimensional motion capture system is used to collect the position and motion information of each marker point of the human lower limbs.

[0009] The control system includes a host computer PC and a slave single-chip computer. The host PC has built-in data processing software and uses serial port communication with the slave single-chip computer to receive and process data. The slave single-chip computer is connected to the stiffness adjustment motor, the position control motor, and the sensor acquisition module respectively.

[0010] The motion intention recognition is to estimate the knee joint torque and stiffness based on surface electromyography signals. By constructing a long short-term memory neural network model, the surface electromyography signals of the four muscles of the lower limb and the knee joint angle signals are used as neural network inputs, and the knee joint torque and stiffness calculated by inverse kinematics are used as model outputs. A training set and a test set are constructed for model training.

[0011] The control strategy refers to obtaining the knee joint torque and stiffness values in real time through motion intention recognition, and then controlling the position control motor admittance according to the output torque value to achieve flexible control of the exoskeleton, assisting the wearer in completing rehabilitation training, and controlling the stiffness adjustment motor according to the output stiffness value to adjust the exoskeleton's knee joint stiffness in real time.

[0012] A control method applicable to the variable stiffness knee joint rehabilitation exoskeleton is also provided, comprising the following steps:

[0013] S10. Collect kinematic data of the wearer while walking on a treadmill through a surface electromyography signal collector, a potentiometer, and a plantar pressure sensor, including surface electromyography signals of the legs collected by the surface electromyography signal collector, knee joint rotation angle collected by the potentiometer, and pressure data collected by the plantar pressure sensor.

[0014] S11, performing filtering preprocessing on the collected knee joint rotation angle data and surface electromyography signal data respectively;

[0015] S12. Calculate knee and ankle joint torques using inverse kinematics: A three-link model of the human lower limb is established using the thigh, calf, and foot. Newton-Euler equations are established for each link component to obtain a human body dynamics model. Plantar pressure is collected and calculated using a plantar pressure sensor. The position and motion state information of each marker point is obtained by attaching marker points to the lower limb using a three-dimensional motion capture system. The velocity and acceleration information of each marker point is calculated using supporting software, thereby obtaining the mass moment of inertia of each marker point. Ultimately, the knee joint torque is obtained by solving the problem.

[0016] S13. Construct a long-short-term memory neural network model to estimate knee joint torque and stiffness. The subject wears a variable-stiffness knee rehabilitation exoskeleton and a surface electromyography (SEM) signal collector and walks on a treadmill to collect data. The first 80% of the collected and processed knee joint angle and electromyographic signal data of the four muscles is used as the training set input data of the neural network model, and the calculated knee joint torque and stiffness values are used as the training set output data. The last 20% of the collected and processed knee joint angle and electromyographic signal data of the four muscles are used as the test set input data, and the calculated knee joint torque and stiffness values are used as the test set output data.

[0017] S14. Use the root mean square error to evaluate the quality of the human knee joint torque and stiffness prediction model and measure the error between the neural network model prediction value and the true value;

[0018] S15. After the knee joint stiffness value is estimated by the long short-term memory neural network, the transmission of the screw nut mechanism is driven by controlling the rotation of the stiffness adjustment motor, thereby adjusting the effective length of the groove rod, realizing real-time adjustment of the joint stiffness, and realizing the variable stiffness function. PID closed-loop control is adopted, and the motor works in speed mode to realize dynamic changes in joint stiffness; an admittance control method with trajectory tracking based on estimated torque is designed in the built-in data processing software of the host PC computer, with the knee joint angle curve based on the walking of different wearers and the estimated torque obtained by the long short-term memory neural network model as input, and the actual knee joint angle as output. The rotation of the gear pair is driven by controlling the rotation of the position control motor, thereby assisting the wearer to complete rehabilitation training through the disc torsion spring.

[0019] Furthermore, in step S11, the specific method of surface electromyography signal data collection and filtering preprocessing is:

[0020] A surface electromyography (SEM) signal acquisition device was attached to the vastus lateralis, semitendinosus, long head of the biceps femoris, and medial head of the gastrocnemius muscle of the right leg. Four channels of the SEM signal acquisition device were used to acquire EMG signals from the four muscles at a rate of 2000 Hz while the subject walked at a speed of 2 m / s for 1 minute. Noise was removed using a 50 Hz notch filter and a Butterworth bandpass filter with a bandwidth of 20 Hz to 500 Hz.

[0021] After that, feature extraction is performed. In order to ensure the continuity of the features, the preprocessed EMG signal is windowed with 30 sampling points, and the root mean square (RMS) of the time domain features is selected for analysis. The RMS calculation formula is as follows:

[0022]

[0023] Where Data[i] is the surface electromyography signal; N is the window size.

[0024] Furthermore, in step S12, the ankle joint torque and knee joint torque are calculated by the human body dynamics model in the three-link model of the human lower limb using the following formula:

[0025]

[0026]

[0027] Among them, M3 is the ankle joint torque, M2 is the knee joint torque, θ3 is the angle between the line connecting the foot center of mass and the ankle joint and the horizontal plane, θ2 is the angle between the line connecting the calf center of mass and the knee joint and the horizontal plane, m3 is the mass of the foot, m2 is the mass of the calf rod, l3 is the height of the ankle joint from the ground, l 2a is the distance from the center of mass of the calf to the center of the knee joint, l2 is the length of the calf, F x is the horizontal component of the ground force on the foot, F y is the vertical force of the ground on the foot, c 2x is the horizontal position of the center of mass of the lower leg, c 2y is the vertical position of the center of mass of the lower leg, c 3x is the horizontal position of the foot's center of mass, c 3y is the vertical position of the foot's center of mass, J3 is the foot's moment of inertia, and J2 is the calf's moment of inertia.

[0028] Furthermore, the hidden layer input of the long short-term memory neural network includes the current input X t , the hidden layer output vector h at the previous moment t-1 , and the hidden layer state C t-1, the hidden layer structure also includes a forget gate f t , update gate i t and output gate o t , forget gate f t The role of C is to determine the state t Which information needs to be discarded, update gate i t The function is to determine whether it can be used in state C t The update, after the forget gate f t and update gate i t , completion status C t Update of the output gate o t To determine the state C t How to affect the hidden layer state h t The output of the forget gate f t , update gate i t and output gate o t And the hidden layer state h t and the updated state C t The calculation expression is as follows:

[0029] f t =σ(W xf x t +W hf h t-1 +b f )

[0030] i t =σ(W xi x t +W hi h t-1 +b i )

[0031] o t =σ(W xo x t +W ho h t-1 +b o )

[0032]

[0033] h t =o t ⊙tanh(C t )

[0034] Among them, σ is the sigmoid function, tanh is the hyperbolic tangent function, W xf 、W xi 、W xo and b f 、b i 、b oThey are the weight parameters and bias parameters that need to be trained in the forget gate, update gate and output gate respectively.

[0035] Furthermore, in step S14, the formula for the root mean square error is:

[0036]

[0037] Among them, Y act Y is the knee joint torque value calculated by inverse kinematics or the stiffness value calculated by the stiffness calculation formula. pre is the knee joint torque value or stiffness value estimated by the neural network model, and N is the data length of the test sample sequence.

[0038] Furthermore, in step S15, the admittance control method with trajectory tracking based on estimated torque adopts the admittance control formula:

[0039]

[0040] Among them, q d and q k represents the target angle and actual angle to be tracked by the exoskeleton, τ h To estimate the moment, M d 、C d and K d are the mass, damping and stiffness parameters of the admittance controlled object, respectively.

[0041] Furthermore, in step S15, the control input of the stiffness control method is:

[0042]

[0043] Where e(t) is the error between the reference stiffness and the neural network estimated stiffness, K p , K i , K d are the parameters of the PID controller.

[0044] A rehabilitation training method based on a variable stiffness knee joint rehabilitation exoskeleton is also provided, comprising the following steps:

[0045] S20: After the relaxation activity, the subject puts on the patch of the surface electromyography signal acquisition instrument, wears the variable stiffness knee rehabilitation exoskeleton, starts the three-dimensional motion capture system, the host PC computer running program and the slave computer switch for initialization;

[0046] S21: A surface electromyography (SEM) signal acquisition instrument acquires and filters the surface electromyography (SEM) signals of the vastus lateralis, semitendinosus, long head of the biceps femoris, and medial head of the gastrocnemius. A potentiometer acquires knee joint angle data. A plantar pressure sensor acquires plantar pressure data. A three-dimensional motion capture system acquires kinematic data. MATLAB software pre-installed on the host PC calculates the subject's knee joint torque and stiffness.

[0047] S22: The subject walks continuously on the treadmill, and the system continuously collects data. The first 80% of the collected and processed knee joint angle and electromyographic signal data of the four muscles are used as the training set input data of the neural network model, and the calculated knee joint torque and stiffness values are used as the training set output data. The last 20% of the collected and processed knee joint angle and electromyographic signal data of the four muscles are used as the test set input and output data, respectively. A program is run on MATLAB to train the model, and the trained model is exported after the model is trained.

[0048] S23: Setting the exoskeleton knee joint reference trajectory in the MATLAB software Simulink module. Based on the knee joint torque and stiffness values estimated in real time in step S22, the lower computer single-chip computer runs the admittance control algorithm to control the position control motor to perform corresponding actions, thereby achieving smooth control of the joint rotation and trajectory tracking of the knee joint rehabilitation exoskeleton. Closed-loop PID control is used to control the stiffness adjustment motor to perform corresponding actions, thereby achieving stiffness tracking and control of the knee joint rehabilitation exoskeleton.

[0049] S24: The subject wears the exoskeleton and wears sensors in accordance with step S21 to collect relevant data. The slave microcontroller transmits the collected data to the host PC via the communication serial port. The host PC processes the data and imports the processed data into the trained neural network model. The neural network model estimates the knee joint torque and stiffness values in real time based on the input data. The host PC sends the output data in real time to the slave microcontroller. The microcontroller executes the program to perform rehabilitation training.

[0050] S25: Repeat steps S23 and S24 until the subject completes rehabilitation training;

[0051] S26: Turn off the host PC, and the subject takes off the variable stiffness knee rehabilitation exoskeleton and removes the lower limb surface electromyography signal acquisition device patch.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0053] 1. The variable-stiffness knee rehabilitation exoskeleton control method of this invention can control the knee joint stiffness in real time based on the subject's movement intentions, further enhancing the safety of the knee rehabilitation exoskeleton and improving the walking ability of patients with knee injuries. Furthermore, the prototype's adjustable stiffness improves the safety, comfort, and versatility of the exoskeleton prototype's interaction with the subject's body.

[0054] 2. The variable stiffness knee rehabilitation exoskeleton control system of the present invention calculates knee joint torque based on the long short-term memory neural network model and inverse kinematics. After the model training is completed, there is no need to calculate the joint torque and stiffness, which reduces the use of sensors. The joint torque and stiffness can be estimated in real time through the model, and the real-time performance of the machine learning algorithm is better.

[0055] 3. The variable-stiffness knee rehabilitation exoskeleton control system of this invention utilizes closed-loop PID control for real-time stiffness adjustment. Stiffness is estimated based on the subject's movement intent, adapting to the environment and rehabilitation task, and better meeting the laws of human movement. During rehabilitation training, the exoskeleton employs a compliant control method, improving the stability and comfort of exoskeleton control, making rehabilitation training more convenient for patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a three-dimensional structural diagram of the variable stiffness knee joint rehabilitation exoskeleton robot of the present invention.

[0057] Figure 2 This is a partial view of the thigh of the variable stiffness knee joint rehabilitation exoskeleton robot of the present invention.

[0058] Figure 3 This is a diagram showing the variable stiffness principle of the variable stiffness knee joint rehabilitation exoskeleton robot of the present invention.

[0059] Figure 4 It is a diagram of the three-link dynamic model of the lower limb.

[0060] Figure 5 This is a schematic diagram of the hidden layer nodes of the LSTM neural network.

[0061] Figure 6 This is a diagram of torque data calculated by inverse kinematics in the test experiment.

[0062] Figure 7 This is a comparison chart of the estimated torque and calculated torque of the LSTM neural network prediction set.

[0063] Figure 8 It is a graph of inverse calculated stiffness data in the test.

[0064] Figure 9 This is a comparison chart of the estimated stiffness and calculated stiffness of the LSTM neural network prediction set.

[0065] Figure 10 This is the model diagram of the admittance control algorithm designed by the MATLAB software simulink module.

[0066] Figure 11 It is the simulation diagram of admittance control trajectory tracking.

[0067] Figure 12 This is a flow chart of the motion control method of the variable stiffness knee joint rehabilitation exoskeleton robot of the present invention.

[0068] In the figure: 1 stiffness adjustment motor encoder; 2 stiffness adjustment motor; 3 coupling; 4 slide; 5 screw-nut mechanism; 6 slider; 7 disc torque; 8 gear pair; 9 position control motor; 10 position control motor encoder; 11 calf rod; 12 ankle joint shaft; 13 ankle joint assembly; 14 sole support; 15 plantar pressure insole; 16 thigh rod; 17 potentiometer. DETAILED DESCRIPTION

[0069] The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more precise definition of the protection scope of the present invention.

[0070] See also Figure 1 and Figure 2 A variable stiffness knee joint rehabilitation exoskeleton includes a variable stiffness knee joint rehabilitation exoskeleton body, a power module, a control system, a sensor acquisition module and a control strategy module based on motion intention recognition.

[0071] Among them, the variable stiffness knee joint rehabilitation exoskeleton body includes a knee joint formed by a thigh rod 16 and a calf rod 11 at the bottom end of the thigh rod 16 being hinged through a knee joint pivot, and an ankle joint formed by a calf rod 11 and a sole support 14 at the bottom end of the calf rod 11 being hinged through an ankle joint pivot 12. A thigh fixing frame is fixedly provided on the inner side of the thigh rod 16, through which the thigh rod 16 can be worn on the thigh of the rehabilitation trainee and fastened by a Velcro strap; a calf fixing frame is fixedly provided on the inner side of the calf rod 11, through which the calf rod 11 can be worn on the calf of the rehabilitation trainee and fastened by a Velcro strap; the inner end of the ankle joint shaft 12 is fixedly connected to the ankle joint assembly 13, and the sole support 14 is fixedly installed on the side of the ankle joint assembly 13, so that the sole support 14 is located directly below the thigh fixing frame and the calf fixing frame; a plantar pressure insole 15 is fixedly provided on the surface of the sole support 14, and after the thigh rod 16 and the calf rod 11 are fixed, the plantar pressure insole 15 is just on the sole of the foot of the rehabilitation trainee.

[0072] A screw-nut mechanism 5 driven by a stiffness adjustment motor 2 is fixedly mounted on the side of the thigh rod 16, and a slider 6 is fixedly connected to the power output end (on the nut seat) of the screw-nut mechanism 5. Specifically, the stiffness adjustment motor 2 adopts a DC motor, and a stiffness adjustment motor encoder 1 is provided at its end for detecting the rotation angle of the stiffness adjustment motor 2. A slide 4 is fixedly mounted on the outer surface of the thigh rod 16, and the screw-nut mechanism 5 is rotatably mounted on the slide 4. The output shaft of the stiffness adjustment motor 2 is transmission-connected to the power input end (the shaft end of the screw) of the screw-nut mechanism 5 through a coupling 3. A guide rod is also provided on the slide 4, which is located on one side of the screw-nut mechanism 5. A guide block is provided on the sliding sleeve of the guide rod and is fixedly connected to the nut seat. The upward and downward displacement of the slider 6 can be achieved through the rotation of the stiffness adjustment motor 2 and the transmission of the screw-nut mechanism, thereby adjusting the position of the slider 6.

[0073] The power module includes a mounting bracket fixedly connected to the side of the calf rod 11, a position control motor 9 fixedly mounted on the mounting bracket, and a disc torsion spring 7 rotatably mounted on the mounting bracket. The output shaft end of the position control motor 9 is connected to the torsion spring torque transmission shaft of the disc torsion spring 7 through a gear pair 8. A groove rod is fixedly provided on the outer circumference of the disc torsion spring 7, and a slider 6 is slidably provided within the groove rod. The position control motor 9 is a DC motor, and a position control motor encoder 10 is provided at its end for detecting the rotation angle of the position control motor 9. The gear pair 8 adopts a bevel gear transmission pair, with a driving pinion fixedly mounted on the output shaft end of the position control motor 9, and a driven gear meshing with the driving pinion is fixedly mounted on the shaft end of the torsion spring torque transmission shaft of the disc torsion spring 7. By controlling the rotation of the position control motor 9 and the transmission of the gear pair 8, the torque can be flexibly transmitted to the slider 6 through the disc torsion spring 7, thereby generating a lateral force on the thigh rod 16. Since the sliding module is fixed on the thigh rod 16, the reaction force exerted by the sliding module will push the calf rod 11 to rotate around the knee joint.

[0074] like Figure 3 As shown, the variable stiffness principle of the variable stiffness knee joint rehabilitation exoskeleton robot of the present invention is as follows:

[0075] When the rotation angle of the outer ring of the disc torsion spring relative to the inner ring is When the linear compression spring produces a corresponding linear displacement Δx, then:

[0076]

[0077] In the formula, R is the outer radius of the disc torsion spring.

[0078] The force F generated by the linear compression spring s A moment T will be generated at the center point O of the disc torsion spring. o2 Equivalent torque T o1, we can get the moment T o1 and T o2 as follows:

[0079] T o =F s R=k s ΔxR

[0080]

[0081] Where, T o is the central moment of the disc torsion spring, k is the stiffness coefficient of the disc torsion spring, k s is the stiffness coefficient of the equivalent linear compression spring.

[0082] Therefore, the relationship between the stiffness coefficient of the disc torsion spring and the stiffness coefficient of the linear compression spring can be derived:

[0083] k s R 2 =k

[0084] The shank rod 11 is fixed in a vertical state, the slider 6 is regarded as the pivot point A, the groove rod and the thigh rod 16 are regarded as the connecting rod rotating around the slider 6, the AB segment represents the thigh rod, and the point O corresponds to the knee joint axis. Then the rotation angle θ of the AB segment connecting rod around the pivot point A is the rotation angle of the shank rod 11 when the thigh rod 16 is fixed. When the position control motor 9 generates an angle variable When the equivalent linear compression spring of the disc torsion spring 7 exerts a force F on the pivot point A, s The torque T caused can be calculated by the following formula:

[0085]

[0086] Where R is the vertical distance between the center axis of the equivalent linear compression spring of the disc torsion spring and the center line of the knee joint axis, k is the stiffness coefficient of the disc torsion spring, k s is the stiffness coefficient of the equivalent linear compression spring of the disc torsion spring, and L is the distance from the central axis of the equivalent linear compression spring of the disc torsion spring to the pivot point A.

[0087] Here, the vertical distance from point B to point A is set to (L+R). Therefore, a force F is generated at the connecting axis position at point B, which can be expressed as follows:

[0088]

[0089] The calculation formula of knee joint stiffness is:

[0090]

[0091] Among them, θ is the deflection angle of the thigh rod, T is the torque at the pivot point A, and L is the effective length of the groove rod adjusted by the stiffness adjustment motor 2. It can be seen from the formula that the stiffness of the knee joint is linearly proportional to the effective length L of the groove rod. When the length L changes, the stiffness will also change accordingly. Therefore, after the joint stiffness is estimated by the neural network, the length of the groove rod can be adjusted by controlling the stiffness adjustment motor 2 to drive the screw nut mechanism 5, thereby achieving real-time adjustment of the joint stiffness and realizing the variable stiffness function of the exoskeleton. When the stiffness adjustment motor 2 rotates at an angle of αrad, the displacement of the slider 6 on the screw can be calculated by the following formula:

[0092]

[0093] Where s is the displacement of the slider and p is the thread pitch of the screw.

[0094] The sensor acquisition module includes a surface electromyography signal collector, a potentiometer 17, a plantar pressure sensor 15 and a three-dimensional motion capture system (all existing equipment). The surface electromyography signal collector is used to collect electromyography signals of the lower limbs, specifically the surface electromyography signals of the vastus lateralis, semitendinosus, long head of the biceps femoris and medial head of the gastrocnemius. The potentiometer is installed at the hinge of the thigh rod and the calf rod, that is, the knee joint of the exoskeleton robot, to collect the rotation angle of the knee joint. The plantar pressure sensor is set on the bottom surface of the sole support to collect the ground reaction force, and the three-dimensional motion capture system is used to collect the kinematic data of the wearer when walking on the treadmill, that is, the position and motion information of each marker point of the human lower limb. The above-collected signal data is used as input data, and the torque of the knee joint is calculated by inverse kinematics. Then, the joint torque and angle can be used to calculate the joint stiffness through the formula, which is used as the output result of the neural network model.

[0095] The control system consists of a host PC and a slave microcontroller. The host PC has built-in data processing software (MATLAB) and communicates with the slave microcontroller via serial ports, responsible for data reception and processing. The serial port number, baud rate, and input / output buffer sizes are set in the communication initialization module of the program running on the host PC. After connecting the host PC to the microcontroller via a USB cable, the program is run and the connection is automatically completed. The slave microcontroller is connected to the stiffness adjustment motor, position control motor, and sensor acquisition module, respectively, responsible for data acquisition, data transmission, and motor control. The data processing software receives data from sensors such as the potentiometer and plantar pressure sensor and transmits the processed data to the slave microcontroller, which then sends control commands to the controllers of the stiffness adjustment motor and position control motor to control the motors.

[0096] Movement intention recognition refers to estimating the knee joint torque and stiffness based on surface electromyography signals. By constructing a long short-term memory neural network (LSTM) model (hereinafter referred to as the LSTM model), the surface electromyography signals of the four muscles of the lower limbs and the knee joint angle signals are used as neural network inputs, and the knee joint torque and stiffness calculated by inverse kinematics are used as model outputs. Training sets and test sets are constructed for model training.

[0097] The control strategy uses the previously mentioned principles of compliant admittance control and variable stiffness to determine the knee joint torque and stiffness values obtained in real time through motion intention recognition. This strategy implements compliant exoskeleton control based on the output torque value, controlling the position control motor's admittance control to assist the wearer in completing rehabilitation training, improving the stability and comfort of the exoskeleton's control, and facilitating rehabilitation training for patients. The stiffness adjustment motor is controlled based on the output stiffness value to adjust the exoskeleton's knee joint stiffness in real time. Closed-loop PID control is used to adjust stiffness in real time, estimating stiffness based on the subject's motion intention and adapting to the variable stiffness behavior of the environment and rehabilitation task, more in line with the laws of human movement.

[0098] See also Figures 4 to 11 A control method for a variable stiffness knee joint rehabilitation exoskeleton comprises the following steps:

[0099] S10. The three-dimensional motion capture system composed of a surface electromyography signal collector, a potentiometer, and a plantar pressure sensor is used to collect the kinematic data of the wearer while walking on a treadmill, including the surface electromyography signals of the legs collected by the surface electromyography signal collector, the knee joint rotation angle collected by the potentiometer, and the pressure data collected by the plantar pressure sensor.

[0100] S11, performing filtering preprocessing on the collected knee joint rotation angle data and surface electromyography signal data respectively;

[0101] (1) Knee joint angle data preprocessing:

[0102] The knee joint angle is measured using a potentiometer at the exoskeleton's knee joint. Before knee joint angle acquisition, the potentiometer's range is calibrated, selecting the middle portion as the new range, as the actual measured angle often differs from the brush rotation angle at both ends of the range. After calibration, the collected angle data is filtered.

[0103] (2) Surface electromyography signal data preprocessing:

[0104] Taking the collection of EMG signals from the wearer's right leg muscles as an example, before collecting the signals, to reduce the impedance between the surface EMG electrodes and the subject's skin, the hair on the relevant skin surface of the subject was shaved, and then the skin was wiped with alcohol cotton. Before the test, the subject did some appropriate relaxation exercises, and the surface EMG signal acquisition device patches were attached to the vastus lateralis, semitendinosus, long head of the biceps femoris, and medial head of the gastrocnemius muscle on the right leg. The surface EMG signal acquisition device was used to collect EMG signals from the four muscles of the subject at a rate of 2000Hz while walking at a speed of 1.5m / s for 1 minute. Noise was removed using a 50Hz notch filter and a Butterworth bandpass filter with a bandwidth of 20Hz-500Hz.

[0105] After that, feature extraction is performed. To ensure the continuity of the features, the preprocessed EMG signal is windowed with 30 sampling points, and the root mean square (RMS) of the time domain features is selected for analysis. The calculation formula of RMS is as follows:

[0106]

[0107] Where Data[i] is the surface electromyography signal and N is the window size.

[0108] S12. Calculate knee and ankle torques by inverse kinematics: Build a three-link model of the human lower limbs using the thigh, calf, and foot, such as Figure 4 As shown, Newton-Euler equations are established for each connecting rod component to obtain a human body dynamics model. Plantar pressure is collected and calculated using a plantar pressure sensor. The position and motion state information of each marker point attached to the human lower limb is obtained using a 3D motion capture system. The velocity and acceleration information of each marker point are calculated using supporting software, and the mass moment of inertia of each marker point is obtained. Finally, the human knee joint torque is solved.

[0109] The formulas used to calculate ankle joint moments and knee joint moments are as follows:

[0110]

[0111] Among them, M3 is the ankle joint torque, M2 is the knee joint torque, θ3 is the angle between the line connecting the foot center of mass and the ankle joint and the horizontal plane, θ2 is the angle between the line connecting the calf center of mass and the knee joint and the horizontal plane, m3 is the mass of the foot, m2 is the mass of the calf rod, l3 is the height of the ankle joint from the ground, l 2a is the distance from the center of mass of the calf to the center of the knee joint, l2 is the length of the calf, F x is the horizontal component of the ground force on the foot, F y is the vertical force of the ground on the foot, c 2x is the horizontal position of the center of mass of the lower leg, c 2y is the vertical position of the center of mass of the lower leg, c 3xis the horizontal position of the foot's center of mass, c 3y is the vertical position of the foot's center of mass, J3 is the foot's moment of inertia, and J2 is the calf's moment of inertia.

[0112] S13. Build a long short-term memory neural network (LSTM) model to estimate knee joint torque and stiffness:

[0113] After wearing a variable-stiffness knee rehabilitation exoskeleton and a surface electromyography signal collector, the subjects walked on a treadmill to collect data. The first 80% of the collected and processed knee joint angle and electromyography signal data of the four muscles were used as the training set input data of the neural network model, and the calculated knee joint torque and stiffness values were used as the training set output data. The last 20% of the collected and processed knee joint angle and electromyography signal data of the four muscles were used as the test set input data, and the calculated knee joint torque and stiffness values were used as the test set output data.

[0114] In this embodiment, the schematic diagram of the hidden layer nodes of the LSTM neural network is as follows: Figure 5 As shown. The input of the hidden layer of the LSTM neural network includes the current input X t , the hidden layer output vector h at the previous moment t-1 , and the hidden layer state C t-1 , the hidden layer structure also includes a forget gate f t , update gate i t and output gate o t , forget gate f t The role of C is to determine the state t Which information needs to be discarded, update gate i t The function is to determine whether it can be used in state C t The update, after the forget gate f t and update gate i t , completion status C t Update of the output gate o t To determine the state C t How to affect the hidden layer state h t The output of the forget gate f t , update gate i t and output gate o t And the hidden layer state h t and the updated state C t The calculation expression is as follows:

[0115] f t =σ(W xf x t +W hf h t-1 +b f )

[0116] i t =σ(W xi x t +W hi h t-1 +b i )

[0117] o t =σ(W xo x t +W ho h t-1 +b o )

[0118]

[0119] h t =o t ⊙tanh(C t )

[0120] Among them, σ is the sigmoid function, tanh is the hyperbolic tangent function, W xf 、W xi 、W xo and b f 、b i 、b o They are the weight parameters and bias parameters that need to be trained in the forget gate, update gate and output gate respectively.

[0121] S14. The root mean square error (RMSE) is used to evaluate the quality of the human knee joint torque and stiffness prediction model, and the RMSE is used to measure the degree of error between the predicted value and the true value of the neural network model.

[0122] The formula for the root mean square error RMSE is:

[0123]

[0124] Among them, Y act Y is the knee joint torque value calculated by inverse kinematics or the stiffness value calculated by the stiffness calculation formula. pre is the knee joint torque value or stiffness value estimated by the neural network model, and N is the data length of the test sample sequence.

[0125] S15. After the knee joint stiffness value is estimated by the long short-term memory neural network, as mentioned above, the transmission of the screw nut mechanism is driven by controlling the rotation of the stiffness adjustment motor, thereby adjusting the effective length of the groove rod, realizing real-time adjustment of the joint stiffness, and realizing the variable stiffness function. PID closed-loop control is adopted, and the motor works in speed mode to realize dynamic changes in joint stiffness; an admittance control method with trajectory tracking based on estimated torque is designed in the MATLAB software simulink module built into the host PC computer, with the knee joint angle data based on the walking of different wearers and the estimated torque obtained by the long short-term memory neural network model as input, and the actual knee joint angle as output. The rotation of the gear pair is driven by controlling the rotation of the position control motor, thereby assisting the wearer to complete rehabilitation training through the disc torsion spring.

[0126] The admittance control formula used is:

[0127]

[0128] Among them, q d and q k represents the target angle and actual angle to be tracked by the exoskeleton, τ h To estimate the moment, M d 、C d and K d are the mass, damping and stiffness parameters of the admittance controlled object, respectively.

[0129] Figure 10 The admittance control algorithm is designed in the simulink module of MATLAB software. In the figure, expect_input represents the reference trajectory, that is, the reference angle of the knee joint; Imported_Signal 1 represents the input estimated knee joint torque; admittance_ctrl represents the admittance control module; position_control represents the position tracking module of the knee rehabilitation exoskeleton; and out.Q represents the actual angle of the knee joint.

[0130] Using the above admittance control method, the reference trajectory is set to a sine signal with an amplitude of 5 degrees, and a verification test is performed. The overall effect is as follows Figure 11 As shown in Figure 3, the solid line represents the reference trajectory curve to be tracked, and the dotted line represents the actual tracking trajectory curve. It can be seen intuitively that through the compliant control performed by the admittance control algorithm, the actual tracking trajectory curve basically matches the reference trajectory curve to be tracked, with a small error and stability within a small and reasonable range.

[0131] The exoskeleton stiffness control method uses PID closed-loop control. The motor works in speed mode and the control input is:

[0132]

[0133] Where e(t) is the error between the reference stiffness and the neural network estimated stiffness, K p , K i , K d are the parameters of the PID controller.

[0134] See also Figure 12 A rehabilitation training method based on a variable stiffness knee joint rehabilitation exoskeleton comprises the following steps:

[0135] S20: After the relaxation activity, the subject puts on the patch of the surface electromyography signal acquisition instrument, wears the variable stiffness knee rehabilitation exoskeleton, starts the three-dimensional motion capture system, the host PC computer running program and the slave computer switch for initialization;

[0136] S21: A surface electromyography (SEM) signal acquisition instrument acquires and filters the surface electromyography (SEM) signals of the vastus lateralis, semitendinosus, long head of the biceps femoris, and medial head of the gastrocnemius. A potentiometer acquires knee joint angle data. A plantar pressure sensor acquires plantar pressure data. A three-dimensional motion capture system acquires kinematic data. MATLAB software pre-installed on the host PC calculates the subject's knee joint torque and stiffness.

[0137] S22: The subject walks continuously on the treadmill, and the system continuously collects data. The first 80% of the collected and processed knee joint angle and electromyographic signal data of the four muscles are used as the training set input data of the neural network model, and the calculated knee joint torque and stiffness values are used as the training set output data. The last 20% of the collected and processed knee joint angle and electromyographic signal data of the four muscles are used as the test set input and output data, respectively. A program is run on MATLAB to train the model, and the trained model is exported after the model is trained.

[0138] Figure 6 and Figure 7 This is a comparison chart of the torque estimated by the LSTM neural network and the torque calculated by inverse dynamics. In this example, a healthy male was selected as the subject and asked to wear a lower limb surface electromyography signal acquisition device and a variable stiffness knee joint rehabilitation exoskeleton to perform a walking experiment. After the data collection is completed, it is imported into the LSTM neural network model for training. The overall knee joint torque dataset is as follows: Figure 6 As shown, the first 80% is the training set, which only shows the calculated torque, and the last 20% is the test set. The test set prediction effect is as follows Figure 7 As shown in the figure, the horizontal axis represents the test set data volume, the vertical axis represents the torque data of each data point, the solid line represents the torque calculated by inverse dynamics, and the dotted line represents the torque predicted by the neural network. A visual comparison shows that the torque predicted by the neural network is basically consistent with the torque calculated by inverse dynamics, with a small error value that is within a reasonable error range.

[0139] Figure 8 and Figure 9 This is a comparison chart of the stiffness estimated and calculated by the LSTM neural network. The overall knee joint torque dataset is as follows: Figure 8 As shown in the figure, the first 80% is the training set, which only shows the calculated stiffness, and the last 20% is the test set. The test set prediction effect is as follows Figure 9 As shown in the figure, the horizontal axis represents the test data volume, the vertical axis represents the stiffness data of each data point, the solid line represents the calculated stiffness, and the dashed line represents the neural network predicted stiffness. Similarly, a visual comparison shows that the neural network predicted stiffness is basically consistent with the inverse dynamics calculated torque, with a small error value that is within a reasonable error range.

[0140] S23: The exoskeleton knee joint reference trajectory is set in the MATLAB software Simulink module. Based on the knee joint torque and stiffness values estimated in real time in step S22, the lower computer single-chip computer runs the admittance control algorithm to control the position control motor (the M1 motor shown in the figure) to perform corresponding actions, thereby achieving smooth control of the joint rotation and trajectory tracking of the knee joint rehabilitation exoskeleton. A closed-loop PID control is used to control the stiffness adjustment motor (the M2 motor shown in the figure) to perform corresponding actions, thereby achieving stiffness tracking and control of the knee joint rehabilitation exoskeleton.

[0141] S24: The subject wears the exoskeleton and wears sensors in accordance with step S21 to collect relevant data. The slave microcontroller transmits the collected data to the host PC via the communication serial port. The host PC processes the data and imports the processed data into the trained neural network model. The neural network model estimates the knee joint torque and stiffness values in real time based on the input data. The host PC sends the output data in real time to the slave microcontroller. The microcontroller executes the program to perform rehabilitation training.

[0142] S25: Repeat steps S23 and S24 until the subject completes rehabilitation training;

[0143] S26: Turn off the host PC, and the subject takes off the variable stiffness knee rehabilitation exoskeleton and removes the lower limb surface electromyography signal acquisition device patch.

[0144] In addition, the variable stiffness knee rehabilitation exoskeleton is also equipped with an emergency stop switch to prevent accidents during rehabilitation training (such as program failure or mechanical component damage, etc.). When an accident occurs, the emergency stop switch can be pressed in time to disconnect the entire rehabilitation training system and ensure the safety of the wearer.

[0145] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention's description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A variable stiffness knee joint rehabilitation exoskeleton, characterized by: It includes a variable stiffness knee joint rehabilitation exoskeleton body, a power module, a control system, a sensor acquisition module, and a control strategy module based on motion intention recognition; The variable stiffness knee joint rehabilitation exoskeleton body includes a knee joint formed by hinged connection of a thigh rod and a calf rod at the bottom end of the thigh rod, and an ankle joint formed by hinged connection of the calf rod and a sole support at the bottom end of the calf rod. A screw-nut mechanism driven by a stiffness adjustment motor is fixedly installed on the side of the thigh rod, and a slider is fixedly connected to the power output end of the screw-nut mechanism. The power module includes a mounting bracket fixedly connected to the side of the calf rod, a position control motor fixedly installed on the mounting bracket, and a disc torsion spring rotatably installed on the mounting bracket. The output shaft end of the position control motor is connected to the torsion spring torque transmission shaft of the disc torsion spring through a gear pair. A groove rod is fixedly provided on the outer circumferential surface of the disc torsion spring, and the slider is slidably provided in the groove rod. The stiffness adjustment motor and the position control motor are both DC motors, and a stiffness adjustment motor encoder is provided at the end of each motor. The stiffness adjustment motor is used to detect the rotation angle of the stiffness adjustment motor, and the position control motor is used to detect the rotation angle of the position control motor. The sensor acquisition module includes a surface electromyography signal acquisition instrument, a potentiometer, a plantar pressure sensor, and a three-dimensional motion capture system. The surface electromyography signal acquisition instrument is used to collect electromyographic signals of the lower limbs. The potentiometer is installed at the hinge of the thigh rod and the calf rod to collect the rotation angle of the knee joint. The plantar pressure sensor is arranged on the bottom surface of the sole support to collect the ground reaction force. The three-dimensional motion capture system is used to collect the position and motion information of each marker point of the human lower limbs. The control system includes a host computer PC and a slave single-chip computer. The host PC has built-in data processing software and uses serial port communication with the slave single-chip computer to receive and process data. The slave single-chip computer is connected to the stiffness adjustment motor, the position control motor, and the sensor acquisition module respectively. The motion intention recognition is to estimate the knee joint torque and stiffness based on surface electromyography signals. By constructing a long short-term memory neural network model, the surface electromyography signals of the four muscles of the lower limb and the knee joint angle signals are used as neural network inputs, and the knee joint torque and stiffness calculated by inverse kinematics are used as model outputs. A training set and a test set are constructed for model training. The control strategy refers to obtaining the knee joint torque and stiffness values in real time through motion intention recognition, and then controlling the position control motor admittance according to the output torque value to achieve flexible control of the exoskeleton, assisting the wearer in completing rehabilitation training, and controlling the stiffness adjustment motor according to the output stiffness value to adjust the exoskeleton's knee joint stiffness in real time.

2. The variable stiffness knee joint rehabilitation exoskeleton according to claim 1, characterized in that: The control method includes the following steps: S10, collecting kinematic data of the wearer while walking on a treadmill using a surface electromyography signal collector, a potentiometer, and a plantar pressure sensor, including surface electromyography signals of the legs collected by the surface electromyography signal collector, knee joint rotation angles collected by the potentiometer, and pressure data collected by the plantar pressure sensor; S11, performing filtering preprocessing on the collected knee joint rotation angle data and surface electromyography signal data respectively; S12. Calculate knee and ankle joint torques using inverse kinematics: A three-link model of the human lower limb is established using the thigh, calf, and foot. Newton-Euler equations are established for each link component to obtain a human body dynamics model. Plantar pressure is collected and calculated using a plantar pressure sensor. The position and motion state information of each marker point is obtained by attaching marker points to the lower limb using a three-dimensional motion capture system. The velocity and acceleration information of each marker point is calculated using supporting software, thereby obtaining the mass moment of inertia of each marker point. Ultimately, the knee joint torque is obtained by solving the problem. S13. Construct a long-short-term memory neural network model to estimate knee joint torque and stiffness. The subject wears a variable-stiffness knee rehabilitation exoskeleton and a surface electromyography (SEM) signal collector and walks on a treadmill to collect data. The first 80% of the collected and processed knee joint angle and electromyographic signal data of the four muscles is used as the training set input data of the neural network model, and the calculated knee joint torque and stiffness values are used as the training set output data. The last 20% of the collected and processed knee joint angle and electromyographic signal data of the four muscles are used as the test set input data, and the calculated knee joint torque and stiffness values are used as the test set output data. S14. Use the root mean square error to evaluate the quality of the human knee joint torque and stiffness prediction model and measure the error between the neural network model prediction value and the true value; S15. After the knee joint stiffness value is estimated by the long short-term memory neural network, the transmission of the screw nut mechanism is driven by controlling the rotation of the stiffness adjustment motor, thereby adjusting the effective length of the groove rod, realizing real-time adjustment of the joint stiffness, and realizing the variable stiffness function. PID closed-loop control is adopted, and the motor works in speed mode to realize dynamic changes in joint stiffness; an admittance control method with trajectory tracking based on estimated torque is designed in the built-in data processing software of the host PC computer, with the knee joint angle curve based on the walking of different wearers and the estimated torque obtained by the long short-term memory neural network model as input, and the actual knee joint angle as output. The rotation of the gear pair is driven by controlling the rotation of the position control motor, thereby assisting the wearer to complete rehabilitation training through the disc torsion spring.

3. The variable stiffness knee joint rehabilitation exoskeleton according to claim 2, characterized in that: In step S11, the specific method of surface electromyography signal data acquisition and filtering preprocessing is: A surface electromyography (SEM) signal acquisition device was attached to the vastus lateralis, semitendinosus, long head of the biceps femoris, and medial head of the gastrocnemius muscle of the right leg. Four channels of the SEM signal acquisition device were used to acquire EMG signals from the four muscles at a rate of 2000 Hz while the subject walked at a speed of 2 m / s for 1 minute. Noise was removed using a 50 Hz notch filter and a Butterworth bandpass filter with a bandwidth of 20 Hz to 500 Hz. After that, feature extraction is performed. In order to ensure the continuity of the features, the preprocessed EMG signal is windowed with 30 sampling points, and the root mean square RMS of the time domain features is selected for extraction. The RMS calculation formula is as follows: ； in, is the surface electromyographic signal; is the window size.

4. The variable stiffness knee joint rehabilitation exoskeleton according to claim 2, characterized in that: In step S12, the ankle joint torque and knee joint torque are calculated by the human body dynamics model in the three-link model of the human lower limb using the following formula: ； ； in, is the ankle joint torque, is the knee joint torque, is the angle between the line connecting the foot center of mass and the ankle joint and the horizontal plane, is the angle between the line connecting the calf center of mass and the knee joint and the horizontal plane, For the quality of the foot, is the mass of the calf member, is the height of the ankle joint from the ground, is the distance from the center of mass of the calf to the center of the knee joint, is the calf length, is the horizontal component of the ground force on the foot, is the vertical component of the force exerted by the ground on the foot, is the horizontal position of the center of mass of the lower leg, is the vertical position of the center of mass of the calf, is the horizontal position of the foot's center of mass, is the vertical position of the foot's center of mass, is the moment of inertia of the foot, is the moment of inertia of the calf.

5. The variable stiffness knee joint rehabilitation exoskeleton according to claim 2, characterized in that: The hidden layer input of the long short-term memory neural network includes the current input , the hidden layer output vector at the previous moment , and the hidden layer state , the hidden layer structure also includes a forget gate , Update Gate and output gate , the Forget Gate The role of determining the state Which information needs to be discarded and the update gate The function is to determine whether it can be used for status Update, through the forget gate and update gate , completed status Update, output gate To determine the status How to affect the hidden layer state The output of the forget gate , Update Gate and output gate and the hidden layer state and the updated status The calculation expression is as follows: ； ； ； ； ； in, is the sigmoid function, is the hyperbolic tangent function, 、 、 and 、 、 They are the weight parameters and bias parameters that need to be trained in the forget gate, update gate and output gate respectively.

6. The variable stiffness knee joint rehabilitation exoskeleton according to claim 2, characterized in that: In step S14, the formula for the root mean square error is: ； in, It is the knee joint torque value calculated by inverse kinematics or the stiffness value calculated by the stiffness calculation formula. is the knee joint torque or stiffness value estimated by the neural network model, is the data length of the test sample sequence.

7. The variable stiffness knee joint rehabilitation exoskeleton according to claim 2, characterized in that: In step S15, the admittance control formula used in the admittance control method with trajectory tracking based on estimated torque is: ； ； in, and Indicates the target angle and actual angle that the exoskeleton is tracking. To estimate the torque, 、 and are the mass, damping and stiffness parameters of the admittance controlled object, respectively.

8. The variable stiffness knee joint rehabilitation exoskeleton according to claim 2, characterized in that: In step S15, the control input of the stiffness control method is: ； in, is the error between the reference stiffness and the neural network estimated stiffness, 、 、 are the parameters of the PID controller.

9. The variable stiffness knee joint rehabilitation exoskeleton according to claim 1, characterized in that: The rehabilitation training method includes the following steps: S20: After the relaxation activity, the subject puts on the patch of the surface electromyography signal acquisition instrument, wears the variable stiffness knee rehabilitation exoskeleton, starts the three-dimensional motion capture system, the host PC computer running program and the slave computer switch for initialization; S21: A surface electromyography (SEM) signal acquisition instrument acquires and filters the surface electromyography (SEM) signals of the vastus lateralis, semitendinosus, long head of the biceps femoris, and medial head of the gastrocnemius. A potentiometer acquires knee joint angle data. A plantar pressure sensor acquires plantar pressure data. A three-dimensional motion capture system acquires kinematic data. MATLAB software pre-installed on the host PC calculates the subject's knee joint torque and stiffness. S22: The subject walks continuously on the treadmill, and the system continuously collects data. The first 80% of the collected and processed data of the knee joint angle and the electromyographic signals of the four muscles are used as the training set input data of the neural network model, and the calculated knee joint torque and stiffness values are used as the training set output data. The last 20% of the collected and processed data of the knee joint angle and the electromyographic signals of the four muscles are used as the test set input and output data, respectively. The model is trained by running a program on MATLAB, and the trained model is exported after the model is trained; S23: Setting the exoskeleton knee joint reference trajectory in the MATLAB software Simulink module. Based on the knee joint torque and stiffness values estimated in real time in step S22, the lower computer single-chip computer runs the admittance control algorithm to control the position control motor to perform corresponding actions, thereby achieving smooth control of the joint rotation and trajectory tracking of the knee joint rehabilitation exoskeleton. Closed-loop PID control is used to control the stiffness adjustment motor to perform corresponding actions, thereby achieving stiffness tracking and control of the knee joint rehabilitation exoskeleton. S24: The subject wears the exoskeleton and wears sensors in accordance with step S21 to collect relevant data. The slave microcontroller transmits the collected data to the host PC via the communication serial port. The host PC processes the data and imports the processed data into the trained neural network model. The neural network model estimates the knee joint torque and stiffness values in real time based on the input data. The host PC sends the output data in real time to the slave microcontroller. The microcontroller executes the program to perform rehabilitation training. S25: Repeat steps S23 and S24 until the subject completes rehabilitation training; S26: Turn off the host PC, and the subject takes off the variable stiffness knee rehabilitation exoskeleton and removes the lower limb surface electromyography signal acquisition device patch.

Images