Upper limb rehabilitation robot fuzzy PID control method based on myoelectricity feedback

Abstract

The invention discloses a fuzzy PID (Proportion Integration Differentiation) control method for an upper limb rehabilitation robot based on myoelectricity feedback. The fuzzy PID control method comprises the following steps: step 1, sampling surface myoelectricity signal data; 2, performing signal processing on the surface electromyogram signal data; 3, performing feature extraction on the surface electromyogram signal data after signal processing through a feature selection method so as to obtain a key feature set; 4, creating a fusion model, training the fusion model by using the fatigue state key feature set, and determining an optimal fusion model; detecting the fatigue state of the person to be detected in real time; and step 5, obtaining a fatigue state corresponding to the currently obtained surface electromyogram signal of the patient in real time, and dynamically adjusting a corresponding PID parameter in combination with a self-adaptive fuzzy PID controller so as to adapt to different fatigue states. Proportional, integral and differential parameters can be dynamically adjusted according to the real-time fatigue state of a patient; the method adapts to different fatigue levels of patients, and control flexibility and response speed are improved.

Description

Technical Field

[0001] The invention relates to the technical field of rehabilitation training, and in particular to a fuzzy PID control method for an upper limb rehabilitation robot based on electromyography feedback. Background Art

[0002] With the aging of the population, changes in lifestyle and eating habits, the incidence of stroke is increasing year by year, and it affects more and more young people. Stroke may cause brain damage, which in turn causes movement disorders in a large number of patients, which is one of the main causes of long-term disability. In particular, the motor function of the upper limbs is greatly affected, making it impossible for patients to complete daily life activities independently. It is crucial to restore the functional ability of the upper limbs of disabled patients because individuals need to use their upper limbs when performing various daily activities. According to the neuroplasticity hypothesis, continuous and high-intensity training can change the structure and function of the nervous system, restore the connection between the affected limbs and the damaged central nervous system of the brain, and thus regain control of the limb motor function. There are two main control methods for upper limb rehabilitation robots, one is admittance impedance control and the other is PID control.

[0003] Specifically, Addisie et al. observed that the robotic arm needs to be precisely controlled during the treatment of stroke patients, and designed a rehabilitation robot based on fuzzy PID control, which enables the rehabilitation robot to drive the patient's elbow and forearm according to the patient's pain level. Modares et al. proposed a novel control strategy that uses machine learning methods to find the optimal impedance parameters and minimize the tracking error of the impedance control system. Luo et al. proposed an adaptive impedance controller that determines the optimal robot stiffness based on an estimate of the patient's stiffness. Experiments show that the adaptive impedance controller brings more significant rehabilitation effects than the constant impedance controller.

[0004] To overcome uncertainty and disturbance, Alireza et al. designed an adaptive fuzzy sliding mode controller to reduce the chattering effect and compensate for the patient's variable force. Experimental results show that the accuracy of tracking the target trajectory is improved by about 80% due to adaptability. Xu et al. proposed a newly developed adaptive impedance controller based on evolutionary dynamic fuzzy neural network, which can adjust the expected impedance between the robot and the damaged limb in real time according to the physical recovery of the limb. The robot can also perform individual rehabilitation for each knuckle, which is particularly important for the rehabilitation of fingers. In addition, Alireza et al. proposed a fuzzy sliding mode controller design method for robots due to model uncertainty, disturbance and chattering in the system. Rehabilitation robots are widely used by stroke patients during rehabilitation.

[0005] However, the disadvantage of traditional control methods is that they cannot be dynamically adjusted according to the patient's fatigue state, which leads to unsatisfactory training results or aggravates patient fatigue. It can be said that managing fatigue during training is crucial to patient safety and rehabilitation effects. In this process, human-computer interaction is particularly important, which can provide patients with safe, effective and economical rehabilitation training. High-intensity exercise can easily cause patients to feel tired or damage their musculoskeletal system. However, low-intensity exercise is often not enough to achieve the desired effect. At the same time, muscle fatigue not only reflects exercise intensity, but also plays a key role in adjusting training plans and preventing excessive fatigue and muscle damage.

[0006] Therefore, the control design of the rehabilitation robot should be able to adjust the training intensity according to the patient's fatigue level and then customize a personalized training plan, which is a necessary condition for achieving the best rehabilitation effect. Summary of the invention

[0007] The purpose of the present invention is to provide a fuzzy PID control method for an upper limb rehabilitation robot based on electromyography feedback, aiming to solve the technical problem that the traditional control method cannot be dynamically adjusted according to the patient's fatigue state, thereby resulting in unsatisfactory training effect or aggravating the patient's fatigue.

[0008] In order to solve the above problems, the present invention provides a fuzzy PID control method for an upper limb rehabilitation robot based on electromyographic feedback, and the corresponding technical solution includes:

[0009] Step 1, sampling surface electromyography signal data;

[0010] Step 2, performing signal processing on the aforementioned surface electromyography signal sampling data;

[0011] Step 3: According to the importance score, feature extraction is performed on the surface electromyography signal sampling data after the above signal processing by a feature selection method, so as to obtain a key feature set;

[0012] Step 4: Create a fusion model, use the fatigue state key feature set to train the fusion model and determine the best fusion model; thereby detecting the fatigue state of the subject in real time and completing the classification of sEMG signals;

[0013] Step 5: According to the aforementioned optimal fusion model, the fatigue state corresponding to the patient's surface electromyography (sEMG) signal currently acquired in real time is obtained, and the corresponding PID parameters are dynamically adjusted in combination with the adaptive fuzzy PID controller based on electromyography feedback to adapt to different fatigue states.

[0014] Furthermore, the steps of processing the aforementioned surface electromyography signal data include: (1) rectifying and filtering the aforementioned surface electromyography signal data, specifically removing the DC component, bandpass filtering and full-wave rectification of the aforementioned surface electromyography signal data in sequence; and (2) performing feature extraction processing on the aforementioned surface electromyography signal data through a sliding window method, thereby converting the continuous original signal into a discrete signal.

[0015] Furthermore, according to the importance score, feature extraction is performed on the surface electromyography signal data after the above-mentioned signal processing through a feature selection method, so as to obtain the processing steps of the key feature set of the fatigue state, including: the feature selection method adopts the SelectFromModel method, thereby using the machine learning model to evaluate the importance of each feature, and select the key features that need to be retained according to the importance score.

[0016] Furthermore, the specific steps of the aforementioned step 3 include: (1) selecting a training model and training according to the selected training model through the feature set obtained above, so as to provide feature importance scores through the trained training model; (2) setting a threshold and selecting key features, that is, setting a threshold according to experience or automatically selecting a suitable threshold by SelectFromModel, so that those features with importance scores higher than this threshold are selected and retained, generating a new data set that only contains those key features that are considered important.

[0017] Furthermore, the fusion model creation method includes selecting multiple basic classifier combinations, and forming multiple evaluation strategies through multiple model fusion technologies, and finally determining the best strategy through a weighted average base learner, detecting the fatigue state of the current person to be tested, and giving a detection result.

[0018] Furthermore, the types of basic classifier combinations include MLP_KNN, LightGBM_LR_MLP, RBF_KNN and SVM_RBF_KNN; model fusion techniques include stacking, voting and averaging.

[0019] Furthermore, the soft voting model SoftVoting is used to make decisions by weighted averaging the prediction probabilities of the base learners. The corresponding model formula is:

[0020]

[0021] Among them, P(class i ) is the final prediction result, i is the category number, i.e., the fatigue level category, N is the number of base learners, P j (class i ) is the predicted probability of the j-th base learner for the i-th class.

[0022] Furthermore, the output formula of the adaptive fuzzy PID controller is:

[0023]

[0024] Among them, K p (e, Δe, f): proportional gain dynamically adjusted according to the system error e(t), error change rate Δe(t) and fatigue state f (no fatigue, mild fatigue, severe fatigue); K i (e, Δe, f): integral gain dynamically adjusted according to the same input as above; K d (e, Δe, f): differential gain dynamically adjusted according to the same input as above; f: fatigue state of EMG signal feedback.

[0025] Implementing the embodiments of the present invention will have the following beneficial effects:

[0026] The present invention can dynamically adjust the proportional, integral and differential parameters according to the patient's real-time fatigue state; because it can cope with complex rehabilitation training scenarios, adapt to the different fatigue levels of patients, and improve the flexibility and response speed of control. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0028] in:

[0029] Figure 1 It is a flow chart of the overall steps of the present invention;

[0030] Figure 2 The sampling step design diagram corresponding to the surface electromyography signal data of the present invention;

[0031] Figure 3 A flow chart of signal processing for surface electromyography signal data of the present invention;

[0032] Figure 4 Importance feature graph of 9 features selected by SelectFromModel of the present invention;

[0033] Figure 5 This is a rendering of the model fusion technology of the present invention;

[0034] Figure 6 It is the control principle diagram of the basic PID controller of the present invention;

[0035] Figure 7 It is the control principle diagram of the fuzzy PID controller of the present invention;

[0036] Figure 8 It is the control principle diagram of the adaptive fuzzy PID controller of the present invention;

[0037] Fig. 9 The simulation results of the end trajectory of the upper limb rehabilitation robot under different fatigue states by the three controllers of the present invention;

[0038] Fig.10 These are the simulation results of the angle control of the shoulder joint of the upper limb rehabilitation robot by the three controllers of the present invention under different fatigue states. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0041] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0042] like Figure 1 , the present invention proposes a fuzzy PID control method for an upper limb rehabilitation robot based on electromyography feedback:

[0043] Step 1, sampling surface electromyography signal data;

[0044] Step 2, performing signal processing on the aforementioned surface electromyography signal sampling data;

[0045] Step 3: According to the importance score, feature extraction is performed on the surface electromyography signal sampling data after the above signal processing by a feature selection method, so as to obtain a key feature set;

[0046] Step 4: Create a fusion model, use the fatigue state key feature set to train the fusion model and determine the best fusion model; thereby detecting the fatigue state of the subject in real time and completing the classification of sEMG signals;

[0047] Step 5: According to the aforementioned optimal fusion model, the fatigue state corresponding to the patient's surface electromyography (sEMG) signal currently acquired in real time is obtained, and the corresponding PID parameters are dynamically adjusted in combination with the adaptive fuzzy PID controller based on electromyography feedback to adapt to different fatigue states.

[0048] The present invention monitors the patient's muscle activity in real time through sEMG signals, and uses the model fusion technology in machine learning to divide the fatigue state into three categories: no fatigue, mild fatigue and severe fatigue; according to different fatigue levels, the fuzzy PID control technology is used to enable it to adaptively adjust the PID control parameters to ensure precise control of the motion trajectory and system stability during training.

[0049] In a further embodiment, the aforementioned step 1 includes: sampling surface electromyography signal data and establishing a sampling database; due to the development of information technology and biometric data acquisition technology, the biomedical field has accumulated a wealth of human data. Human biopotential data is crucial in medicine and rehabilitation because it reflects the changes in the potential and polarity of life activities. Therefore, the potential record of muscle activity can be determined by objectively evaluating signals of muscle energy and fatigue, such as surface electromyography signals-sEMG signals. sEMG helps to understand someone's fatigue level and makes rehabilitation safer. Therefore, the sampling signal selects the surface electromyography signal to detect fatigue information.

[0050] More specifically, in the embodiment, in upper limb rehabilitation robot training, the characteristic parameters of the electromyographic signal are extracted through the dumbbell lifting action to evaluate and monitor the fatigue level and training effect of the sampled person. The specific sampling data collection process is: using multivariate and multi-level factor design, controlling variables such as dumbbell weight (2kg to 8kg), lifting frequency (2 to 10 times / minute), and lifting time (5 minutes), observing the changes in electromyographic signals and other physiological signals, in order to explore the effects of different training conditions on fatigue level and training effect. Figure 2As shown in the figure, first, the physiological signals of the sampled person in a relaxed and calm state are collected; then, several groups of electromyographic signal sampling data of the sampled person performing dumbbell lifting training are collected and the data files are saved, that is, the sampled person lifts dumbbells of different weights, and the electromyography (sEMG) signal of his biceps brachii is recorded through the surface electromyography system. After each group of experiments, the sampled person scores the difficulty of the training, and the experimenter records the completion of the training task. Through the above process, the real upper limb movement scene is simulated to increase the reliability and effectiveness of the experiment.

[0051] Further embodiments, such as Figure 3 As shown, the aforementioned step 2, performing signal processing on the aforementioned surface electromyography signal data;

[0052] More specifically, the steps of processing the surface electromyography signal data include: (1) rectifying and filtering the surface electromyography signal data, that is, in order to improve the signal quality and reduce the noise, the surface electromyography signal data is sequentially subjected to the processes of removing the DC component, bandpass filtering and full-wave rectification, thereby enhancing the signal availability through this step and laying a solid foundation for the subsequent data analysis. (2) feature extraction processing is performed on the surface electromyography signal data by the sliding window method, thereby converting the continuous original signal into a discrete signal that is more suitable for analysis and classification, that is, dividing the continuous signal into a series of fixed-length time periods or "windows", more specifically, dividing the signal after rectification and filtering into 5-second windows, with 50% overlap between two adjacent windows - the latter window starts 2.5 seconds before the end of the previous window. This segmentation method helps feature extraction in the process of sEMG signal processing, converting continuous raw signals into discrete segments that are more suitable for analysis and classification. Through this processing, more training samples can be obtained from the same raw data, window overlap can better capture the instantaneous changes of features, and the signal is segmented with different starting points, which helps to reduce the deviation that may be introduced by selecting a specific starting point, thereby improving the accuracy of subsequent steps (such as classification). Finally, after the above data processing process, the feature set shown in Table 1 was obtained, which provides a reliable basis for subsequent analysis.

[0053] Table 1 Time domain characteristics and calculation formula

[0054]

[0055] In a further embodiment, in the aforementioned step 3, according to the importance score, feature extraction is performed on the surface electromyography signal data after the aforementioned signal processing by a feature selection method, so as to obtain a set of key features of the fatigue state; more specifically, the feature selection method can adopt the SelectFromModel method, thereby using a machine learning model to evaluate the importance of each feature, and select the key features that need to be retained according to the importance score.

[0056] In machine learning, feature selection is a key step in building efficient and accurate models. Too many features will increase model complexity, leading to overfitting, long training time, waste of computing resources, and reduce model efficiency, scalability and interpretability. Therefore, feature selection is crucial. More specifically, the aforementioned step 3 includes: (1) selecting a training model and training the selected training model using the feature set obtained above, thereby providing a feature importance score through the trained training model. The training model can be a random forest model or a linear model (Logistic regression with L1 regularization, etc.); for the random forest model, the feature importance is obtained by looking at the number of times each feature is used to split nodes in all trees or by reducing the sum of impurities; for the linear model, it is measured by the absolute value of the coefficient. (2) Setting a threshold and selecting key features, that is, setting a threshold based on experience or automatically selecting a suitable threshold by SelectFromModel, so that those features with importance scores higher than this threshold are selected and retained, generating a new data set that only contains those key features that are considered important.

[0057] More specifically, this application uses SelectFromModel to select 9 features. Figure 4 The importance of the features is shown in Figure 1. Among them, F Mean Frequency (0.153) and F Median Frequency (0.132) are the most important features, which contribute the most to the model prediction and reflect the inherent patterns in the data. Although the importance of other features is lower than 0.13, they may still have some impact on the model or provide additional information, such as DP Max (μV) and DN Min (μV), which represent the maximum voltage in the positive and negative directions, respectively, and are related to the volatility or peak of the data. DP Mean (μV) and DN Variances (μV2) represent the average voltage and variance in the positive and negative directions, respectively, and are related to the stability or distribution of the data. DN RMS (μV) and DP Min (μV) represent the root mean square voltage in the negative direction and the minimum voltage in the positive direction, respectively, and are related to the intensity or range of the data. DP Max (μV) is an important feature because it can help us understand the working status and potential of muscles. DP Max (μV) can be used to assess muscle strength, endurance, fatigue, injury, and recovery. DP Max (μV) can also guide exercise, rehabilitation, and treatment plans to improve muscle efficiency and health.

[0058] In a further embodiment, the aforementioned step 4, creates a fusion model, uses a set of key features of fatigue state to train the fusion model and determine the best fusion model; thereby detecting the fatigue state of the subject to be tested in real time and completing the classification of sEMG signals; the fusion model creation method includes selecting a combination of multiple basic classifiers, and forming a variety of evaluation strategies through multiple model fusion techniques, and finally determining the best strategy through a weighted average base learner, detecting the fatigue state of the current subject to be tested, and giving a detection result.

[0059] More specifically, the types of basic classifier combinations include MLP_KNN, LightGBM_LR_MLP, RBF_KNN and SVM_RBF_KNN, MLP: Multilayer Perceptron, KNN: K-Nearest Neighbors, SVM: Support Vector Machine, LightGBM: Light Gradient Boosting Machine, LR: Logistic Regression, RBF: Radial Basis Function; model fusion techniques include stacking, voting and averaging, which aim to comprehensively consider the performance, diversity and complementarity of the model, and improve the classification performance by adopting multiple basic classifiers and different model fusion techniques. 12 strategies have been formed to improve the accuracy and robustness of the model, and then determine which fusion strategy can provide the best performance. More specifically, 5 cross-validations are used to evaluate the performance indicators such as accuracy, recall, F1 value, etc. of each fusion model. Figure 5 As shown in the figure, the voting method performs best among all fusion methods. Specifically, the voting method performs well in various indicators such as accuracy, recall, F1 value and precision. In all evaluations, the performance of the voting method is significantly better than the stacking and averaging methods. The voting method combined with the LightGBM_RBF_MLP model achieved the highest scores in all evaluation indicators, and the accuracy, recall, F1 score and accuracy of the fused model were 0.93, 0.93, 0.94 and 0.93 respectively.

[0060] More specifically, a soft voting model can also be constructed to determine which fusion strategy is used for fatigue state detection, that is, the soft voting model is used to make a decision by weighted averaging the prediction probability of the base learner. The corresponding model formula is:

[0061]

[0062] Among them, P(class i) is the final prediction result, i is the category number, i.e., the fatigue level category, category refers to the fatigue state, including no fatigue (NF), mild fatigue (MF), severe fatigue (SF), N is the number of base learners, P j (class i ) is the predicted probability of the j-th base learner for the i-th class.

[0063] Through the soft voting model method, the base classifiers are trained---four base classifier combinations are trained separately: MLP_KNN, LightGBM_LR_MLP, RBF_KNN and SVM_RBF_KNN. For each combination, ensure that each base classifier can output the predicted probability-the probability of belonging to each category (for example, using the predict_proba method); these probabilities are weighted averaged according to the weight. If all base classifiers have the same weight, it is a simple average; weights can also be assigned according to the performance of the base classifier (such as the cross-validation score). Classifiers with better performance can be given higher weights; the category with the highest average probability is selected as the final prediction result. Use 5-fold cross validation to evaluate the performance of the SoftVoting method and calculate and record performance indicators such as accuracy, recall, and F1 value. The adopted soft voting fusion model shows high accuracy, recall, precision, and F1 value in fatigue detection, demonstrating its robustness in complex sEMG signal processing, and providing a reliable basis for precise control of rehabilitation robots. This method not only improves the effect of rehabilitation training, but also provides an important reference for the design and optimization of future intelligent rehabilitation systems and has broad application prospects.

[0064] A further embodiment obtains the fatigue state corresponding to the patient's surface electromyography (sEMG) signal currently acquired in real time according to the aforementioned optimal fusion model, and dynamically adjusts the corresponding PID parameters in combination with an adaptive fuzzy PID controller based on electromyography feedback to adapt to different fatigue states.

[0065] More specifically, the basic PID controller control principle is:

[0066] The relationship between the output u(t) of the PID controller and the control error e(t) can be expressed as follows:

[0067]

[0068] Where u(t) is the PID controller output, e(t) is the control error, which is defined as the difference between the desired value and the actual output, i.e., e(t) = r(t) - y(t); where r(t) is the desired output and y(t) is the actual output. p is the proportional gain, the direct effect of the regulation error; K i is the integral gain, eliminating steady-state error; Kd It is the differential gain, which predicts the trend of error change and reduces the overshoot of the system.

[0069] Specific application in rehabilitation robot control, combined with Figure 6 In this figure, r(t) represents the expected value (set value, such as target position or speed), and y(t) represents the actual output of the system (such as the current position signal of the robot); the PID controller is responsible for calculating the control signal u(t), based on the error e(t) = r(t)-y(t), the controlled object (rehabilitation robot) receives the control signal u(t), and the output signal y(t) is fed back to the controller to form a closed loop.

[0070] Under the control principle of PID controller, the output formula of fuzzy PID controller is similar to that of traditional PID, but its proportional, integral and differential parameters are no longer fixed, but are dynamically adjusted according to fuzzy logic rules. Figure 7 The output u(t) of the fuzzy PID controller can be expressed as:

[0071]

[0072] Among them, K p (e) K i (e) K d (e) are the proportional, integral and differential gains obtained through fuzzy reasoning, which are functions of the error e(t) and the error change rate Δe(t), respectively; they change dynamically according to the current state of the rehabilitation robot control system (such as error and error change rate); e(t) is the current error; Δe(t) is the error change rate, defined as

[0073] In rehabilitation robots, the physiological state of patients is uncertain and nonlinear, and traditional PID control design is difficult to adapt. Since the fuzzy PID controller can flexibly respond to changes in the patient's state by adjusting the PID parameters in real time, it is possible to consider designing a controller formed by combining adaptive fuzzy PID control technology with fatigue state detection. When the patient is fatigued, the controller will automatically reduce K p and K d value, thereby reducing excessive load on the patient and ensuring that the entire system responds more gently and adaptively.

[0074] Among them, Figure 8 The adaptive fuzzy PID controller is responsible for fuzzifying the input variables and generating fuzzy reasoning results; the PID controller dynamically adjusts the control parameter K p , K i , K d ; The output signal y(t) represents the actual output of the rehabilitation robot system, such as joint angle or velocity.

[0075] The output formula of the adaptive fuzzy PID controller is:

[0076]

[0077] Among them, K p (e, Δe, f): proportional gain dynamically adjusted according to the system error e(t), error change rate Δe(t) and fatigue state f (no fatigue, mild fatigue, severe fatigue); K i (e, Δe, f): integral gain dynamically adjusted according to the same input as above; K d (e, Δe, f): differential gain dynamically adjusted according to the same input as above; f: fatigue state fed back by electromyographic signal. By collecting and analyzing the electromyographic signal of the patient, the fatigue state is divided into non-fatigue (NF), mild fatigue (MF), and severe fatigue (SF). The fuzzy PID rule table is shown in Table 2-4.

[0078] Table 2 Fuzzy rule table of proportional gain Kp based on electromyographic feedback

[0079]

[0080] When the muscles are in a state of mild fatigue, the adaptive fuzzy PID controller needs to gradually reduce Kp to reduce excessive response and appropriately increase the integral action (Ki) to compensate for the response delay caused by fatigue.

[0081] Table 3 Proportional gain K based on myoelectric feedback i Fuzzy rule table

[0082]

[0083] When the muscles are severely fatigued, the adaptive fuzzy PID controller must further reduce Kp and significantly increase the integral coefficient Ki to extend the memory characteristics of the controller, thereby better compensating for the impact of muscle fatigue on control accuracy. In addition, Kd (differential action) also needs to be appropriately increased to reduce system oscillation.

[0084] Table 4 Proportional gain K based on myoelectric feedback d Fuzzy rule table

[0085]

[0086]

[0087] The fuzzy PID rule table based on electromyographic feedback can dynamically adjust the proportional, integral and differential gains according to the user's fatigue level by adding the input of fatigue status. In the non-fatigue state, the system can maintain a high control force; in the state of mild fatigue and severe fatigue, the system will gradually reduce the control force to ensure the safety and comfort of the human body. Through the above three tables, we can see that each gain (K p , K i , K d ) How to make corresponding adjustments based on the electromyographic feedback information so that the fuzzy PID controller can adapt to changes in the human body's fatigue state.

[0088] In order to verify the performance of this application, a simulation experiment was conducted on the controller formed by the adaptive fuzzy PID control method based on electromyographic feedback to verify its effectiveness in the control of rehabilitation robots. The experiment simulated the influence of different patients' muscle fatigue states on the motion control of rehabilitation robots through the Simulink simulation environment, and compared the simulation results of fixed PID controller and fuzzy PID controller to illustrate the advantages of adaptive fuzzy PID controller in dealing with system nonlinearity and changes in patient fatigue states.

[0089] Simulation experiment settings:

[0090] This experiment verifies the performance of the adaptive fuzzy PID controller based on electromyography feedback in the motion control of the rehabilitation robot through simulation experiments.

[0091] The experimental design includes the following key parts:

[0092] The rehabilitation robot system is simulated using a second-order linear model, and its transfer function is expressed as follows:

[0093]

[0094] Among them, K is the system gain; ζ is the damping ratio, which determines the oscillation characteristics of the system; ω n is the natural frequency of the system, which determines the response speed of the system. This model represents the dynamic characteristics of the joint motion control of the rehabilitation robot. The input control signal is force or torque, and the output is joint position or angular velocity.

[0095] The performance of three controllers is compared: first, a fixed PID controller whose PID parameters are set manually and whose proportional gain K p , integral gain K i , differential gain K dAll remain fixed. The second is the fuzzy PID controller, which dynamically adjusts the PID parameters through fuzzy control rules, but does not rely on the patient's electromyographic signals. The last is the adaptive fuzzy PID controller based on electromyographic feedback, which dynamically adjusts the PID parameters to adapt to different fatigue states by obtaining the patient's surface electromyographic signal (sEMG) feedback in real time.

[0096] During the simulation, three types of patient muscle fatigue states are simulated: Non-fatigue state: muscle activity is normal, the system is expected to respond quickly and overshoot is small. Mild fatigue state means that muscle activity is weakened, and the controller needs to appropriately reduce the proportional gain K p , increase the integral K i To compensate for the slow movement. Severe fatigue means that the muscle fatigue is obvious, and the controller needs to further reduce the proportional gain K p , enhance the integral and differential effects to maintain the stability of the system. By setting different input signals (such as step signals and sinusoidal signals), the dynamic response performance of the three controllers is tested.

[0097] Simulation results:

[0098] Fig. 9 The simulation results of the upper limb rehabilitation robot terminal trajectory under different fatigue states are shown in Figure 1. The trajectory tracking performance of the fixed PID controller, fuzzy PID controller and adaptive fuzzy PID controller based on electromyography feedback are compared under the non-fatigue state, mild fatigue state and severe fatigue state.

[0099] As can be seen from the figure, in the non-fatigue state, the trajectory tracking effects of the three controllers are similar, and all can follow the expected trajectory well (black dotted line). However, the adaptive fuzzy PID controller based on EMG feedback (red curve) shows higher accuracy and the smallest deviation from the expected trajectory. In the mild fatigue state, the tracking error of the fixed PID controller (blue curve) begins to increase, especially in the later stage of movement. The fuzzy PID controller (yellow curve) has improved, but the adaptive fuzzy PID controller based on EMG feedback still maintains the best tracking effect with a small error, which can better adapt to the influence of mild fatigue. When entering the severe fatigue state, the error of the fixed PID controller increases further, and the deviation from the expected trajectory is obvious. Although the fuzzy PID controller has improved compared with the fixed PID, it still has a certain error. In contrast, the adaptive fuzzy PID controller based on EMG feedback can dynamically adjust the control parameters according to the real-time EMG signal, and still maintains a good trajectory tracking ability under severe fatigue.

[0100] In summary, the adaptive fuzzy PID controller based on electromyographic feedback showed better trajectory tracking performance under various fatigue conditions, especially when the patient was fatigued, its dynamic adjustment ability effectively improved the control accuracy and robustness of the system. This shows that this control method has significant advantages in rehabilitation training and helps to improve the rehabilitation effect.

[0101] At the same time, combined Fig.10 , for the angle control of the shoulder joint of the upper limb rehabilitation robot, the performance of three different PID control methods (fixed PID, fuzzy PID and fuzzy PID based on electromyographic feedback) in three states of non-fatigue, mild fatigue and severe fatigue were compared. The expected trajectory is a smooth sine wave, which represents the motion trajectory of the shoulder joint under normal conditions. The study found that as the fatigue state worsens, the errors and oscillations of various control methods show significant differences. In the non-fatigue state, although the fixed PID shows slight oscillation, the error is small and it can follow the expected trajectory well; the fuzzy PID has a faster response, smaller oscillation and lower error; and the fuzzy PID based on electromyographic feedback almost perfectly follows the expected trajectory and performs best. As the fatigue state enters mild fatigue, the oscillation and error of the fixed PID increase significantly, and the response speed slows down; although the fuzzy PID has a slight lag, its oscillation amplitude is still small, and it can maintain good control accuracy; the fuzzy PID based on electromyographic feedback can still achieve good tracking effect in this state, showing slight errors and small fluctuations. Under severe fatigue conditions, the control accuracy of fixed PID is significantly reduced, the oscillation is violent, the error is large, and it is difficult to accurately follow the desired trajectory; although the fuzzy PID oscillation increases, it can reduce the error to a certain extent; and the fuzzy PID based on electromyography feedback still performs well under severe fatigue conditions, with a small oscillation amplitude and the smallest error, and can better track the desired motion trajectory. In summary, the fuzzy PID method based on electromyography feedback shows the best control effect under different fatigue conditions, especially under severe fatigue conditions, its superiority is more obvious. Therefore, the fuzzy PID method based on electromyography feedback has broad application prospects in upper limb rehabilitation training, and can provide patients with a more accurate and stable rehabilitation training experience.

[0102] The simulation results show that the fuzzy PID control method based on sEMG feedback is significantly better than the traditional fixed PID and fuzzy PID control methods under different fatigue conditions. Its average tracking error and maximum deviation are both at a low level, especially under mild and severe fatigue conditions. This method greatly improves the trajectory tracking accuracy of the system through real-time adaptive adjustment, enhances the intelligence level and overall stability of the rehabilitation robot, and can better adapt to the real-time fatigue state changes of patients, thereby enhancing the personalized effect of rehabilitation training. At the same time, the system is more robust, effectively avoiding the problems of motion instability and decreased control accuracy caused by fatigue.

[0103] In summary, the present application proposes an adaptive fuzzy PID control method based on surface electromyography (sEMG) and applies it to an upper limb rehabilitation robot, which can effectively solve the problems of decreased trajectory tracking accuracy and unstable rehabilitation effect caused by changes in fatigue state during rehabilitation training. At the same time, by constructing a soft voting model / multiple model fusion technology for fatigue state detection, and combining the fuzzy PID control method, the control parameters are adjusted according to real-time sEMG feedback, and the system can effectively respond to the dynamic changes of users in non-fatigue, mild fatigue and high fatigue states.

[0104] The above disclosure is only the preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A fuzzy PID control method for an upper limb rehabilitation robot based on electromyographic feedback, the corresponding technical solution includes: Step 1, sampling surface electromyography signal data; Step 2, performing signal processing on the aforementioned surface electromyography signal sampling data; Step 3: According to the importance score, feature extraction is performed on the surface electromyography signal sampling data after the above signal processing by a feature selection method, so as to obtain a key feature set; Step 4: Create a fusion model, use the fatigue state key feature set to train the fusion model and determine the best fusion model; thereby detecting the fatigue state of the subject in real time and completing the classification of sEMG signals; Step 5: According to the aforementioned optimal fusion model, the fatigue state corresponding to the patient's surface electromyography signal currently acquired in real time is obtained, and the corresponding PID parameters are dynamically adjusted in combination with the adaptive fuzzy PID controller based on electromyography feedback to adapt to different fatigue states.

2. The fuzzy PID control method for an upper limb rehabilitation robot based on electromyographic feedback as claimed in claim 1, characterized in that: The steps of processing the aforementioned surface electromyography signal data include: (1) rectifying and filtering the aforementioned surface electromyography signal data, specifically removing the DC component, bandpass filtering and full-wave rectification of the aforementioned surface electromyography signal data in sequence; and (2) performing feature extraction processing on the aforementioned surface electromyography signal data by a sliding window method, thereby converting the continuous original signal into a discrete signal.

3. The fuzzy PID control method for an upper limb rehabilitation robot based on electromyographic feedback as claimed in claim 1, characterized in that: According to the importance score, the surface electromyography signal sampling data after the above-mentioned signal processing is subjected to feature extraction through the feature selection method, so as to obtain the key feature set of the fatigue state. The processing steps include: the feature selection method adopts the SelectFromModel method, thereby using the machine learning model to evaluate the importance of each data feature, and selects the key features to be retained according to the importance score.

4. The fuzzy PID control method for an upper limb rehabilitation robot based on electromyographic feedback as claimed in claim 3, characterized in that: The specific steps of the aforementioned step 3 include: (1) selecting a training model and training according to the selected training model using the feature set obtained above, thereby providing feature importance scores through the trained training model; (2) setting a threshold and selecting key features, that is, setting a threshold based on experience or automatically selecting a suitable threshold by SelectFromModel, so that features with importance scores higher than this threshold are selected and retained, thereby generating a new data set.

5. The fuzzy PID control method for an upper limb rehabilitation robot based on electromyographic feedback as claimed in claim 1, characterized in that: The fusion model creation method includes selecting a combination of multiple basic classifiers, forming multiple evaluation strategies through multiple model fusion technologies, determining the best strategy, detecting the fatigue state of the current person to be tested, and providing the detection result.

6. The fuzzy PID control method for an upper limb rehabilitation robot based on electromyographic feedback as claimed in claim 5, characterized in that: The types of basic classifier combinations include MLP_KNN, LightGBM_LR_MLP, RBF_KNN and SVM_RBF_KNN; the model fusion techniques include stacking, voting and averaging.

7. The fuzzy PID control method for an upper limb rehabilitation robot based on electromyographic feedback as claimed in claim 5, characterized in that: The soft voting model SoftVoting is used to make the best strategy by weighted averaging the prediction probability of the base learner. The corresponding model formula is: Among them, P(class i ) is the final prediction result, i is the category number, i.e., the fatigue level category, N is the number of base learners, P j (class i ) is the predicted probability of the j-th base learner for the i-th class.

8. The fuzzy PID control method for an upper limb rehabilitation robot based on electromyographic feedback as claimed in claim 1, characterized in that: The output formula of the adaptive fuzzy PID controller is: Among them, K p (e, Δe, f): proportional gain dynamically adjusted according to the system error e(t), error change rate Δe(t) and fatigue state f, which can be divided into no fatigue, mild fatigue and severe fatigue; K i (e, Δe, f): integral gain dynamically adjusted according to the same input as above; K d (e, Δe, f): differential gain dynamically adjusted according to the same input as above; f: fatigue state of EMG signal feedback.

Images