A knee joint rehabilitation exoskeleton control method based on variable stiffness elastic driver

Abstract

The application discloses a knee joint rehabilitation exoskeleton control method based on a variable stiffness elastic driver, establishes a chaos grey wolf algorithm optimized support vector machine regression (CGWO-SVR) model, further improves the accuracy and processing speed of the support vector machine regression model through the chaos grey wolf optimization algorithm, thereby obtaining expected torque and stiffness, and in addition, utilizes an improved whale optimization algorithm (IWOA) to perform parameter setting on admittance parameters in admittance control, improves the accuracy of the output expected angle of a driving motor, and solves the problem that the tension on the transmission rope disappears in the transmission of the joint center problem, through establishing the relationship between the tension on the transmission rope and the angle of the power driving motor, the power driving motor is pre-rotated by a certain angle to pre-tighten the transmission rope before each driving, and the compensation angle and the expected angle are used for closed loop PIDA control on the power driving motor, so that the motion intention of the patient can be quickly and accurately recognized, and the compliance of the exoskeleton power output is improved.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation exoskeleton robot technology, and in particular to a control method for a knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator. Background Technology

[0002] With the increasing aging population globally, the number of patients suffering from lower limb movement disorders caused by diseases such as muscular dystrophy and stroke is on the rise. According to relevant medical theories and clinical experience, rehabilitation training can prevent muscle atrophy and promote the recovery of muscle function. Lower limb rehabilitation exoskeleton robots are mechatronic systems that interact with humans. By simulating the movements of human lower limb muscles, they can help patients exercise their lower limb muscles and restore their function. To better leverage the advantages of rehabilitation exoskeleton robots and improve the safety and comfort of patients wearing them, it is necessary to equip them with control methods adapted to different patient conditions.

[0003] To meet the wearer's needs in different situations, coordinating the movement of the wearer and the exoskeleton is crucial. The key technology in rehabilitation exoskeletons lies in accurately identifying the patient's movement intentions and effectively controlling the exoskeleton accordingly to meet the patient's needs in various environments. This is both the focus and the challenge in designing rehabilitation exoskeletons. Currently, methods for movement intention recognition are generally divided into two categories: thresholding and machine learning. Thresholding involves observing some human movement information and manually setting thresholds to classify different gaits. However, due to the subjective factors of different designers, the classification results will vary, and this classification algorithm is too coarse to meet the needs of complex situations, exhibiting poor generalization ability. Machine learning, which has emerged in recent years, is superior to thresholding. Machine learning involves building structural models of varying depths, collecting different features, adaptively learning the model's parameters, and establishing the relationship between the model's input and output. This method can adapt to the needs of different patients in different situations and also has higher accuracy.

[0004] In real-world environments, patients wearing exoskeletons encounter various errors. Even small positional errors, given the exoskeleton's high rigidity, can generate significant contact forces, posing a significant safety hazard. Therefore, it's crucial that the exoskeleton, while moving along a fixed trajectory, can dynamically adjust its angle and position according to different situations—a requirement known as compliant exoskeleton control. For knee rehabilitation exoskeletons, due to individual differences, patients' movement intentions are often unpredictable. Generally, torque sensors or inverse dynamics models can be used to obtain the desired torque. However, due to external interference, the desired torque obtained using torque sensors is often inaccurate. If an inverse dynamics model continuously outputs the desired torque, the system response becomes slow and inaccurate due to computational complexity and individual differences in knee joint movement angles. Furthermore, using general admittance control to obtain the desired trajectory often requires manual adjustment of the admittance parameters based on experience, which is time-consuming and often yields unsatisfactory results. Additionally, during knee joint rotation, the knee joint cannot be equated to a typical revolute joint; its rotation center constantly changes, causing discontinuities in the exoskeleton's transmission. Summary of the Invention

[0005] The technical problem to be solved by this invention is as follows: In order to quickly and accurately identify the patient's movement intention, a Chaotic Gray Wolf Algorithm-Optimized Support Vector Machine Regression (CGWO-SVR) model is established. The accuracy and processing speed of the support vector machine regression model are further improved by the Gray Wolf optimization algorithm, thereby obtaining the desired torque and stiffness. In addition, the Improved Whale Algorithm (IWOA) is used to tune the admittance parameters in admittance control, thereby improving the accuracy of the desired angle of the output drive motor. In order to solve the problem of the loss of tension on the transmission rope during transmission caused by the joint alignment problem, the relationship between the tension on the transmission rope and the angle of the power drive motor is established. The power drive motor is pre-rotated by a certain angle before each drive to pre-tighten the transmission rope. This compensation angle and the desired angle are used to perform closed-loop PIDA control on the power drive motor.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] A knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator includes a variable stiffness knee joint rehabilitation exoskeleton body, a power drive module, a stiffness adjustment module, a control module, a signal acquisition module, a data processing module, and a control module based on motion intention recognition.

[0008] The variable stiffness knee joint rehabilitation exoskeleton body includes a knee joint composed of a lower leg rod, a hinge shaft fixedly set on one side of the top of the lower leg rod, and a thigh rod movably hinged to the hinge shaft. The thigh rod is connected to an adjustable thigh mounting frame on its side, and the lower leg rod is connected to an adjustable lower leg mounting frame on the same side. The thigh mounting frame and the lower leg mounting frame can be fixedly set on the thigh and lower leg of the human body respectively by tightening straps.

[0009] The power drive module includes a mounting frame fixedly connected to the front side of the thigh mounting frame, a power drive motor fixedly connected to the mounting frame, a disc torsion spring and a planetary gear system rotatably mounted on the mounting frame, one shaft end of the disc torsion spring being driven to the output end of the power drive motor, and the other shaft end being driven to the shaft end of the sun gear of the planetary gear system, the shaft ends of the three planet gears of the planetary gear system being connected through a planet carrier, the shaft end of the central rotating shaft of the planet carrier being fixedly connected to a drive rope pulley, and a driven rope pulley being fixedly connected to the hinge shaft, and the drive rope pulley and the driven rope pulley being driven to each other through a transmission rope;

[0010] The stiffness adjustment module includes a stiffness adjustment motor fixedly connected to the mounting bracket and a stiffness adjustment gear fixedly connected to the output shaft end of the stiffness adjustment motor. The stiffness adjustment gear meshes with the gear ring of the planetary gear system.

[0011] The signal acquisition module includes a surface electromyography (EMG) signal acquisition device, a potentiometer, a tension sensor, and a motion capture system. The EMG signal acquisition device is used to acquire surface EMG signals of the lower limbs. The potentiometer is mounted on the hinge shaft and is used to acquire the relative rotation angle between the lower leg rod and the thigh rod. The tension sensor is set on the transmission rope and is used to measure the tension on the transmission rope. The motion capture system is used to acquire the position, velocity, and acceleration of each marker point on the lower limbs of the human body.

[0012] The control module includes a host PC, a slave microcontroller, a first encoder mounted on the power drive motor, and a second encoder mounted on the stiffness adjustment motor. The host PC has built-in data processing software and communicates with the slave microcontroller via serial port to receive and process data. The slave microcontroller communicates with the power drive motor and the stiffness adjustment motor via CAN bus and collects data from the signal acquisition module.

[0013] The aforementioned motion intention recognition refers to estimating the torque and stiffness of the knee joint by collecting surface electromyography signals and knee joint angles, optimizing the support vector machine regression model by establishing a chaotic gray wolf algorithm, using the collected signals as the input of the model, and using the torque and stiffness calculated by inverse dynamics as the output of the model, and training the support vector machine regression model by constructing training and testing sets;

[0014] The control module refers to tuning the admittance parameters of the admittance control using an improved whale algorithm. The torque estimated by the support vector machine regression model is used as the input of the admittance control, and the knee joint angle is used as the output of the admittance control. The admittance parameters are adjusted by evaluating the error between the admittance output angle and the actual knee joint angle. At the same time, the drive motor compensates for a certain angle to tighten the transmission rope. A closed-loop PIDA is constructed with the compensation angle and the output angle of the admittance control to perform compliant control of the drive motor.

[0015] Furthermore, a thigh connecting plate is provided between the thigh member and the thigh connecting frame, and a calf connecting plate is provided between the calf member and the calf connecting frame. Several evenly distributed bolt holes are provided on the side walls of the thigh member, calf member, thigh connecting frame, and calf connecting frame. The two sides of the thigh connecting plate are connected to the thigh member and the thigh connecting frame respectively by bolts, and the two sides of the calf connecting plate are connected to the calf member and the calf connecting frame respectively by bolts. By adjusting the assembly position of the bolts and bolt holes, the installation position of the thigh mounting frame on the thigh member and the installation position of the calf mounting frame on the calf member can be adjusted respectively.

[0016] Furthermore, the stiffness of the planetary frame is expressed as the output stiffness of the knee rehabilitation exoskeleton based on a variable stiffness elastic actuator, and is:

[0017]

[0018] Among them, K H M represents the output stiffness of the exoskeleton. H θ is the torque acting on the planet carrier. H The rotation angle of the planet carrier;

[0019] Based on the characteristics of the gear train transmission, the torque and angle relationship between the disc torsion spring and the planetary carrier can be derived. Furthermore, the relationship between the output stiffness of the exoskeleton and the stiffness of the disc torsion spring can be obtained as follows:

[0020] K H =K s ·i 2

[0021] Among them, K s Let be the torsional spring stiffness, and i be the transmission ratio between the sun gear and the planet carrier in the planetary gear train. The formula for calculating i is:

[0022]

[0023] Where Z1, Z3, and Z4 represent the number of teeth of the sun gear, the ring gear, and the stiffness adjustment gear, respectively, and the number of teeth on the inner and outer sides of the ring gear is equal. n1 and n4 are the rotational speeds of the sun gear and the stiffness adjustment gear, respectively.

[0024] Therefore, the output stiffness of the exoskeleton is:

[0025]

[0026] A method for controlling a knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator, applied to the aforementioned knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator, includes the following steps:

[0027] S10, The signal acquisition module collects kinematic data of the wearer while walking, including the surface electromyography signals of the leg collected by the surface electromyography signal acquisition device and the knee joint rotation angle collected by the potentiometer;

[0028] S11. Preprocess the collected knee joint rotation angle data and surface electromyography signal data respectively;

[0029] S12. Establish a Lagrange dynamic model of the lower limb two-linkage system through the thigh and lower leg links, obtain the motion information of the lower limb markers through the motion capture system, and solve the knee joint torque through the Euler-Lagrange equation in the data processing software.

[0030] S13. Construct a chaotic gray wolf algorithm to optimize the support vector machine regression model for motion intention recognition: Build the corresponding model on the data processing software, collect knee joint angle and surface electromyography signal data of the wearer in the walking state after wearing the exoskeleton, use the first percent of the collected and processed data as the input of the training set, and use the knee joint torque and stiffness values ​​calculated by the model as the output of the training set to train the prediction model, use the last percent of the collected and processed data as the input of the test set, and use the knee joint torque and stiffness values ​​calculated by the model as the output of the test set.

[0031] S14. Calculate the root mean square error of the true values ​​of knee joint torque and stiffness calculated by the Lagrange dynamics model and the estimated values ​​of knee joint torque and stiffness output by the support vector machine regression model. Use the root mean square error to evaluate the merits of the chaotic gray wolf algorithm-optimized support vector machine regression prediction model.

[0032] S15. During one gait cycle of the wearer wearing the exoskeleton, the angle of rotation of the power drive motor at some sampling time points is collected to tighten the transmission rope. Several sets of data points of time and power drive motor rotation angle are obtained. These data points are fitted with spline curves to obtain the motor rotation angle-time equation to compensate for the joint centering. Before each drive of the transmission rope by the power drive motor, the power drive motor is rotated in advance by the compensation angle to achieve continuous output of driving force on the transmission rope.

[0033] S16. After estimating the knee joint torque by optimizing the support vector machine regression model using the chaotic gray wolf algorithm, an admittance control model with the admittance parameters tuned by the improved whale optimization algorithm is first built on the host computer MATLAB software. After inputting the knee joint torque into the admittance control model, the admittance control model outputs the corresponding knee joint rotation angle.

[0034] S17. After the knee joint stiffness is estimated by optimizing the support vector machine regression model using the chaotic gray wolf algorithm, the stiffness adjustment motor is controlled by the lower-level microcontroller to drive the stiffness adjustment gear to rotate, thereby rotating the gear ring of the planetary gear train to achieve real-time stiffness adjustment. Real-time nonlinear adjustment of knee joint stiffness is achieved through closed-loop PID control.

[0035] S18. The knee joint rotation angle output by the admittance control model in step S16 and the compensation angle in step S15 to solve the problem of joint corona are used as inputs for closed-loop PIDA control to control the rotation of the power drive motor to drive the knee joint rotation and assist the wearer in rehabilitation training.

[0036] Furthermore, in step S11, the specific method for acquiring and filtering the surface electromyography signal data is as follows:

[0037] The surface electromyography (EMG) signal acquisition device patches were applied to the vastus lateralis, long head of biceps femoris, and medial head of gastrocnemius muscles of the right leg, respectively. The EMG signal acquisition device was used to acquire the EMG signals of the leg muscles of the wearer walking at a speed of 2m / s for 1 minute at a rate of 2000Hz using four channels. Then, a 50Hz band-stop filter and a 20Hz-500Hz Butterworth bandpass filter were used to remove noise from the signals.

[0038] The filtered electromyography signal was then divided into windows of 30 sampling points, and the root mean square (RMS) of the time-domain features was extracted. The formula for calculating RMS is as follows:

[0039]

[0040] Where X(i) is the electromyographic signal intensity value; N is the number of electromyographic signals within a window.

[0041] Furthermore, in step S12, the formula used to calculate the knee joint torque in the lower limb two-link Lagrange dynamics model is:

[0042]

[0043] Where M is the knee joint torque, J1 is the moment of inertia of the thigh member, J2 is the moment of inertia of the lower leg member, m1 is the mass of the thigh member, m2 is the mass of the lower leg member, θ1 is the angle between the line connecting the center of mass of the thigh member and the knee joint and the horizontal plane, θ2 is the angle between the line connecting the center of mass of the lower leg member and the knee joint and the horizontal plane, l 1a Let l1 be the distance from the center of mass of the thigh member to the center of rotation of the knee joint, l2 be the length of the thigh member, and F be the length of the lower leg member. x F is the horizontal component of the force acting on the knee joint. y c is the vertical component of the force acting on the knee joint. 1x Let c be the horizontal position of the center of mass of the thigh member pair. 1y Let c be the vertical position of the center of mass of the thigh member pair. 2x Let c be the horizontal position of the center of mass of the lower leg member pair. 2y The vertical position of the center of mass of the lower leg member.

[0044] Furthermore, in step S14, the formula for calculating the root mean square error is:

[0045]

[0046] Among them, X ref The knee joint stiffness is obtained by solving the Euler-Lagrange equations for the knee joint torque or by using the exoskeleton stiffness calculation formula. X pre It is the knee joint torque or stiffness estimated by the support vector machine regression model, and N is the data length of the test sample sequence.

[0047] Furthermore, in step S16, the admittance control model uses the following admittance control formula:

[0048]

[0049] Where e = q r -q a , representing the knee joint rotation angle error, q r q represents the desired rotation angle of the knee joint. a Let M be the desired rotation angle of the knee joint, τ be the estimated torque, and M be the torque. d C d and K d These are the mass, damping, and stiffness parameters of the admittance control object.

[0050] Furthermore, in step S17, the stiffness control formula for adjusting knee joint stiffness is:

[0051]

[0052] Where e(t) is the error between the predicted stiffness estimated by the support vector machine regression model and the reference stiffness calculated by the Lagrange dynamics model, and K p K i K d These are the proportional, integral, and derivative parameters of the PID controller.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0054] (1) The knee joint rehabilitation exoskeleton based on variable stiffness elastic actuator of the present invention uses the support vector machine regression model in the machine learning algorithm and optimizes it with the gray wolf optimization algorithm, which improves the training speed and accuracy of the model. After the model is trained, it is not necessary to recalculate the knee joint torque and angle, which reduces the use of sensors. It can quickly and accurately identify the wearer's movement intention, continuously estimate the torque and stiffness required for knee joint movement, improve the real-time performance of the exoskeleton's compliant control, and enhance the safety of the knee joint rehabilitation exoskeleton. In addition, based on the variable stiffness of the knee joint, the wearer's comfort can be improved. Only by retraining the model can the movement intention of different wearers be predicted, which improves the versatility of the exoskeleton.

[0055] (2) In the knee joint rehabilitation exoskeleton control method based on variable stiffness elastic actuator of the present invention, the improved whale optimization algorithm is used to tune the parameters of the admittance control model. After the torque is predicted, there is no need to manually select the admittance parameters, which greatly reduces the trial and error time of selecting the admittance parameters and improves the accuracy of the desired angle of the admittance control output.

[0056] (3) The knee joint rehabilitation exoskeleton based on the variable stiffness elastic actuator of the present invention, by making the power drive motor rotate a certain compensation angle in advance each time it rotates and outputs, keeps the transmission rope in a taut state, solves the problem of discontinuous transmission of the exoskeleton caused by the joint alignment problem, and improves the smoothness of the power output of the exoskeleton.

[0057] (4) The knee joint rehabilitation exoskeleton based on variable stiffness elastic actuator of the present invention predicts stiffness based on the wearer’s different movement intentions, which can flexibly adapt to different environments and the need for variable stiffness. It uses closed-loop PID to dynamically control the stiffness motor, which improves the stability and comfort of the exoskeleton. Attached Figure Description

[0058] Figure 1 This is a three-dimensional structural diagram of the knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator of the present invention.

[0059] Figure 2 This is a schematic diagram of the main structure of the knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator of the present invention.

[0060] Figure 3 This is a three-dimensional structural diagram of the knee joint rehabilitation exoskeleton of the present invention in the human leg wearing state;

[0061] Figure 4 This is a schematic diagram of the lower limb two-link Lagrange dynamics model established in this invention;

[0062] Figure 5 This is a schematic diagram of the tolerance principle of Support Vector Machine Regression (SVR) used in this invention;

[0063] Figure 6 This is a schematic diagram of the gray wolf position update method in the gray wolf optimization algorithm used in this invention;

[0064] Figure 7 This is a comparison chart of estimated torque and calculated torque in the CGWO-SVR simulation experiment;

[0065] Figure 8 This is a comparison chart of estimated stiffness and calculated stiffness in the CGWO-SVR simulation experiment;

[0066] Figure 9 This is a diagram of the admittance control model based on the WOA optimization algorithm in the Simulink module of MATLAB software.

[0067] Figure 10 This is a comparison chart of the actual angle and the desired angle obtained from the admittance control simulation experiment in Simulink;

[0068] Figure 11 This is a schematic diagram of the motion control system of a knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator.

[0069] In the diagram: 1. First encoder; 2. Power drive motor; 3. Sun gear; 4. Worm gear mechanism; 5. Disc torsion spring; 6. Planetary carrier; 7. Second encoder; 8. Stiffness adjustment motor; 9. Stiffness adjustment gear; 10. Transmission rope; 11. Tension sensor; 12. Potentiometer; 13. Gear ring; 14. Planetary gear; 15. Thigh rod; 16. Reducer; 17. Thigh connecting plate; 18. Drive rope pulley; 19. Lower leg rod; 20. Lower leg connecting plate; 21. Driven rope pulley. Detailed Implementation

[0070] The preferred embodiments of the present invention will now be described in detail 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 providing a clearer and more explicit definition of the scope of protection of the present invention.

[0071] Please see Figure 1 and Figure 2A knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator includes a variable stiffness knee joint rehabilitation exoskeleton body, a power drive module, a stiffness adjustment module, a control module, a signal acquisition module, a data processing module, and a control module based on motion intention recognition.

[0072] The variable stiffness knee joint rehabilitation exoskeleton body includes a knee joint composed of a lower leg member 19, a hinge shaft fixedly disposed on one side of the top of the lower leg member 19, and a thigh member 15 movably hinged to the hinge shaft. The side of the thigh member 15 (e.g.) Figure 1 The left side shown is connected to an adjustable thigh mounting frame, and the same side of the lower leg member 19 (as shown) is connected to the same side of the lower leg member 19. Figure 1 An adjustable calf mounting frame is connected to the left side (as shown in the diagram). The thigh mounting frame and calf mounting frame can be respectively fixed to the thigh and calf of the human body via tightening straps. Figure 3 As shown, several evenly distributed bolt holes are provided on the side walls of the thigh member 15, the lower leg member 19, the thigh connecting frame, and the lower leg connecting frame. By matching the bolt holes and connecting them with bolts, the thigh mounting frame can be connected and fixed on the thigh member 15, and the lower leg mounting frame can be connected and fixed on the lower leg member 19. By adjusting the matching positions between the bolt holes, the connection position of the thigh mounting frame on the thigh member 15 and the connection position of the lower leg mounting frame on the lower leg member 19 can be adjusted.

[0073] Preferably, a thigh connecting plate 17 is provided between the thigh rod 15 and the thigh connecting frame, and a calf connecting plate 20 is provided between the calf rod 19 and the calf connecting frame. Both the thigh connecting plate 17 and the calf connecting plate 20 are C-shaped sheet metal parts. The two sides of the thigh connecting plate 17 are connected to the thigh rod 15 and the thigh connecting frame respectively by bolts, and the two sides of the calf connecting plate 20 are connected to the calf rod 19 and the calf connecting frame respectively by bolts. This positions the thigh rod 15 at a certain distance outside the thigh mounting frame, and the calf rod 19 at a certain distance outside the calf mounting frame. This provides ample space for the installation of functional components such as the power drive module and the stiffness adjustment module, and also ensures that the knee joints of the two rods are parallel to each other on the outside of the human leg, preventing direct collision between the rods and the human leg during rotation and improving safety during use. By adjusting the assembly position of the bolts and bolt holes, the installation position of the thigh mounting frame on the thigh rod 15 and the installation position of the calf mounting frame on the calf rod 29 can be adjusted respectively.

[0074] The power drive module includes a mounting bracket fixedly connected to the front side of the thigh mounting frame, a power drive motor 2 fixedly connected to the mounting bracket, a disc torsion spring 5 and a planetary gear train rotatably mounted on the mounting bracket. The mounting bracket is composed of multiple sheet metal frames, which are fixedly connected by bolt pairs and are used for mounting mechanical structures such as the power drive motor 2, the stiffness adjustment motor 8, the disc torsion spring 5 and the planetary gear train. In this embodiment, the power drive motor 2 is fixedly mounted on the top front side of the mounting bracket, and its output shaft is vertically downward. A first encoder 1 is provided at the top shaft end of the power drive motor 2 for detecting the rotation angle of the power drive motor 2. The disc torsion spring 5 is located on one side below the power drive motor 2, with its central axis horizontally set and parallel to the rotation axis of the knee joint. Its input shaft and output shaft are rotatably connected to the two vertical side walls of the mounting bracket. The disc torsion spring 5 and the power drive motor 2 are connected by a worm gear mechanism 4. Specifically, the worm of the worm gear mechanism 4 is coaxially fixedly connected to the output shaft of the power drive motor 2, and the worm wheel of the worm gear mechanism 4 is coaxially fixedly connected to the input shaft of the disc torsion spring 5. This satisfies the requirement that the power drive motor 2 is positioned above the disc torsion spring 5 to make the structure compact, and also achieves smooth power transmission and unidirectional power transmission with a large transmission ratio.

[0075] The planetary gear train consists of a ring gear 13, a sun gear 3 coaxially mounted within the ring gear 13, and three planet gears 14 meshing with both the sun gear 3 and the ring gear 13. The outer surface of the ring gear 13 has an outer set of teeth equal in number to its inner teeth. A ring gear support is fixedly connected to the inner end face of the ring gear 13, and the ring gear support is rotatably mounted on the output shaft of the disc torsion spring 5, fixing the axis of the ring gear 13 and aligning it with the output shaft axis of the disc torsion spring 5, allowing the ring gear 13 to rotate around its axis. The inner shaft end of the sun gear 3 is fixedly connected to the output shaft end of the disc torsion spring 5. The disc torsion spring 5 flexibly transmits the driving force of the power drive motor 2 to the sun gear 3, thus providing power input to the planetary gear train. The shaft ends of the three planet gears 14 are connected by a planet carrier. When the ring gear 13 is relatively fixed and the sun gear 3 rotates, the three planet gears 14, through their revolution around the sun gear 3, drive the planet carrier to rotate, thus providing power output to the planetary gear train.

[0076] The stiffness of the planetary carrier, expressed as the output stiffness of this knee rehabilitation exoskeleton based on a variable stiffness elastic actuator, is:

[0077]

[0078] Among them, K H M represents the output stiffness of the exoskeleton. H θ is the torque acting on the planet carrier. H The rotation angle of the planet carrier;

[0079] Based on the characteristics of the gear train transmission, the torque and angle relationship between the disc torsion spring and the planetary carrier can be derived. Furthermore, the relationship between the output stiffness of the exoskeleton and the stiffness of the disc torsion spring can be obtained as follows:

[0080] K H =K s ·i 2

[0081] Among them, K s Let be the torsional spring stiffness, and i be the transmission ratio between the sun gear and the planet carrier in the planetary gear train. The formula for calculating i is:

[0082]

[0083] Where Z1, Z3, and Z4 represent the number of teeth of the sun gear, the ring gear, and the stiffness adjustment gear, respectively, and the number of teeth on the inner and outer sides of the ring gear is equal. n1 and n4 are the rotational speeds of the sun gear and the stiffness adjustment gear, respectively.

[0084] Therefore, the output stiffness of the exoskeleton is:

[0085]

[0086] A drive pulley 18 is fixedly connected to the central axis of the planetary frame, and a driven pulley 21 is fixedly connected to the hinge axis of the knee joint. The drive pulley 18 and the driven pulley 21 are connected by a transmission rope 10. When the planetary frame rotates and outputs power, it can drive the drive pulley 18 to rotate synchronously, which in turn drives the driven pulley 21 to rotate through the transmission rope 10. This causes the hinge axis and the lower leg member 19 to rotate synchronously with the driven pulley 21, ultimately achieving rotation of the lower leg member 19 relative to the thigh member 15. Since the lower leg member 19 is fixed to the lower leg through the lower leg mounting frame, and the thigh member 15 is fixed to the thigh through the thigh mounting frame, when the downward push member 19 rotates relative to the thigh member 15, it can actively apply bending and stretching power to the lower leg, thereby realizing leg-assisted rehabilitation training for the exoskeleton wearer.

[0087] The stiffness adjustment module includes a stiffness adjustment motor 8 fixedly connected to a mounting bracket and a stiffness adjustment gear 9 fixedly connected to the output shaft end of the stiffness adjustment motor 8. The stiffness adjustment gear 9 meshes with the ring gear 13 of the planetary gear train. Specifically, the stiffness adjustment motor 8 is located below the disc torsion spring 5, and its output shaft axis is parallel to the output shaft axis of the disc torsion spring 5. A reducer 16 is fixedly connected to the output shaft end of the motor 8, and the stiffness adjustment gear 9 is fixedly mounted on the output shaft end of the reducer 16. A second encoder 7 is provided at the end of the stiffness adjustment motor 8 to detect the rotation angle of the stiffness adjustment motor 8. The stiffness adjustment gear 9 and the ring gear 13 form a reduction transmission, and together with the reducer 16, they form a two-stage reduction structure. The stiffness adjustment motor 8 drives the stiffness adjustment gear 9 to rotate through the reducer 16, and the stiffness adjustment gear 9 drives the ring gear 13 to rotate slightly through meshing transmission, thereby realizing real-time stiffness adjustment.

[0088] The signal acquisition module includes a surface electromyography (EMG) signal acquisition device, a potentiometer 12, a tension sensor 11, and a motion capture system (all existing equipment). The EMG signal acquisition device is used to acquire surface EMG signals of the lower limbs, specifically the EMG signals of the vastus lateralis, long head of the biceps femoris, and medial head of the gastrocnemius muscle in the right leg. The potentiometer 12 is mounted on the hinge axis, i.e., at the knee joint of the exoskeleton robot, and is used to acquire the relative rotation angle (knee joint rotation angle) between the lower leg link 19 and the thigh link 15. The tension sensor 11 is set on the transmission rope 10 and is used to measure the tension on the transmission rope 10. The motion capture system is used to acquire the position, velocity, and acceleration of each marker point on the lower limb. Using the acquired signal data as input data, the torque and stiffness of the knee joint are calculated through inverse dynamics.

[0089] The control module includes a host PC, a slave microcontroller, a first encoder 1 mounted on the drive motor 2, and a second encoder 7 mounted on the stiffness adjustment motor 8. The host PC has built-in data processing software (MATLAB) and communicates with the slave microcontroller via serial port for data reception and processing. The host PC's communication module is configured with the serial port number, baud rate, and input / output buffer size, and is connected to the slave microcontroller via USB. The slave microcontroller communicates with the drive motor 2 and the stiffness adjustment motor 8 via CAN bus and collects data from the signal acquisition module.

[0090] Motion intention recognition refers to estimating the torque and stiffness of the knee joint by collecting surface electromyography signals and knee joint angles. A chaotic gray wolf algorithm is used to optimize the support vector machine regression model. The collected signals are used as the input of the model, and the torque and stiffness calculated by inverse dynamics are used as the output of the model. The support vector machine regression model is trained by constructing training and test sets.

[0091] The control module refers to the use of an improved whale algorithm to tune the admittance parameters for admittance control. The torque estimated by the support vector machine regression model is used as the input of the admittance control, and the knee joint angle is used as the output of the admittance control. The admittance parameters are adjusted by evaluating the error between the admittance output angle and the actual knee joint angle. At the same time, the drive motor 2 compensates for a certain angle to tighten the transmission rope 10. A closed-loop PIDA is constructed using the compensation angle and the output angle of the admittance control to perform compliant control on the drive motor 2.

[0092] The following section details the specific content of the control module based on motion intention recognition, using a knee joint rehabilitation exoskeleton control method based on variable stiffness elastic actuators.

[0093] Please see Figure 11 A control method for a knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator, applied to the aforementioned knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator, includes the following steps:

[0094] S10, the signal acquisition module collects kinematic data of the wearer while walking, including the surface electromyography (EMG) signals of the leg collected by the surface EMG signal acquisition device and the knee joint rotation angle collected by the potentiometer; the surface EMG signal acquisition device patch is attached to the vastus lateralis, long head of biceps femoris, and medial head of gastrocnemius muscle of the right leg respectively, and the four channels of the surface EMG signal acquisition device are used to collect the EMG signals of the leg muscles of the wearer walking at a speed of 2m / s for 1 minute at a rate of 2000Hz;

[0095] S11. Preprocess the collected knee joint rotation angle data and surface electromyography signal data respectively;

[0096] (1) Knee joint angle data preprocessing:

[0097] The knee joint angle was measured using a potentiometer located at the knee joint of the exoskeleton robot structure. Before acquiring the knee joint angle, the actual measured angle often differed from the brush rotation angle due to the limitations of the potentiometer's range. Therefore, the potentiometer's range needed to be calibrated, and a portion of the potentiometer's middle range was selected as the new range. Noise in the signal was removed using a 50Hz band-stop filter and a 20Hz-500Hz Butterworth bandpass filter.

[0098] (2) Preprocessing of surface electromyography signal data:

[0099] Noise in the signal was removed using a 50Hz band-stop filter and a 20Hz-500Hz Butterworth bandpass filter. The filtered electromyography (EMG) signal was then divided into windows of 30 sampling points, and the root mean square (RMS) of the time-domain features was extracted. The formula for calculating RMS is as follows:

[0100]

[0101] Where X(i) is the electromyographic signal intensity value; N is the number of electromyographic signals within a window.

[0102] S12. Establish a two-link Lagrangian dynamic model of the lower limb using the thigh and lower leg links, such as... Figure 4 As shown, motion information of lower limb markers is acquired through a motion capture system, and the knee joint torque is solved using the Euler-Lagrange equation in data processing software.

[0103] The formula used to calculate the knee joint torque in the lower limb two-link Lagrange dynamic model is:

[0104]

[0105] Where M is the knee joint torque, J1 is the moment of inertia of the thigh member, J2 is the moment of inertia of the lower leg member, m1 is the mass of the thigh member, m2 is the mass of the lower leg member, θ1 is the angle between the line connecting the center of mass of the thigh member and the knee joint and the horizontal plane, θ2 is the angle between the line connecting the center of mass of the lower leg member and the knee joint and the horizontal plane, l 1a Let l1 be the distance from the center of mass of the thigh member to the center of rotation of the knee joint, l2 be the length of the thigh member, and F be the length of the lower leg member. x F is the horizontal component of the force acting on the knee joint. y c is the vertical component of the force acting on the knee joint. 1x Let c be the horizontal position of the center of mass of the thigh member. 1y Let c be the vertical position of the center of mass of the thigh member. 2x Let c be the horizontal position of the center of mass of the lower leg member. 2y The vertical position of the center of mass of the lower leg member.

[0106] S13. Construct a chaotic gray wolf algorithm to optimize the support vector machine regression model for motion intention recognition: Build the corresponding model on the data processing software, collect knee joint angle and surface electromyography signal data of the wearer in the walking state after wearing the exoskeleton, use the first 70% of the collected and processed data as the input of the training set, and use the knee joint torque and stiffness values ​​calculated by the model as the output of the training set to train the prediction model, use the last 30% of the collected and processed data as the input of the test set, and use the knee joint torque and stiffness values ​​calculated by the model as the output of the test set.

[0107] Support Vector Machines (SVMs) are a type of generalized linear classifier that performs binary classification of data using supervised learning. Their decision boundary is the hyperplane with the maximum margin found among the training samples, which can be transformed into a convex quadratic programming problem. The addition of a kernel function can solve nonlinear problems and avoid the curse of dimensionality. The dimension of the feature subspace is closely related to the kernel function σ; the smaller the σ, the more detailed the classification, but excessively small σ can lead to overfitting. Slack variables are used to allow for some special samples, thus avoiding the impact of these special samples on the overall model performance. A penalty factor C is introduced to describe the model's tolerance to errors in the training samples. A larger C value indicates a lower tolerance for errors in the training samples, resulting in stronger model fitting accuracy but weaker generalization ability and a higher risk of overfitting. A smaller C value indicates a higher tolerance for errors in the training samples. Although the model has strong generalization ability, the sample fitting accuracy decreases, increasing empirical risk and making it prone to underfitting. To establish the optimal SVM regression model, the key is to find the optimal penalty factor C and kernel function parameter σ.

[0108] SVR is a branch of Support Vector Machine (SVM). Unlike SVM, which aims to maximize the distance from the nearest sample to the hyperplane, SVR aims to minimize the distance from the farthest sample to the hyperplane. D={((x i ,y i ),i=1,2,...,m},

[0109] Given a training set D = {(x i ,y i )}, i = 1, 2, ..., m, where x i y i Let the input vector and the actual value be represented respectively, and establish the objective function:

[0110] f(x)=ω T Φ(x)+b

[0111] Where x = (x1, x2, ..., x m ) T Let f(x) represent m sets of input vectors, Φ(x) represent the mapping function from Euclidean space to Hilbert space, ω represent the weights, and b is the bias value. In traditional problems, the loss is usually calculated using the difference between the output of the model f(x) and the true value y; the loss is zero only when f(x) equals y. However, SVR assumes that f(x) can tolerate a maximum deviation ε from y, i.e., the loss is only calculated when |f(x) - y| > ε. This is equivalent to using f(x) = ω... T Φ(x)+b constructs a gap band with a width of 2ε. If a sample falls within the gap band, the loss is considered to be 0. Figure 5 As shown.

[0112] At this point, the linear hard-spaced SVR problem can be transformed into:

[0113]

[0114] st|(ω·x i +b)-y i |≤ε,i=1,2,…,m

[0115] We need to maximize the interval band and minimize the loss to determine ω and b.

[0116] SVR loss function:

[0117]

[0118]

[0119] This leads to the original problem of constrained optimization:

[0120]

[0121] Where C is the penalty factor, l ε This is the loss function.

[0122] Introducing slack variables ξ≥0 increases the model's tolerance for individual samples, allowing some samples to be outside the interval band. If different degrees of relaxation are allowed on both sides of the hyperplane, two slack variables ξ are introduced. i ≥0, Therefore, the problem can be transformed into a linear soft margin:

[0123]

[0124] At this point, we introduce the Lagrange multiplier α. i ≥0, β i ≥0, Constructing an unconstrained Lagrangian function:

[0125]

[0126] Using the KKT optimality conditions, the original problem is transformed into the dual problem:

[0127]

[0128] Where, α i , β i , They are Lagrange multipliers, ω, b, ξ i , These are the main problem parameters.

[0129] For ω, b, ξ respectively i , Taking the partial derivative and substituting it into the Lagrange function, we get:

[0130]

[0131] At this point, the formula contains only α. i , Then seek For α i , The maximum value of .

[0132]

[0133] Equivalent to:

[0134]

[0135] The above equation is the dual optimization problem, assuming the optimal solution is α. * , The optimal solution ω of the original problem can be obtained. * and b * ,Right now:

[0136]

[0137] The optimal hyperplane can then be represented as:

[0138]

[0139] The classification decision function can be written as:

[0140]

[0141] Considering the nonlinear mapping Φ(x) and the kernel function K(x,y), it is easy to obtain the dual form of nonlinear support vector regression:

[0142]

[0143] K(x,x i )=Φ(x)·Φ(x i )

[0144] Common forms of kernel functions include polynomial functions, Laplace kernel functions, sigmoid kernel functions, and Gaussian radial basis function (RBF) kernel functions. The RBF function is as follows:

[0145]

[0146] The RBF function has the characteristics of wide convergence region, few parameters, and strong versatility, so it is used to build SVR models.

[0147] The Grey-Wolf Optimizer (GWO) algorithm is inspired by the pack hunting behavior of grey wolves. The hunting process of grey wolves includes surrounding the prey, chasing the prey, and attacking the prey.

[0148] During the encirclement, the wolf pack will adjust its position, that is:

[0149] D = |C·X p (t)-X p (t)|

[0150] X(t+1)=X p (t)-A·D

[0151] Where D is the distance between the wolf and its prey, t represents the number of iterations, A and C are coefficients, and X... p Let A(t) be the position of the prey, and let X(t) be the position of the gray wolf. The calculations for A and C are as follows:

[0152] A = 2a·r1-a

[0153] C = 2r²

[0154] Where r1 and r2 are vectors randomly generated between [0, 1], A is a random value in [-a, a], and the convergence factor a decreases linearly from 2 to 0 during the iteration process according to the following formula.

[0155]

[0156] In the formula, t represents the current iteration number, and T is the set maximum iteration number. As the value of a decreases from 2 to 0, the corresponding value of A also changes in the interval [-a, a]. The larger the value of a, the more the gray wolves will move away from the prey, hoping to find a more suitable prey, thus prompting the wolf pack to conduct a global search (||>1). If the value of a is smaller, the gray wolves will move closer to the prey, prompting the wolf pack to conduct a local search (||<1).

[0157] During the hunt, wolves a, β, and δ continuously approach the prey's location, while the position of the remaining wolves ω is constantly adjusted based on the positions of wolves a, β, and δ. The mathematical model for wolves a, β, and δ tracking their prey is as follows:

[0158]

[0159] Among them, D α D β D δ Let X represent the distances between wolves a, β, and δ and other gray wolves, respectively. αX β X δ Let C1, C2, and C3 represent the current positions of wolves a, β, and δ, respectively. C1, C2, and C3 are three randomly generated vectors, and X is the current position of the individual gray wolf.

[0160]

[0161] Where X1, X2, and X3 represent the adjusted positions of wolf ω after being influenced by wolf a, wolf β, and wolf δ, respectively. Here, the average value is taken to obtain the final position of wolf ω:

[0162]

[0163] Grey Wolf's location update method is as follows Figure 6 As shown.

[0164] Using chaotic sequences for population initialization, selection, crossover, and mutation operations affects the entire algorithm process and often achieves better results than pseudo-random numbers. Since the Logistic mapping has good traversal consistency, this paper uses the Logistic mapping to generate the initial position of the wolf pack, as shown in the following equation:

[0165] y j+1 =μy j (1-y j ),y j ∈(0,1),μ∈[0,4]

[0166] In the formula, μ is called the branch parameter. Related studies indicate that the Logistic mapping only exhibits chaotic properties when 3.5699456 < μ ≤ 4. This paper uses the Logistic mapping to map it to the population search space.

[0167] x i,j =lb j,min +y i,j (ub j,max -lb j,min )

[0168] Among them, [lb] j,min ub j,max [This is the position of the gray wolf, x] i,j The upper and lower limits.

[0169] S14. Calculate the root mean square error of the true values ​​of knee joint torque and stiffness calculated by the Lagrange dynamics model and the estimated values ​​of knee joint torque and stiffness output by the support vector machine regression model. Use the root mean square error to evaluate the merits of the chaotic gray wolf algorithm-optimized support vector machine regression prediction model.

[0170] The formula for calculating the root mean square error is:

[0171]

[0172] Among them, X ref The knee joint stiffness is obtained by solving the Euler-Lagrange equations for the knee joint torque or by using the exoskeleton stiffness calculation formula. X pre It is the knee joint torque or stiffness estimated by the support vector machine regression model, and N is the data length of the test sample sequence.

[0173] like Figure 7 The figure shows a comparison between the predicted torque from the test set and the actual knee joint torque calculated from the dynamic equation in the CGWO-SVR simulation experiment. The horizontal axis represents the amount of test set data, and the vertical axis represents the knee joint torque. The solid line represents the torque predicted by the support vector machine regression model, and the dashed line represents the knee joint torque calculated from the dynamic equation. It can be seen that the two are basically consistent with each other, with a small error.

[0174] like Figure 8 The figure shows a comparison between the predicted stiffness of the test set and the actual knee joint stiffness calculated from the dynamic equations in the CGWO-SVR simulation experiment. The horizontal axis represents the amount of test set data, and the vertical axis represents stiffness. The solid line represents the stiffness predicted by the support vector machine regression model, and the dashed line represents the calculated knee joint stiffness. It can be seen that the two are basically consistent, with small errors, and the prediction effect is good.

[0175] S15. During one gait cycle of the wearer wearing the exoskeleton, due to the joint alignment issue, the center distance of the transmission cable 10 changes, and the tension on the transmission cable 10 disappears. Assuming that when the tension sensor 11 on the transmission cable 10 displays a certain tension value, it indicates that the transmission cable 10 is already taut, then at the point when the tension on the transmission cable 10 disappears, rotating the power drive motor 2 by a certain angle can bring the transmission cable 10 back to a taut state, thus achieving transmission. The angles of rotation of the power drive motor 2 at some sampling time points to taut the transmission cable 10 are collected, obtaining several sets of time and power drive motor rotation angle data points. These data points are fitted using spline curves to obtain the motor rotation angle-time equation for compensating for joint alignment. Before each drive of the transmission cable, the power drive motor rotates by a compensation angle in advance to achieve continuous output of driving force on the transmission cable.

[0176] The motor rotation angle-time equation for the compensating joint relative to the center is expressed as:

[0177] θ * =Ψ(t)

[0178] Where, θ * It is the motor rotation angle used to compensate for the tension on the pre-tension rope, and Ψ(t) is an equation about time.

[0179] S16. After estimating the knee joint torque using the support vector machine regression model optimized by the chaotic gray wolf algorithm, an admittance control model with the admittance parameters tuned by the improved whale optimization algorithm is first built on the host computer MATLAB software. After inputting the knee joint torque into the admittance control model, the admittance control model outputs the corresponding knee joint rotation angle.

[0180] The standard WOA (Wide Object Orientation) simulates the unique search methods and hunting mechanisms of humpback whales, mainly including three important stages: hunting prey, bubble netting, and prey searching. In the WOA, the position of each humpback whale represents a potential solution. By continuously updating the whale's position in the solution space, the globally optimal solution is eventually obtained. The specific steps are as follows:

[0181] 1) Encirclement and capture of prey

[0182] The whale's search scope is the global solution space, assuming the current best candidate solution is the target prey or a near-optimal solution. After defining the best candidate solution, the mathematical expression for updating the positions of other whales is as follows:

[0183]

[0184] Where t represents the current iteration number, and It is a coefficient vector. It is a position vector. This represents the position vector of the optimal solution. This indicates the current optimal location of the whale pod and the distance between it and its prey. Among these, and The calculation formula is as follows:

[0185]

[0186] Throughout the iteration process, and It is a random vector in [0,1]. The linear decrease from 2 to 0 can be represented as: t is the current iteration number, and T is the maximum iteration number.

[0187] 2) Predation using bubble nets

[0188] First, a mathematical model of the positions between the whale pod and its prey is established, expressed using a logarithmic spiral equation:

[0189]

[0190] in Let represent the distance between the i-th individual and the current optimal solution, b be the spiral shape parameter, and l be a random number between [0,1]. The calculation formula is as follows:

[0191]

[0192] When whales contract and surround prey, they can choose to either surround the prey or hunt along a spiral path of bubble nets. The probability of updating their position is the same for both hunting methods. The mathematical model for this position update is as follows:

[0193]

[0194] Where p is a random number in [0,1].

[0195] 3) Searching for prey

[0196] Besides using bubble nets for predation, whale pods can update their positions based on their distance from each other and randomly search for prey. Their position update model is as follows:

[0197]

[0198] in, This indicates the location of a randomly selected individual whale. This indicates the distance between the whale's current optimal position and its prey. When, update the location using a random prey search method; when At that time, the location is updated using bubble net hunting.

[0199] In the original WOA algorithm, constant weights are used and do not change during algorithm iterations. Therefore, the convergence speed and global search capability of the algorithm do not change with the search position of the population. In order to improve the convergence speed and global search capability of WOA, and to improve the tracking accuracy and robustness of the exoskeleton motion trajectory after dynamically adjusting the admittance control parameters using WOA, the constant weights are changed to adjustable weights.

[0200] Dynamically adjustable weights are the control factor that balances the algorithm's global search capability and local evolution capability. Larger weights can enhance the algorithm's global search capability, while smaller weights can improve the algorithm's local evolution capability and convergence speed. This is illustrated in the following equation:

[0201]

[0202] Where, ω t It is a dynamic weight, ω max It is the maximum weight, ω min It is the minimum weight. It is a random vector in [0,1]. Adding dynamic weights to the position update formula can improve the convergence speed of the algorithm. The weights are larger in the early stage of the algorithm, which can enhance the global search capability. The weights are smaller in the later stage of the algorithm, and introducing random vectors can avoid getting trapped in local optima.

[0203] With the introduction of dynamic weights, the mathematical model for updating the position of whales during encirclement predation and bubble net predation is as follows:

[0204]

[0205] The location of the random search is updated as follows:

[0206]

[0207] The admittance control formula used in the admittance control model is:

[0208]

[0209] Where e = q r -q a , representing the knee joint rotation angle error, q r q represents the desired rotation angle of the knee joint. a Let M be the desired rotation angle of the knee joint, τ be the estimated torque, and M be the torque. d C d and K d These are the mass, damping, and stiffness parameters of the admittance control object. The magnitude of the error e is used to evaluate the quality of the improved whale optimization algorithm's tuning of the admittance control parameters.

[0210] S17. After estimating the knee joint stiffness by optimizing the support vector machine regression model using the chaotic gray wolf algorithm, the stiffness adjustment motor is controlled by the lower-level microcontroller to drive the stiffness adjustment gear, thereby causing the gear ring of the planetary gear train to rotate to achieve real-time stiffness adjustment. Through closed-loop PID control, real-time nonlinear adjustment of knee joint stiffness is achieved.

[0211] The stiffness control formula for achieving knee joint stiffness adjustment is:

[0212]

[0213] Where e(t) is the error between the predicted stiffness estimated by the support vector machine regression model and the reference stiffness calculated by the Lagrange dynamics model, and K p K i K d These are the proportional, integral, and derivative parameters of the PID controller.

[0214] S18. The knee joint rotation angle output by the admittance control model in step S16 and the compensation angle in step S15 to solve the problem of joint corona are used as inputs for closed-loop PIDA control to control the rotation of the power drive motor to drive the knee joint rotation and assist the wearer in rehabilitation training.

[0215] The PIDA control formula is:

[0216]

[0217] Where e(t) represents the error between the admittance control output angle and the compensation angle and the actual angle, K a It is an acceleration parameter.

[0218] like Figure 9 The image shows an admittance control algorithm for a motor built using the Simulink module in MATLAB software. Except_T is the estimated torque obtained from GWO-SVM, M. d C d and K d The angle of rotation of the driving motor is obtained through optimization using the IWOA algorithm in MATLAB. Figure 10 The figure shows a comparison between the actual angle and the desired angle obtained from the admittance control simulation experiment in Simulink. The solid line represents the actual angle, and the dashed line represents the desired angle. It can be seen that the two curves are basically consistent, indicating that the admittance control parameter tuning effect is good.

[0219] When undergoing rehabilitation training using this knee joint rehabilitation exoskeleton based on variable stiffness elastic actuators, the wearer applies the surface electromyography (EMG) signal acquisition patch after relaxation activities, puts on the variable stiffness knee joint rehabilitation exoskeleton, and activates the exoskeleton switch, motion capture system, and lower-level computer switch. The upper-level PC runs the program for initialization. As the wearer walks continuously, the EMG signal acquisition device acquires and filters the surface EMG signals from the vastus lateralis, long head of the biceps femoris, and medial head of the gastrocnemius muscle. The potentiometer acquires knee joint angle data, and the motion capture system acquires the position, velocity, and acceleration data of various marker points on the lower limbs. This data is then transmitted via a single chip on the lower-level computer. The machine transmits data to a host PC; the host PC processes the data and imports it into a trained Chaotic Gray Wolf Algorithm-Optimized Support Vector Machine Regression (CGWO-SVR) model. The prediction model will predict the knee joint torque and stiffness in real time based on the input; the stiffness control and position control parts of the control module respectively realize the stiffness adjustment and trajectory tracking of the exoskeleton, and transmit the corresponding commands to the corresponding motor controller to realize the wearer's rehabilitation training. This process is repeated until the rehabilitation training is completed; after the rehabilitation training is completed, the host PC is turned off, the wearer removes the knee joint rehabilitation exoskeleton, and the electromyography signal acquisition device patch on the lower limb surface is removed.

[0220] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator, characterized in that, It includes a variable stiffness knee joint rehabilitation exoskeleton body, a power drive module, a stiffness adjustment module, a control module, a signal acquisition module, a data processing module, and a control module based on motion intention recognition; The variable stiffness knee joint rehabilitation exoskeleton body includes a knee joint consisting of a lower leg rod (19), a hinge shaft fixedly set on one side of the top of the lower leg rod (19), and a thigh rod (15) movably hinged to the hinge shaft. The thigh rod (15) is connected to a thigh mounting frame with adjustable position on its side, and the lower leg rod (19) is connected to a lower leg mounting frame with adjustable position on the same side. The thigh mounting frame and the lower leg mounting frame can be fixedly set on the thigh and lower leg of the human body respectively by tightening straps. The power drive module includes a mounting frame fixedly connected to the front side of the thigh mounting frame, a power drive motor (2) fixedly connected to the mounting frame, a disc torsion spring (5) and a planetary gear system respectively rotatably mounted on the mounting frame. One shaft end of the disc torsion spring (5) is connected to the output end of the power drive motor (2) and the other shaft end is connected to the shaft end of the sun gear (3) of the planetary gear system. The shaft ends of the three planet gears (14) of the planetary gear system are connected through the planet carrier. The central rotating shaft of the planet carrier is fixedly connected to the active rope wheel (18), and the driven rope wheel (21) is fixedly connected to the hinge shaft. The active rope wheel (18) and the driven rope wheel (21) are connected through the transmission rope (10). The stiffness adjustment module includes a stiffness adjustment motor (8) fixedly connected to the mounting frame and a stiffness adjustment gear (9) fixedly connected to the output shaft end of the stiffness adjustment motor (8). The stiffness adjustment gear (9) meshes with the gear ring (13) of the planetary gear system. The signal acquisition module includes a surface electromyography (EMG) signal acquisition device, a potentiometer (12), a tension sensor (11), and a motion capture system. The EMG signal acquisition device is used to acquire surface EMG signals of the lower limbs. The potentiometer (12) is mounted on the hinge shaft and is used to acquire the relative rotation angle between the lower leg rod (19) and the thigh rod (15). The tension sensor (11) is set on the transmission rope (10) and is used to measure the tension on the transmission rope (10). The motion capture system is used to acquire the position, velocity, and acceleration of each marker point of the lower limbs of the human body. The control method for the knee joint rehabilitation exoskeleton includes the following steps: S10, The signal acquisition module collects motion data of the wearer while walking, including surface electromyography signals of the leg collected by the surface electromyography signal acquisition device and knee joint rotation angle collected by the potentiometer; S11. Preprocess the collected knee joint rotation angle data and surface electromyography signal data respectively; S12. Establish a Lagrange dynamic model of the lower limb two-linkage system through the thigh and lower leg links, obtain the motion information of the lower limb markers through the motion capture system, and solve the knee joint torque through the Euler-Lagrange equation in the data processing software. S13. Construct a chaotic gray wolf algorithm to optimize the support vector machine regression model for motion intention recognition: Build the corresponding model on the data processing software, collect knee joint angle and surface electromyography signal data of the wearer in the walking state after wearing the exoskeleton, use the first 70% of the collected and processed data as the input of the training set, and use the knee joint torque and stiffness values ​​calculated by the model as the output of the training set to train the prediction model, use the last 30% of the collected and processed data as the input of the test set, and use the knee joint torque and stiffness values ​​calculated by the model as the output of the test set. S14. Calculate the root mean square error of the true values ​​of knee joint torque and stiffness calculated by the Lagrange dynamics model and the estimated values ​​of knee joint torque and stiffness output by the support vector machine regression model. Use the root mean square error to evaluate the merits of the chaotic gray wolf algorithm-optimized support vector machine regression prediction model. S15. During one gait cycle of the wearer wearing the exoskeleton, the angle of rotation of the power drive motor at some sampling time points is collected to tighten the transmission rope. Several sets of data points of time and power drive motor rotation angle are obtained. These data points are fitted with spline curves to obtain the motor rotation angle-time equation to compensate for the joint centering. Before each drive of the transmission rope by the power drive motor, the power drive motor is rotated in advance by the compensation angle to achieve continuous output of driving force on the transmission rope. S16. After estimating the knee joint torque by optimizing the support vector machine regression model using the chaotic gray wolf algorithm, an admittance control model with the admittance parameters tuned by the improved whale optimization algorithm is first built on the host computer MATLAB software. After inputting the knee joint torque into the admittance control model, the admittance control model outputs the corresponding knee joint rotation angle. S17. After the knee joint stiffness is estimated by optimizing the support vector machine regression model using the chaotic gray wolf algorithm, the stiffness adjustment motor is controlled by the lower-level microcontroller to drive the stiffness adjustment gear to rotate, thereby rotating the gear ring of the planetary gear train to achieve real-time stiffness adjustment. Real-time nonlinear adjustment of knee joint stiffness is achieved through closed-loop PID control. S18. The knee joint rotation angle output by the admittance control model in step S16 and the compensation angle in step S15 to solve the problem of joint corona are used as inputs for closed-loop PIDA control to control the rotation of the power drive motor to drive the knee joint rotation and assist the wearer in rehabilitation training.

2. The knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator according to claim 1, characterized in that: In step S11, the specific method for acquiring and filtering the surface electromyography signal data is as follows: The surface electromyography (EMG) signal acquisition device patches were applied to the vastus lateralis, long head of biceps femoris, and medial head of gastrocnemius muscles of the right leg, respectively. The EMG signal acquisition device was used to acquire the EMG signals of the leg muscles of the wearer walking at a speed of 2m / s for 1 minute at a rate of 2000Hz using four channels. Then, a 50Hz band-stop filter and a 20Hz-500Hz Butterworth bandpass filter were used to remove noise from the signals. The filtered electromyography signal was then divided into windows of 30 sampling points, and the root mean square (RMS) of the time-domain features was extracted. The formula for calculating RMS is as follows: ； in, This represents the intensity value of the electromyographic signal. The number of electromyographic signals within a window.

3. The knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator according to claim 1, characterized in that: In step S12, the formula used to calculate the knee joint torque in the lower limb two-link Lagrange dynamics model is: ； in, For knee joint torque, Let the moment of inertia of the thigh member be denoted as . Let be the moment of inertia of the lower leg member. For the mass of the thigh member, For the mass of the lower leg member, The angle between the line connecting the center of mass of the thigh member and the knee joint and the horizontal plane. Let be the angle between the line connecting the center of mass of the lower leg member and the knee joint and the horizontal plane. Let be the distance from the center of mass of the thigh member to the center of rotation of the knee joint. The length of the thigh member, The length of the lower leg member. The horizontal component of the force acting on the knee joint. The vertical component of the force acting on the knee joint. Let this be the horizontal position of the center of mass of the thigh member pair. The vertical position of the center of mass of the thigh member pair. Let be the horizontal position of the center of mass of the lower leg member. The vertical position of the center of mass of the lower leg member.

4. The knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator according to claim 1, characterized in that: In step S14, the formula for calculating the root mean square error is: ； in, The knee joint stiffness is obtained by solving the Euler-Lagrange equations for the knee joint torque or by using the exoskeleton stiffness calculation formula. It is the knee joint torque or stiffness estimated by the support vector machine regression model. It is the data length of the test sample sequence.

5. The knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator according to claim 1, characterized in that: In step S16, the admittance control model uses the following admittance control formula: ； in, This indicates the error in the knee joint rotation angle. The desired rotation angle of the knee joint. This represents the desired rotation angle of the knee joint. To estimate the torque, , and These are the mass, damping, and stiffness parameters of the admittance control object.

6. The knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator according to claim 1, characterized in that: In step S17, the stiffness control formula for adjusting knee joint stiffness is: ； in, The error between the predicted stiffness estimated by the support vector machine regression model and the reference stiffness calculated by the Lagrange dynamics model. , , These are the proportional, integral, and derivative parameters of the PID controller.

7. The knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator according to any one of claims 1 to 6, characterized in that: The control module includes a host PC, a slave microcontroller, a first encoder (1) mounted on the power drive motor (2), and a second encoder (7) mounted on the stiffness adjustment motor (8). The host PC has built-in data processing software and communicates with the slave microcontroller via serial port to receive and process data. The slave microcontroller communicates with the power drive motor (2) and the stiffness adjustment motor (8) via CAN bus and collects data collected by the signal acquisition module. The aforementioned motion intention recognition refers to estimating the torque and stiffness of the knee joint by collecting surface electromyography signals and knee joint angles, optimizing the support vector machine regression model by establishing a chaotic gray wolf algorithm, using the collected signals as the input of the model, and using the torque and stiffness calculated by inverse dynamics as the output of the model, and training the support vector machine regression model by constructing training and testing sets; The control module refers to tuning the admittance parameters of the admittance control through the improved whale algorithm, using the torque estimated by the support vector machine regression model as the input of the admittance control, and the knee joint angle as the output of the admittance control. The admittance parameters are adjusted by evaluating the error between the output angle of the admittance control and the actual angle of the knee joint. At the same time, the transmission rope (10) is tightened by compensating a certain angle through the power drive motor (2). A closed-loop PIDA is constructed with the compensation angle and the output angle of the admittance control to perform compliant control on the power drive motor (2).

8. The knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator according to claim 7, characterized in that: A thigh connecting plate (17) is provided between the thigh rod (15) and the thigh connecting frame, and a calf connecting plate (20) is provided between the calf rod (19) and the calf connecting frame. Several evenly distributed bolt holes are provided on the side walls of the thigh rod (15), the calf rod (19), the thigh connecting frame and the calf connecting frame. The two sides of the thigh connecting plate (17) are connected to the thigh rod (15) and the thigh connecting frame by bolts, respectively. The two sides of the calf connecting plate (20) are connected to the calf rod (19) and the calf connecting frame by bolts, respectively. By adjusting the assembly position of the bolts and bolt holes, the installation position of the thigh mounting frame on the thigh rod (15) and the installation position of the calf mounting frame on the calf rod (19) can be adjusted respectively.

9. The knee joint rehabilitation exoskeleton based on a variable stiffness elastic actuator according to claim 7, characterized in that: The stiffness of the planetary frame is expressed as the output stiffness of the knee rehabilitation exoskeleton based on a variable stiffness elastic actuator, and is: ； in, For the output stiffness of the exoskeleton, The torque acting on the planet carrier, The rotation angle of the planet carrier; Based on the characteristics of the gear train transmission, the torque and angle relationship between the disc torsion spring and the planetary carrier can be derived. Furthermore, the relationship between the output stiffness of the exoskeleton and the stiffness of the disc torsion spring can be obtained as follows: ； in, For the stiffness of the torsion spring, This refers to the transmission ratio between the sun gear and the planet carrier in a planetary gear train. The calculation formula is: ； in, , , These represent the number of teeth on the sun gear, ring gear, and stiffness adjusting gear, respectively, with the number of teeth on the inner and outer sides of the ring gear being equal. , These are the rotational speeds of the sun gear and the stiffness adjustment gear, respectively. Therefore, the output stiffness of the exoskeleton is: 。

Images