A sports protection bandage system

Abstract

This invention provides a sports protective bandage system, relating to the field of sports protective bandage technology. The invention acquires user movement and physiological data in real time through multimodal sensing and signal processing, constructs and continuously updates a personalized dynamic safety boundary model, solving the problem that general protection cannot adapt to individual differences. By calculating psychological stress levels, psychological safety is incorporated into decision-making, avoiding secondary risks caused by panic. Based on real-time data and rehabilitation trends, a desensitization guidance strategy is generated, which, through tactile and pressure linkage regulation, actively guides muscle exertion and center of gravity adjustment while ensuring joint safety, effectively overcoming the "disuse" problem caused by traditional protection. Finally, through a safety guidance arbitration mechanism, intelligent decision-making is made in coordination of physical safety boundaries and psychological state, achieving a dynamic and personalized balance between immediate protection and long-term active rehabilitation.

Description

Technical Field

[0001] This invention relates to the field of sports protective bandage technology, and more specifically to an intelligent pressure-sensing sports protective bandage system. Background Technology

[0002] Sports protective bandages, especially those designed for chronic instability joints such as the ankle and knee (e.g., habitual ankle sprains), primarily function to provide external support for joints during daily activities or sports to enhance stability and prevent secondary injuries. Traditional elastic bandages or protective gear with adjustable straps often provide static support pressure or require manual adjustment by the user, and cannot be dynamically adjusted according to the joint's real-time load and stability.

[0003] Existing technologies include some smart protective bandages with pressure sensing and simple adjustment functions. These products typically integrate pressure sensors within the bandage to monitor pressure distribution at the joints and adjust the bandage pressure using actuators such as miniature air pumps. However, the adjustment logic of these products is mostly based on increasing support or simple movement pattern recognition when excessive pressure is detected. This is a passive response or fixed program control, failing to address the "disuse atrophy" problem caused by long-term wear, i.e., "disuse muscle atrophy" or "disuse functional degeneration." This manifests as the muscles, ligaments, and proprioceptive system around the joint gradually atrophying, becoming sluggish, or experiencing decreased coordination due to over-reliance on external support, which is detrimental to the long-term recovery of joint function. Furthermore, there is a lack of consideration for the user's psychological state and personalized safety boundaries. During rehabilitation training, the psychological panic caused by the user's fear of re-injury can lead to muscle stiffness and decreased coordination, increasing the risk. Simultaneously, there are significant differences in joint stability, muscle strength, and psychological resilience among individuals, and existing products cannot define and adhere to personalized safety and challenge balance boundaries. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention provides a sports protective bandage system.

[0005] The technical solution adopted in this invention is as follows:

[0006] A sports protective bandage system includes a multimodal sensor array, a central processing unit, and an actuator array. It employs an intelligent pressure-sensing sports protective bandage linkage adjustment method, the method comprising the following steps:

[0007] S1: The user's biomechanical and physiological signals are collected in real time through the multimodal sensor array, and the central processing unit processes the collected signals to obtain processed biomechanical and physiological data; wherein, the processed biomechanical data includes at least the real-time motion trajectory and posture data of the joints and the activation feature data of the target muscle group.

[0008] S2: Based on processed biomechanical data from historical records, a personalized dynamic safety boundary model representing the safe range of motion of a user's joints is constructed and continuously updated through machine learning algorithms.

[0009] S3: Based on the processed physiological data acquired in real time, a psychological stress level that quantifies the user's panic or tension is calculated.

[0010] S4: Based on the current processed biomechanical data and historical rehabilitation trends, generate a desensitization guidance strategy aimed at guiding users to actively shift their center of gravity or adjust their posture.

[0011] S5: Execute safety guidance arbitration, specifically: compare the predicted user movement state corresponding to the desensitization guidance strategy with the real-time safety range defined by the personalized dynamic safety boundary model, and simultaneously combine the psychological stress level, generating a post-arbitration adjustment instruction based on preset collaborative arbitration logic; wherein, the collaborative arbitration logic is configured as follows: when the predicted user movement state is within the real-time safety range and the psychological stress level is lower than a first threshold, output a first arbitration result allowing the execution of the desensitization guidance strategy; when the predicted user movement state exceeds the real-time safety range or the psychological stress level is higher than a second threshold, output a second arbitration result to tune, terminate, or convert the desensitization guidance strategy to a soothing mode;

[0012] S6: Based on the arbitrated adjustment command generated in step S5, drive the actuator array to perform the corresponding adjustment operation.

[0013] Step S1 specifically includes:

[0014] S11: Acquire multi-axis kinematic signals of the joint through the inertial measurement unit, acquire electromyographic signals of the target muscle group through the electromyographic sensor module, and acquire skin conductance signals of the user through the skin conductance sensor.

[0015] S12: The central processing unit processes the multi-axis kinematic signals to obtain the real-time motion trajectory and posture data of the joint, processes the electromyographic signals to obtain the activation characteristic data of the target muscle group, and processes the skin conductance signals to obtain the physiological data.

[0016] In step S2, the personalized dynamic safety boundary model is constructed and continuously updated. Specifically, by analyzing the historical sequence of the real-time motion trajectory and posture data of the joint, a multi-dimensional parameter space boundary is constructed using clustering or probability distribution methods to characterize the safe range of motion of the joint under different activity modes. This multi-dimensional parameter space boundary is used as the personalized dynamic safety boundary model. The threshold of the multi-dimensional parameter space boundary is periodically updated using the newly acquired real-time motion trajectory and posture data of the joint.

[0017] The psychological stress level is calculated in step S3 by analyzing the amplitude change rate and spectral characteristics of the skin conductance signal in the physiological data to obtain a quantitative value of the psychological stress level that reflects the degree of excitation of the user's autonomic nervous system.

[0018] In step S4, the desensitization guidance strategy is generated by generating guiding tactile feedback instructions and / or target support pressure parameters to guide the user to perform center of gravity shifting or posture adjustment, based on the real-time motion trajectory and posture data of the joint and the activation characteristic data of the target muscle group, combined with historical rehabilitation trend data. The guiding tactile feedback instructions include information on the site of action, stimulation mode and intensity.

[0019] In step S5, when the second arbitration result is output, the collaborative arbitration logic is further configured to generate a soothing tactile feedback instruction to reduce the user's tension, the soothing tactile feedback instruction including rhythmic vibration or mild thermal stimulation mode.

[0020] Step S6 specifically includes: when the arbitrated adjustment instruction includes the first arbitration result, driving the distributed tactile actuator group in the actuator array to generate directional tactile stimulation at the position corresponding to the action site information in the guided tactile feedback instruction on the user's body surface, and / or driving the pressure regulating actuator in the actuator array to adjust the bandage pressure to the target support pressure parameter; when the arbitrated adjustment instruction includes the second arbitration result, driving the distributed tactile actuator group to generate tactile stimulation corresponding to the soothing tactile feedback instruction, and / or driving the pressure regulating actuator to adjust the bandage pressure to a preset safe support pressure.

[0021] The method further includes: S7: storing the user response data and adjustment result data of each adjustment process into the data storage module; the user response data is the biomechanical data newly collected and processed after executing step S6, and the difference data obtained by comparing it with the corresponding biomechanical data before executing S6; the adjustment result data includes at least the type and parameters of the adjustment command that have been executed.

[0022] The method further includes: S8: using the user response data and adjustment result data stored in step S7, periodically optimizing and updating the generation logic of the personalized dynamic safety boundary model and the desensitization guidance strategy through a machine learning algorithm; wherein, the optimization and updating of the generation logic includes adjusting the intensity and frequency parameters of the guiding haptic feedback instruction and the adjustment step size of the target support pressure parameter.

[0023] The beneficial effects of this invention are:

[0024] This invention acquires real-time user motion and physiological data through multimodal sensing and signal processing, constructs and continuously updates a personalized dynamic safety boundary model, solving the problem that general protection cannot adapt to individual differences; by calculating psychological stress levels, it incorporates psychological safety into decision-making, avoiding secondary risks caused by panic; based on real-time data and rehabilitation trends, it generates desensitization guidance strategies, and through the linkage of tactile and pressure regulation, it actively guides muscle exertion and center of gravity adjustment while ensuring joint safety, effectively overcoming the "disuse" problem caused by traditional protection; finally, through a safety guidance arbitration mechanism, it coordinates physical safety boundaries and psychological states to make intelligent decisions, achieving a dynamic and personalized balance between immediate protection and long-term active rehabilitation. Attached Figure Description

[0025] Figure 1 A flowchart of the intelligent pressure-sensing motion protective bandage linkage adjustment method according to an embodiment of the present invention;

[0026] Figure 2 This is an overall flowchart of the intelligent pressure-sensing motion protection bandage linkage adjustment method according to an embodiment of the present invention;

[0027] Figure 3 This is a flowchart illustrating the security-guided arbitration decision-making process according to an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] like Figures 1-3 As shown, an embodiment of the present invention provides a sports protective bandage system, comprising a multimodal sensor array, a central processing unit, and an actuator array, and using an intelligent pressure-sensing sports protective bandage linkage adjustment method, which specifically includes the following steps:

[0030] S1: A multimodal sensor array is used to acquire the user's biomechanical and physiological signals in real time. The central processing unit processes these signals to obtain processed biomechanical and physiological data. The processed biomechanical data includes at least real-time joint motion trajectories and posture data, as well as activation characteristic data of the target muscle groups. This step uses a multimodal sensor array to synchronously acquire signals related to the user's joint movements, muscle activity, and autonomic nervous system responses. The algorithms of the central processing unit (CPU / MCU) then transform the raw sensor signals into standardized data with clear physical or physiological meaning.

[0031] Specifically, step S1 includes: S11: acquiring multi-axis kinematic signals of the joint through the inertial measurement unit, acquiring electromyographic signals of the target muscle group through the electromyographic sensor module, and acquiring skin conductance signals of the user through the skin conductance sensor.

[0032] This step, serving as the raw signal acquisition stage, simultaneously collects raw signals of user joint movement, muscle activity, and autonomic nervous system responses using three types of dedicated sensors. The inertial measurement unit (IMU) can be a miniature six-axis IMU (integrating a three-axis accelerometer and a three-axis gyroscope), and can optionally be equipped with a three-axis magnetometer to form a nine-axis IMU to improve attitude calculation accuracy. Its package size is ≤10mm×10mm×3mm, suitable for bandage-wearing scenarios, and the sampling frequency can be set to 100-200Hz. The inertial measurement unit (IMU) collects multi-axis kinematic signals (such as those from the ankle and knee joints), specifically including three-axis acceleration signals reflecting instantaneous acceleration changes in the joint's three-dimensional space. , , (Unit: m / s) 2 Examples include vertical acceleration during walking and horizontal acceleration during turning; triaxial angular velocity signals reflect the rotational rate of joints around three-dimensional coordinate axes. , , (Unit: rad / s), such as the angular velocity around the x-axis during ankle plantar flexion / dorsiflexion; and a triaxial magnetic field strength signal that can be used to correct attitude drift. , , (Unit: μT), suitable for long-term wear scenarios. The IMU can be attached to the inside of the bandage and fixed coaxially with the joint rotation center. For example, the ankle joint IMU can be fixed to the front of the lower leg, aligned with the center of the talus, to ensure that the sensor moves in sync with the joint.

[0033] The electromyography (EMG) sensor module can use dry electrode EMG sensors, eliminating the need for conductive gel and adapting to dynamic motion scenarios. The number of electrodes can be configured from 2 to 4 depending on the target muscle group. For example, in ankle protection, signals from the peroneus longus and brevis muscles and the tibialis anterior muscle need to be collected. The sampling frequency can be set to 200-500Hz, covering the effective frequency band of EMG signals from 20-500Hz, with a signal resolution ≥12bit to ensure effective identification of weak EMG signals. Electrodes can be attached along the muscle fiber direction, with a spacing of 2-3cm between adjacent electrodes. The reference electrode is attached to an inactive area of ​​the muscle (such as the periosteum on the lateral side of the lower leg) to reduce motion artifact interference. It primarily collects surface EMG signals (sEMG) of the target muscle group. (Unit: μV) This signal reflects the intensity of electrical activity in muscle fibers. The amplitude of the electromyographic signal increases during muscle contraction and approaches the resting potential during relaxation.

[0034] The skin conductance sensor employs a dual-electrode structure, with electrodes made of medical-grade silver / silver chloride, exhibiting both good biocompatibility and conductivity stability. It features a sampling frequency of 50-100Hz, a measurement range of 0.01-100μS, and a signal noise level ≤0.001μS, covering the normal range of human skin conductance from 0.1-50μS. The two electrodes are symmetrically attached to the inside of a bandage, such as the sides of the dorsum of the foot on an ankle bandage, ensuring close contact with the skin to avoid signal fluctuations caused by friction during movement. Its primary function is to collect the conductance signal from the user's skin. (Unit: μS) This signal is positively correlated with sweat gland secretion activity. Since sweat gland secretion is regulated by the autonomic nervous system, when users experience emotions such as panic and tension, the sympathetic nervous system is excited, sweat gland secretion increases, and skin conductance rises significantly. Therefore, this signal is the core physiological indicator for subsequent quantification of psychological stress levels.

[0035] S12: The central processing unit processes multi-axis kinematic signals to obtain real-time motion trajectory and posture data of the joints, processes electromyographic signals to obtain activation characteristic data of the target muscle group, and processes skin conductance signals to obtain physiological data.

[0036] This step uses a series of algorithms, including filtering, computation, and quantization, to remove noise interference from the original signal, extract key features, and ultimately output biomechanical and physiological data. Multi-axis kinematic signal processing converts the acceleration and angular velocity signals acquired by the IMU into quantitative data characterizing the joint's spatial state. First, signal preprocessing is required to remove noise and calibration errors. A 4th-order Butterworth low-pass filter (cutoff frequency 10Hz) can be used to remove high-frequency vibration noise. Simultaneously, zero-bias calibration (acquiring 100 samples in a static state, averaging, and subtracting the zero-bias value) can eliminate inherent IMU errors. For the magnetometer, ellipsoidal fitting calibration can be used to eliminate hard / soft iron interference.

[0037] Attitude calculation can be performed using the quaternion method to calculate the 3D attitude of the joints (avoiding the Euler angle gimbal lock problem), where a quaternion is defined as... ,in For the real part, , , The imaginary part satisfies Its update equation is In the formula This is the quaternion multiplication operator; The quaternion for angular velocity; The quaternion is the time derivative (unit: 1 / s); then the quaternion is converted to Euler angles (roll angle). Pitch angle Yaw angle The unit is rad, which directly represents the spatial posture of the joint. For example, the pitch angle of the ankle joint corresponds to the plantar flexion / dorsiflexion angle, and the roll angle corresponds to the inversion / eversion angle.

[0038] Real-time joint motion trajectory data can be obtained through the second integral of acceleration. First, the velocity is calculated from the integral of acceleration, using the following formula: In the formula This is the initial velocity (default value is 0). Let the velocity be the integral variable; then, the position is obtained by integrating the velocity, using the following formula: In the formula The initial position coordinates (set as the origin) are used to output the final three-dimensional position coordinates. (Unit: m); To reduce integral drift, this invention uses Kalman filtering to smooth the position data. The filtering equation is:

[0039] ;

[0040] In the formula This is the optimal location estimate. Here is the state transition matrix. For the control matrix, For Kalman gain, The original location data obtained by integration. For the observation matrix, For control input (external control signal of the system).

[0041] The final output includes biomechanical data encompassing the joint's three-dimensional pose (Euler angles) and three-dimensional spatial trajectory (position coordinates). .

[0042] The core objective of electromyography (EMG) signal processing is to extract the "activation intensity" feature of muscle contraction. First, bandpass filtering is performed. A 128th-order FIR bandpass filter (20-500Hz) can be used to preserve the effective frequency band of the EMG signal while filtering out power line interference (50Hz) and motion artifacts. The filtering equation is as follows: In the formula This is the filtered electromyography signal. These are the filter coefficients. Let the filter order be . The sampling period (f) s =500Hz), The original electromyographic signal is shown. Full-wave rectification is then performed to convert the bidirectional electromyographic signal into a unidirectional signal to highlight its amplitude characteristics. The formula is as follows: Then, a moving average filter is used to eliminate high-frequency fluctuations, as shown in the formula: In the formula The window length is 50 samples, corresponding to 0.1s. This represents the envelope of electromyographic signals.

[0043] Muscle activation (Dimensionless, range [0,1]) represents the ratio of the current contractile strength of a muscle to its maximum contractile strength. It is a core indicator reflecting the functional state of the muscle, and its calculation equation is: In the formula The mean value of the muscle envelope in the resting state (collected 30 seconds after wearing). Peak envelope value at maximum spontaneous contraction (MVC) of the user (calibrated during first use).

[0044] The final output is the activation feature data of the target muscle group. , For the number of muscle groups, such as ankle protection This corresponds to the peroneus longus and brevis muscles and the tibialis anterior muscle.

[0045] The core objective of skin conductance signal processing is to remove signal noise and obtain stable, standardized skin conductance data. First, an exponential smoothing filter is used to eliminate motion artifacts and baseline drift. The filter equation is as follows: In the formula This is the filtered skin conductance signal. T is a smoothing coefficient (which can be taken as 0.05 to balance response speed and stability). s =0.02s (sampling frequency 50Hz) This is the original skin conductance signal. Baseline calibration is then performed, using the average conductance 10 seconds after the first wear as the baseline. Calculate the relative conductivity value This highlights the dynamic changes. The final output is physiological data. ,in This is an absolute conductivity value, reflecting the basic autonomic nervous system state. It is a relative conductivity value, reflecting the dynamic changes under stress.

[0046] S2: Based on processed biomechanical data from historical records, a personalized dynamic safety boundary model representing the safe range of motion of a user's joints is constructed and continuously updated using machine learning algorithms.

[0047] The safe range of motion of joints varies significantly from person to person and changes dynamically as muscle strength and proprioception recover. This step addresses the shortcomings of the traditional product’s “generalized safety boundary” by using the user’s own historical biomechanical data, selecting a clustering and probability distribution fusion machine learning model that is adapted to multiple activity modes, and using newly collected data to regularly update model parameters and boundary thresholds, thus achieving the personalization and dynamic evolution of the safety range.

[0048] Specifically, the construction and continuous updating of a personalized dynamic safety boundary model involves: analyzing the historical sequence of real-time motion trajectory and posture data of joints, using clustering or probability distribution methods to construct a multi-dimensional parameter space boundary to characterize the safe range of motion of joints under different activity modes, which serves as a personalized dynamic safety boundary model; and periodically updating the threshold of the multi-dimensional parameter space boundary using newly acquired real-time motion trajectory and posture data of joints.

[0049] Step S2 follows a closed-loop process of historical data preprocessing and feature engineering, model building, safety boundary definition, and dynamic updates. The historical data preprocessing and feature engineering stage provides high-quality input data for model building. Its core is extracting key features characterizing joint motion states from the original historical sequences. The specific steps are: first, historical data collection and cleaning are performed. Assuming the user's cumulative wearing time is within a certain range, the biomechanical data sequence output by S1 is... ,in The total number of historical samples must meet the following requirements. To ensure model reliability, For the first The timestamp of each sample collection. To ensure data quality, the 3σ criterion can be used to remove abnormal data caused by sensor malfunctions and motion artifacts. The judgment logic is as follows: ;in For feature set, Features Historical average, Features The historical standard deviation; at the same time, the historical data of different activity modes (walking / running / going up and down stairs) are uniformly resampled to 100Hz to keep in line with the sampling frequency of step S1, so as to ensure alignment of the time dimension.

[0050] The specific method for collecting historical biomechanical data is as follows: Data collection is automatically triggered each time the user wears the bandage and starts the system. The collection duration covers the entire wearing cycle (from startup to shutdown), and the sampling frequency remains consistent with step S1 (100-200Hz) to ensure the continuity and consistency of the data over time. The collected data consists of real-time biomechanical data and scene tags. The scene tags are automatically generated using the activity pattern clustering algorithm in step S2, such as walking, going up and down stairs, and remaining stationary, enabling historical data to be stored according to movement scenarios. To ensure the security and validity of historical data, only data from a system in a safe operating state is collected, i.e., when the arbitration result in step S5 is R1 (allowing guidance) or R... 21 Biomechanical data during (tuning), excluding R 22 (Suspension), R 23 Abnormal data in the (soothing + suspension) state is used to avoid interference from abnormal movement states on model training. At the same time, to ensure the reliability of model training, a minimum data requirement can be set: after accumulating ≥5000 valid samples (approximately corresponding to 50 minutes of effective wearing time), the first Gaussian Mixture Model (GMM) construction can be initiated, enabling the model to fully learn the user's safe movement characteristics.

[0051] Subsequently, key features were extracted to characterize the safe range of motion of joints under different activity modes. We extract 7 core features (covering pose, position, and velocity) to construct a multidimensional feature space. The feature vector expression is as follows:

[0052] ;

[0053] in , where is the roll angle (unit: rad), representing the angle of rotation of the joint around the x-axis; The pitch angle (unit: rad) represents the angle of rotation of the joint about the y-axis; Yaw angle (unit: rad) represents the angle of joint rotation about the z-axis; , , These are the three-dimensional spatial coordinates of the joint (unit: m); The instantaneous velocity (unit: m / s) is derived from step S1. Calculated.

[0054] To eliminate the impact of dimensional differences on the model, the feature vectors were modified. Perform Z-score standardization: ;

[0055] in The historical mean vector of 7-dimensional features. The historical standard deviation vector of the 7-dimensional features. This is the standardized feature vector.

[0056] In the personalized dynamic security boundary model construction phase, this invention specifically employs a Gaussian Mixture Model (GMM) (a probability distribution-based method). The GMM model assumes historical standardized feature vectors. The model expression for a mixture distribution that follows multiple multivariate normal distributions is:

[0057] ;

[0058] In the formula The probability density of the eigenvector (unit: 1 / feature unit) represents the standardized eigenvector. The probability density of movement within a safe range; the higher the value, the more likely one is to be in a safe state. The mixed component number (positive integer) represents the number of user activity patterns, which is automatically determined by the Bayesian Information Criterion (BIC), with a default range of 3 to 5, such as walking, running, going up and down stairs, etc. For the first The weight of the Gaussian component (range [0,1]) represents the weight of the Gaussian component. The proportion of various activity modes in users' daily exercise needs to meet ; Let be a multivariate normal distribution function, describing the th Distribution patterns of safe movement characteristics under various activity modes; For the first The component mean vector (unit: standardized feature units) represents the first... Typical characteristic values ​​of the activity mode, such as the average pitch angle and average speed during walking, are 7×1. For the first Component covariance matrix (unit: (standardized characteristic units)) 2 ), representing the first The correlation between features under different activity modes, such as the correlation between pitch angle and velocity, is shown in a 7×7 dimension, where the covariance matrix... It is the core of characterizing the boundary of the multidimensional parameter space, and directly determines the first The shape of the safety range under each activity mode is expressed mathematically as follows:

[0059] ;

[0060] diagonal elements (like ) is the first Features of various activity modes The variance represents the safety fluctuation range of this feature; the larger the variance, the higher the safety tolerance of this feature; off-diagonal elements

[0061] (like ) is the first Features of various activity modes With features The covariance represents the linked safety constraints of the two features, such as the change in the safety upper limit of speed when the pitch angle increases; This is the expectation operator, which calculates the mean of a random variable; here, it is used to solve for the covariance.

[0062] This invention employs a Gaussian mixture model (GMM) to construct personalized dynamic safety boundaries. Its core principle is to quantify the safe movement characteristics under different user activity patterns through model parameters, thereby achieving a complete safety representation logic for safety pattern classification, fluctuation range definition, and matching degree determination. The GMM model assumes that the historical standardized feature vector follows a mixture distribution of multiple multivariate normal distributions. Each mixture component of the GMM corresponds to a user's 'safe activity pattern' (such as walking, slowly going up and down stairs). The component weight represents the frequency of use of this pattern in the user's daily activities; a higher weight indicates a more common pattern and a higher safety priority.

[0063] The mean vector of each hybrid component is the core quantitative indicator of the "optimal safe joint movement state" in that movement mode. It integrates key features such as joint posture, spatial position, and movement speed. For example, the mean vector corresponding to the walking mode corresponds to the roll angle, pitch angle, three-dimensional position coordinates, and instantaneous velocity at which the joints are most stable and the muscle load is most balanced when the user walks, providing a clear safety benchmark for the user's movement. The closer the user's movement state in this mode is to this mean vector, the more reasonable the joint force and the lower the movement risk.

[0064] The covariance matrix of the mixed components further defines the safe fluctuation range of each motion feature, ensuring that the safety boundary is not a rigid value, but an elastic range that conforms to the laws of human movement. The diagonal elements of the matrix correspond to the variance of a single motion feature, such as the variance of the roll angle, which directly determines the safe angle range of joint inversion and eversion in this motion mode. Exceeding this range means that the force on the joint may be unbalanced. The off-diagonal elements represent the linkage safety constraints between different motion features. For example, when the joint pitch angle increases, the safe upper limit of speed will be adjusted accordingly to avoid the risk caused by the single feature meeting the standard but multiple features being unbalanced, and to achieve a comprehensive safety judgment of multi-dimensional motion states.

[0065] The safety of real-time motion is determined by the probability density value calculated by the model. This probability density value essentially represents the degree of matching between the real-time motion state and all safe motion modes. When the probability density value is not lower than a preset safety probability threshold, it indicates that the real-time motion state highly matches at least one safe motion mode and is within the elastic safety range of that mode. When the probability density value is lower than the safety probability threshold, it indicates that the real-time motion state deviates from the normal fluctuations of all safe motion modes, and the joints face potential damage risks. In this case, the system will intervene through a subsequent safety guidance arbitration mechanism to ensure user motion safety. This hierarchical representation logic based on model parameters allows the safety range to not only conform to the individual user's motion characteristics but also possess dynamic adaptability, solving the shortcomings of traditional general safety boundaries that cannot adapt to individual differences.

[0066] In this invention, model parameters The solution is obtained using the Expectation-Maximization (EM) algorithm. Specifically, during the initialization phase, each sample is randomly assigned to belong to the first... posterior probability of a component ( (For sample indexing), and initialize model parameters. , , Then calculate the posterior probability of each sample, using the following formula:

[0067] ;

[0068] in For the first The sample belongs to the first The posterior probability of a component. For the first The standardized feature vector of each sample. For component indexes.

[0069] Next, update the model parameters. The weight update formula is: ,in Given the total number of historical samples, the formula for updating the mean vector is: The formula for updating the covariance matrix is: When the parameter update amount is less than the convergence threshold When the iteration stops, the final model parameters are obtained.

[0070] The safety boundary determination logic is implemented based on the trained GMM model, that is, it determines whether the real-time motion state is within a safe range by using a probability density threshold. First, a safety probability threshold is set. (Default 0.7, can be adjusted according to the user's recovery stage), its physical meaning is "the minimum probability density requirement for real-time motion state to fall within a safe range"; then, the real-time standardized feature vector is... (Calculated from S1 real-time data), substituted into the GMM model to calculate the probability density. The corresponding judgment rule is: Conversely, it is beyond the safe range.

[0071] The model's dynamic update mechanism ensures that the safety boundary is continuously optimized as the user recovers. Update triggering conditions include time-based triggers and event-based triggers, with time-based triggers occurring after each cumulative data collection. An update is triggered for each new sample (approximately corresponding to 10 minutes of wear time); an event trigger occurs when the user's recovery stage upgrades, such as switching from "passive rehabilitation" to "active training," manually triggering a full update. To avoid the high power consumption of full training, this invention uses the incremental EM algorithm to update model parameters.

[0072] The weight update formula is:

[0073] ;

[0074] The formula for updating the mean vector is:

[0075] ;

[0076] The formula for updating the covariance matrix is:

[0077] ;

[0078] in The number of historical samples before the update. This represents the number of newly added samples; For the newly added The sample belongs to the first The posterior probability of the component; , For the newly added sample set in the 1st Mean vector and covariance matrix of the components; , , These are the updated model parameters.

[0079] For boundary threshold updates, the safety probability threshold The formula is dynamically adjusted according to the recovery process and updated as follows: In the formula Adjust the threshold step size (default 0.05). As a recovery trend factor, The older the child, the better their recovery. The smaller the size, the wider the safety range, encouraging users to gradually expand their athletic abilities. , The safety probability threshold before / after the update.

[0080] S3: Based on real-time acquired and processed physiological data, a psychological stress level is calculated to quantify the user's panic or tension.

[0081] As a physiological indicator reflecting the level of autonomic nervous system excitation, the changes in skin conductance are directly related to psychological stress. When users experience panic or tension, sympathetic nerve excitation promotes increased sweat gland secretion, leading to an increase in skin conductance amplitude and a faster rate of change, with specific spectral characteristics. However, the raw skin conductance signal is easily affected by motion friction, environmental humidity, and individual physiological differences. Therefore, preprocessing is necessary to enhance signal purity. In the secondary denoising stage, based on S1 exponential smoothing filtering, median filtering is added to remove impulse noise such as instantaneous spikes caused by motion friction. The formula is:

[0082]

[0083] in This is the median filtering operator; The filter window half-width is set to 2 (corresponding to a window length of 5 samples); T s =0.02s (the skin conductance sensor sampling frequency in step S1 is 50Hz). The smoothed signal output from step S1 The signal is after secondary denoising (unit: μS).

[0084] It should be noted that the baseline skin conductivity (resting state) varies significantly among different users (e.g., the baseline difference between dry and moist skin can be up to 10 times), requiring baseline calibration and normalization. First, after the user's initial wear and 30 seconds of rest, the mean of the denoised signal during this period is collected as the resting baseline. The calculation formula is:

[0085] ;

[0086] in Resting sample size One sample.

[0087] Then, the relative skin conductance signal was calculated to eliminate baseline differences and highlight the dynamic changes caused by stress. The calculation formula is as follows:

[0088] ;

[0089] like Then take (Since stress only causes an increase in conductivity, a negative change is noise).

[0090] The psychological stress level was calculated as follows:

[0091] By analyzing the amplitude change rate and spectral characteristics of skin conductance signals in physiological data, a quantitative value of psychological stress level reflecting the degree of excitation of the user's autonomic nervous system is calculated.

[0092] This invention calculates psychological stress levels by analyzing the amplitude change rate and spectral characteristics of skin conductance signals. These two features capture stress states from both instantaneous dynamic and steady-state rhythmic dimensions, forming a complementary relationship and avoiding the limitations of a single feature. The amplitude change rate characterizes the rate of increase / decrease of skin conductance over time; this value increases significantly during stress, and its mathematical expression is: In the formula Set the calculation window for the rate of change (default 0.5s, balancing response speed and stability). The amplitude change rate (unit: μS / s) is used, and only positive values ​​are taken. (Since stress only leads to an increase in electrical conductance, negative changes are part of the recovery process and are not included in the stress level). This represents the relative skin conductance signal.

[0093] To achieve standardization, Normalized to the [0,1] interval, we get : ;in This represents the normalized rate of change of amplitude. This represents the user's historical maximum rate of change (which can be collected through simulated stress tests during initial use and continuously updated with subsequent data; the default value is 5 μS / s).

[0094] Regarding spectral characteristics, the power spectral density (PSD) of the skin conductance signal in the 0.04-0.2 Hz band is directly related to autonomic nervous activity, and the power in this band significantly increases during stress. During the extraction process, first... Segment by sliding window (window length) (Contains 250 samples, with a 50% overlap), plus the Hanning window. ( Reduce spectral leakage and obtain Then, the time-domain signal is converted to a frequency-domain signal using a Fast Fourier Transform (FFT). ;in For FFT operators; For the number of samples in the window, Frequency (unit: Hz); To determine the length of the spectrum analysis window; The sampling period.

[0095] The power spectral density was then calculated. (Unit: μS) 2 ( / Hz), and integrate to obtain the cumulative power spectral density in the 0.04-0.2Hz frequency band. Finally, normalize to the [0,1] interval to obtain ,in The normalized spectral characteristics (dimensionless, range [0,1]); The user's historical maximum cumulative power spectral density (collected during initial use via simulated stress testing, and subsequently updated via step S7; default value is 10 μS). 2 ).

[0096] In this invention, the final quantification of psychological stress level adopts a weighted summation model, which fuses two normalized features into the final psychological stress level. The formula is: ;in The amplitude change rate weight (dimensionless, range [0,1], default 0.6) emphasizes the instantaneous dynamic response to stress (such as rapid triggering during sudden panic). This represents the spectral feature weights (dimensionless, range [0,1], default 0.4), emphasizing the steady-state persistence of stress (e.g., maintaining a high stress level during prolonged tension). The weight value can be determined by training with stress test data (watching horror videos, simulating sports risks) from 30 subjects, ensuring that the Pearson correlation coefficient between the fused value and the subjective stress score (1-10 points) is ≥0.85, thus guaranteeing the accuracy of the quantification.

[0097] To adapt to long-term changes in the user's physiological state (such as changes in skin moisture and metabolic levels), this invention also incorporates a dynamic calibration mechanism that updates periodically. , To avoid quantification bias, the updated formulas are as follows:

[0098] ;

[0099] ;

[0100] in This is the smoothing factor (default 0.8), which balances the weights of historical and new data. , The maximum rate of change of amplitude before / after the update; , The maximum cumulative power before / after the update; This refers to the last update time. This is the current update time (the default update cycle is 7 days). This is the maximum value operator, which extracts the maximum feature value between two updates.

[0101] S4: Based on the current processed biomechanical data and historical rehabilitation trends, a desensitization guidance strategy is generated to guide users to actively shift their center of gravity or adjust their posture. The core of this desensitization guidance is to achieve a dynamic balance between safety and challenge. Within personalized safety boundaries, subtle and precise tactile cues and progressive adjustments in support pressure encourage users to actively control their joints, avoiding disuse atrophy or proprioceptive degeneration caused by over-reliance on external support. This strategy is driven by dual-dimensional data of real-time movement status and long-term rehabilitation trends, ensuring that the guidance is both adapted to the current movement scenario and consistent with the user's overall rehabilitation process. Simultaneously, through the linked design of tactile feedback (actively guiding how to adjust) and support pressure (passively assisting for safety), it enhances user participation while ensuring safety during the adjustment process, achieving a smooth transition from external support to autonomous control.

[0102] The desensitization guidance strategy is specifically generated as follows: based on the real-time motion trajectory and posture data of the joints and the activation characteristic data of the target muscle groups, combined with historical rehabilitation trend data, guiding tactile feedback instructions and / or target support pressure parameters are generated to guide users to perform center of gravity shift or posture adjustment. The guiding tactile feedback instructions include information on the site of action, stimulation mode and intensity.

[0103] In this invention, the recovery trend is the core basis for the personalized adaptation of desensitization guidance strategies. It quantifies the degree of joint function recovery in users through historical biomechanical data, especially muscle activation characteristic data. (Recovery Trend Factor) This represents the proportion of the user's current joint function recovered relative to its initial state. It is a dimensionless parameter with a value range of [0,1]. This is the initial recovery phase (users need to rely on strong support). This represents a fully recovered state (able to stabilize joints independently without additional guidance). This factor is calculated using historical activation data from multiple cycles and multiple muscle groups, ensuring the stability and accuracy of the rehabilitation trend assessment. The calculation formula is as follows:

[0104] ;

[0105] In the formula This is the number of historical assessment cycles, with a default value of 4 (unit: weeks). By using multi-cycle data statistics, we can avoid the interference of short-term exercise fluctuations on the judgment of rehabilitation trends. To determine the target number of muscle groups, select 2 to 4 key stabilizing muscle groups depending on the joint being protected. For example, for ankle protection, select the two core muscle groups: the peroneus longus and brevis and the tibialis anterior. For the first Cycle number The initial average activation level of the muscle group, i.e., the first... When the week of rehabilitation began, the first Average activation of muscle groups during standard movements (such as walking, slow squatting); For the first Cycle number The current average activation level of the muscle group, i.e., the first... At the current stage of the week, the first The average activation level of a muscle group during the same standard movement. Among them, , All data were taken from historical muscle activation feature data stored in step S7. Standard movements are determined through S2 activity pattern clustering (such as walking patterns). To ensure the reasonableness of the physical meaning of the parameters, if the calculation results... Then take (Full recovery); Then take (Initial stage).

[0106] To further guide muscle function recovery more precisely, dynamic standard movement activation thresholds are set for each target muscle group. This threshold gradually increases as rehabilitation progresses, guiding muscle activation levels towards a normal functional state. The calculation formula is as follows: In the formula For the first The basic activation threshold for muscle groups is set to 0.3 by default to ensure that muscles maintain a minimum level of effective contraction. For the first The activation level of a muscle group can be increased by a default value of 0.5, meaning that the maximum target activation level for that muscle group is 0.8, to avoid setting too high a target that could lead to excessive muscle fatigue or injury. For the first The target activation level of the muscle group at the current stage is constrained to the range of [0.3, 0.8].

[0107] Guided tactile feedback instructions are the core execution vehicle of desensitization guidance strategies. They apply targeted stimulation to the user's body surface through distributed tactile actuators, directly prompting the user to adjust their center of gravity or posture. Its core parameters include the site of action, stimulation pattern (rhythm), and stimulation intensity, all of which are dynamically adjusted based on real-time data to ensure the accuracy and adaptability of the guidance.

[0108] The site of action The determination follows the directional guidance logic of "stimulating where adjustment is needed", based on the real-time joint position coordinates output in step S1. and attitude angle A mathematical mapping is performed to ensure a precise match between the stimulation site and the joint adjustment requirements. The mapping model is as follows:

[0109] ;

[0110] In the formula The set of tactile zones for bandages is pre-defined based on the anatomical structure of the joint being protected; for example, ankle bandages are divided into medial and lateral zones. outer side Ahead ,rear Four partitions (n=4); For the first The reference coordinates of the partitions are determined based on human anatomy before leaving the factory. For example, the reference coordinates of the medial partitions correspond to the surface projection position of the medial condyle of the ankle joint. For the first Reference attitude angles corresponding to different zones, such as the outward tilting attitude angle reference value for the outer zone. ; This is the attitude angle weighting coefficient, with a default value of 0.02m / rad, used to balance the influence of position coordinates and attitude angles on the determination of the affected part; This is the Euclidean distance operator. This refers to the partitioning operator that takes the minimum value. For example, when the ankle inversion angle... (Approaching the safety boundary), and location coordinates When biased inward, the result can be calculated using this model. (Outer zone) Uses tactile stimulation on the outer side to prompt the user to adjust their center of gravity outwards, thus avoiding excessive inversion.

[0111] Stimulation patterns based on stimulation rhythm Using these as core parameters, a faster rhythm is used in the early stages of rehabilitation to enhance the guiding effect, while a slower rhythm is used in the later stages to weaken the guiding effect, encouraging users to control their joints voluntarily. The mathematical expression is:

[0112] ;

[0113] In the formula The maximum stimulation rhythm is set to 3Hz by default, corresponding to the early stage of recovery. The strong guiding demand; The minimum stimulation rhythm is set to 0.5Hz by default, corresponding to complete recovery ( A subtle hint of need; The amplitude of rhythm adjustment is set to 2.5Hz by default, meaning the total adjustment range of the rhythm as it progresses with the recovery trend is [value missing]. For example, when (Initial rehabilitation) f stim =3-0.2×2.5=2.5Hz (fast rhythm, strong guidance); when (During the later stages of rehabilitation) f stim=3-0.8×2.5=1Hz (slow rhythm, weak guidance), to achieve dynamic adaptation of guidance intensity as rehabilitation progresses.

[0114] Stimulation intensity The design adapts to the user's real-time muscle activation status and recovery trend. When muscle activation is insufficient, the stimulation intensity is increased to enhance the guidance effect; the better the recovery, the weaker the stimulation intensity is to encourage voluntary control. Upper and lower limits are set to prevent excessively strong or weak stimulation from causing guidance failure. The mathematical expression is:

[0115] ;

[0116] In the formula The maximum stimulation intensity is set to 0.9 (dimensionless, in the range of [0,1]), based on the human tactile comfort threshold, to avoid excessive stimulation that could cause discomfort. The minimum stimulus intensity is set to 0.1, ensuring that even a weak cue can still be perceived by the user. The baseline stimulus intensity is set at 0.5, serving as the benchmark for standard guided intensity. This is the intensity adjustment coefficient, with a value of 0.8, used to adjust the weight of the effect of muscle activation deviation on stimulus intensity; For the first Target activation level of muscle groups; The output of step S1 Real-time muscle group activation level. Among them, if... Insufficient muscle activation , Increase accordingly and strengthen guidance for users to exert their efforts; if Muscle activation achieved. Approaching Maintain basic guidance.

[0117] Integrating parameters of site of action, stimulation rhythm, and stimulation intensity, guiding tactile feedback instructions. The final form is defined as: The instruction is sent directly to the distributed tactile actuator group of the actuator array to ensure that the stimulation parameters are accurately matched with the user's current adjustment needs.

[0118] The core of the target support pressure parameter is progressive desensitization. This means providing strong support in the early stages of rehabilitation to ensure joint stability and safety, and gradually reducing the support intensity in the later stages to encourage users to control their joints independently. Simultaneously, dynamic compensation is achieved by incorporating real-time muscle activation status to avoid over-reliance on external support and balance the needs of safety protection and active rehabilitation. The formula for calculating the target support pressure is:

[0119] ;

[0120] In the formula, The target support pressure is measured in kPa and its range is constrained to [7, 15] kPa (based on the safety threshold of blood circulation in the lower limbs and the joint support requirements; below 7 kPa, the support is insufficient, and above 15 kPa, it may compress blood vessels and affect blood circulation). The basic safety support pressure is set at a default value of 15 kPa, corresponding to the early stage of recovery. The maximum support pressure ensures that the joint receives sufficient stability. The default value for adjusting the pressure desensitization range is 8 kPa, which represents the maximum reduction in support pressure as the recovery trend progresses. , where P min =7kPa is the minimum effective support pressure); This is the muscle activation pressure coefficient, with a default value of 3kPa. It is used to compensate for insufficient muscle activation and to prevent joint instability caused by insufficient muscle activation. This is a muscle activation deviation term, and only non-negative values ​​are taken. That is, pressure compensation is only performed when muscle activation is insufficient, and no additional compensation is performed when activation reaches the target.

[0121] Furthermore, this formula can be broken down into a desensitization core term and a security compensation term, where the desensitization core term... , The larger the size (the better the recovery), the lower the support pressure, gradually reducing the user's reliance on external support; safety compensation items. In cases of insufficient muscle activation, appropriately increasing support pressure can assist joint stability and prevent joint injury due to insufficient muscle exertion during adjustment. For example, in the early stages of rehabilitation ( )and hour, Because it exceeded the maximum pressure threshold, the final pressure was set at 15 kPa (strong support + compensation); in the later stages of recovery ( )and hour, (Weak support, no compensation) to achieve precise adaptation of support pressure as the rehabilitation process progresses.

[0122] In this invention, to ensure that the generated desensitization guidance strategy always remains within the safety boundary and to avoid invalid guidance or security risks due to parameter calculation deviations, a validity pre-verification is required after the strategy is generated; the logical judgment condition for the pre-verification is as follows:

[0123] ;

[0124] That is, the target support pressure must be within the range of [7, 15] kPa, the stimulation intensity must be within the range of [0.1, 0.9], and the application site must belong to the preset tactile zone set S. If the verification strategy is invalid (e.g., the calculation result exceeds the above parameter constraints), it will automatically adjust to a "conservative strategy": target support pressure (15kPa), Stimulation Intensity (0.5) Site of action (Default partition, such as the area directly in front of the joint), and record the reasons for invalidity, such as parameter calculation exceeding constraints, abnormal real-time data, etc., to provide data basis for model and strategy optimization.

[0125] S5: Execute safety guidance arbitration, specifically: compare the predicted user motion state corresponding to the desensitization guidance strategy with the real-time safety range defined by the personalized dynamic safety boundary model, and combine it with the psychological stress level, and generate a post-arbitration adjustment instruction according to the preset collaborative arbitration logic; wherein, the collaborative arbitration logic is configured as follows: when the predicted user motion state is within the real-time safety range and the psychological stress level is lower than the first threshold, output the first arbitration result that allows the execution of the desensitization guidance strategy; when the predicted user motion state exceeds the real-time safety range or the psychological stress level is higher than the second threshold, output the second arbitration result that adjusts, terminates or converts the desensitization guidance strategy to a soothing mode.

[0126] To avoid the lag in real-time response, the system first predicts the joint motion state in the near future based on the real-time biomechanical data output in step S1, thus determining in advance whether it will exceed the safety boundary. The prediction target is the future... Motion state feature vector at time (default 0.3s, balancing motion real-time performance and actuator response cycle) This vector has the same dimension as the feature vector constructed in step S2 for the personalized dynamic safety boundary model, and includes roll angle, pitch angle, yaw angle, three-dimensional position coordinates, and instantaneous velocity. .

[0127] The prediction process employs a linear prediction model (LPC), using the latest d+1 time-sequence feature vector sequence output in step S2. The input is T, where d is the prediction order (default 3, balancing prediction accuracy and computational cost). s =0.01s (consistent with the sampling frequency in step S1), the mathematical expression of the linear prediction model is:

[0128] ;

[0129] In the formula The prediction coefficients are obtained through training using the least squares method; 1000 sets of "real-time feature vectors" are collected during the first use. Post-actual feature vector The data pairs are solved using the least squares method. This causes prediction error Minimum, default post-training coefficients [0.8, 0.15, 0.03, 0.02]. For the first Each historical moment feature vector, and the historical feature data output by S1, provide a basis for prediction.

[0130] To ensure that the prediction results are consistent with the dimensions of the personalized dynamic safety boundary model in step S2, the following measures are taken: Perform Z-score standardization, ensuring the standardization formula remains consistent with step S2: ;in This is the historical feature mean vector calculated in step S2. This is the standard deviation vector of historical features.

[0131] Collaborative arbitration logic centers on physical safety constraints (personalized dynamic safety boundaries), psychological safety constraints (psychological stress levels), and rehabilitation goal constraints (desensitization guidance strategies). It achieves a dynamic balance among these three through explicit mathematical logic. The arbitration outcome is divided into a first arbitration outcome (…). ) and the second arbitration result ( Two categories.

[0132] The core judgment parameters include safety probability density, safety probability threshold, psychological stress threshold, and psychological stress level; these are determined by the standardized prediction feature vector. Substituting into a Gaussian mixture model (GMM), the safety probability density for predicting motion states is calculated. The security probability threshold defined in step S2 (Default 0.7) represents the minimum probability requirement for predicting that the motion state falls within a safe range; the psychological stress level output in step S3. (Range [0,1]) must be equal to the first threshold. (Default 0.3, low stress threshold, at which point the user's mindset is stable and accepting guidance), Second threshold (Default 0.7, high stress threshold, at which point the user is panicked / nervous and requires intervention) for comparison.

[0133] Arbitration Result There are two categories, and the triggering logic is defined using the "AND / OR" logical operator, whose mathematical expression is:

[0134] ;

[0135] In the formula The "AND" statement indicates that the output is only generated when both "the predicted motion state is within a safe range" and "the psychological stress level is below a low threshold" are simultaneously satisfied. Allows the implementation of desensitization guidance strategies; "" indicates a logical "OR" condition; the output is triggered if either the predicted motion state exceeds the safe range or the psychological stress level exceeds the high threshold. The desensitization guidance strategy needs to be adjusted, suspended, or switched to a soothing mode. For the intermediate state ( but (i.e., safe movement but moderate psychological stress), default output. However, the intensity of desensitization guidance needs to be reduced, such as multiplying the stimulus intensity by 0.8 and the pressure adjustment step size by 0.5, to balance rehabilitation guidance and psychological adaptation.

[0136] In this embodiment of the invention, to avoid a "one-size-fits-all" approach to strategy processing, the second arbitration result ( The gradient processing logic is used to refine the tuning based on the triggering conditions. ), abort ( ), appeasement + termination ( There are three categories, and their mathematical expressions are:

[0137] ;

[0138] In the formula The fault tolerance threshold for safety probability (default 0.1). The stress level tolerance threshold is 0.1 by default. The tuning operation specifically involves reducing the intensity of desensitization guidance (stimulus intensity × 0.5, target support pressure + 2 kPa) and narrowing the guidance range; the cessation operation refers to pausing the desensitization guidance strategy and maintaining the current support pressure and tactile feedback state; the soothing + cessation operation involves pausing guidance while initiating a soothing mode of rhythmic vibration or gentle thermal stimulation.

[0139] In one embodiment of the invention, when a second arbitration result is output, the collaborative arbitration logic is further configured to: generate a soothing tactile feedback instruction to reduce the user's tension, the soothing tactile feedback instruction including a rhythmic vibration or a gentle thermal stimulation pattern. The soothing tactile feedback instruction is part of the second arbitration result. The system provides customized commands, the core of which is to alleviate the user's tension and panic through rhythmic vibration or gentle heat stimulation. The command parameters are related to the level of psychological stress. The higher the stress level, the more soothing the reassurance signal. Rhythmic vibration is the core reassurance mode, with a vibration frequency... (Unit: Hz) and vibration intensity The mathematical model for (dimensionless, range [0,1]) is as follows:

[0140] ;

[0141] ;

[0142] In the formula The maximum soothing vibration frequency (default 1.0Hz, corresponding to the stress level) is set. ), The minimum soothing vibration frequency (default 0.3Hz, corresponding to the stress level) is used. ), Maximum soothing vibration intensity (default 0.4, set based on human tactile comfort threshold).

[0143] In this invention, mild thermal stimulation is used as an alternative soothing mode, suitable for scenarios where users are sensitive to vibration or have extremely high stress levels (L(t)>0.9), and the thermal stimulation temperature... The mathematical model for (unit: °C) is:

[0144] ;

[0145] In the formula (normal human body temperature), (Use gentle heat to avoid burns). The higher the stress level, the closer the temperature should be to the body's normal temperature to avoid exacerbating discomfort due to high temperatures.

[0146] The final form of a soothing tactile feedback instruction is:

[0147] ;

[0148] This instruction is issued in conjunction with the "stop guidance" instruction to ensure that users can relax psychologically while pausing rehabilitation guidance.

[0149] In this embodiment of the invention, post-arbitration mediation This is the direct input for the actuator array operation in step S6. The corresponding operations are integrated based on different arbitration results, with a unified format and no additional conversion required, as detailed below:

[0150] when hour, This allows the execution of the guided haptic feedback instructions generated in step S4. With target support pressure parameters ;when (Tuning) That is, executing the tuned guided tactile feedback command and the target support pressure parameters; when (When) it is suspended, That is, maintaining the current tactile feedback state and supporting pressure. (Output pressure at the previous moment in step S6); when (When appeasing and suspending) This means executing soothing tactile feedback instructions and maintaining the current support pressure.

[0151] S6: Based on the arbitrated adjustment command generated in step S5, the actuator array is driven to perform the corresponding adjustment operation. Specifically, when the arbitrated adjustment command includes a first arbitration result, the distributed tactile actuator group in the actuator array generates directional tactile stimulation at the location corresponding to the action site information in the command on the user's body surface according to the guided tactile feedback command, and / or the pressure regulating actuator in the actuator array adjusts the bandage pressure to the target support pressure parameter; when the arbitrated adjustment command includes a second arbitration result, the distributed tactile actuator group generates tactile stimulation corresponding to the soothing tactile feedback command, and / or the pressure regulating actuator adjusts the bandage pressure to a preset safe support pressure.

[0152] In this embodiment of the invention, the number of distributed tactile actuator groups in the actuator array is consistent with the number of preset zones of the bandage. For example, the ankle joint protective bandage has four actuators corresponding to the inner, outer, front, and rear sides. A miniature eccentric motor can be used, with a maximum vibration amplitude of 1.5 mm and a response time of no more than 50 ms, which can quickly output directional tactile feedback. The pressure regulating actuator can be a combination of a miniature air pump and a solenoid valve, with a pressure regulating range of 5~20 kPa, an regulating accuracy of ±0.5 kPa, and a response time of ≤100 ms, which can smoothly adjust the bandage support pressure. A flexible PTC heating pad can be selected as a thermal stimulation actuator, with a temperature regulating range of 36~40℃, a temperature control accuracy of ±0.3℃, and a response time of ≤200 ms, which is used for gentle soothing under high stress.

[0153] Among them, when the post-arbitration mediation order is the first arbitration result ( When the S4-generated desensitization guidance strategy is executed, the coordinated action of tactile guidance and pressure adaptation is achieved. For the distributed tactile actuator group, the target area is activated first. The bandage's preset partitions correspond one-to-one with the tactile actuators. The activation logic is as follows:

[0154] ;

[0155] in The partition number corresponding to the executor is used to activate only the part that functions in the instruction. Matching actuators ensure the directionality of haptic feedback.

[0156] For vibration rhythm, the actuator vibration frequency is directly adapted to the stimulation rhythm in the command. ,Right now , The actual vibration frequency of the actuator is controlled by the duty cycle of the PWM signal. For example, a frequency of 3Hz corresponds to a PWM period of 333ms.

[0157] Vibration intensity is then determined through a mathematical model. The formula for converting the range [0,1] to the actual vibration amplitude is as follows: In the formula The maximum vibration amplitude of the actuator (default 1.5mm) is constrained. A min =0.15mm), ensuring that the minimum strength is perceptible and the maximum strength does not cause discomfort. For example ,but The actuator operates at an amplitude of 0.9mm. Frequency at Vibration of the affected area.

[0158] Pressure regulating actuator In this state, its core is to achieve the target and support the pressure. Precise and stable adjustment. First, based on the current bandage pressure... (Data collected in real time by a pressure sensor) and The deviation determines the direction of the action, when Inflation is performed at the appropriate time. Maintain pressure holding state when The deflation action is performed at the time, where δ P =0.5kPa is the allowable pressure deviation.

[0159] To avoid discomfort caused by sudden pressure changes, a regulating rate is set. In the formula To maximize the adjustment rate and avoid excessively rapid inflation / deflation that could compress blood vessels or cause insufficient support, ; For maximum pressure difference, adjust at maximum rate when the pressure difference exceeds 8 kPa; otherwise, reduce the rate proportionally to ensure smooth operation. For example, P current =8kPa, P target =12kPa, then , The inflation process lasted 4 seconds, and the pressure steadily increased from 8 kPa to 12 kPa.

[0160] It should also be noted that data is collected every 10ms during the execution process. ,pass Dynamically adjust the adjustment rate to ensure precise pressure approximation. The error does not exceed .

[0161] When the post-arbitration mediation order is the second arbitration result ( When this occurs, adjust, stop, or soothe actions based on the specific sub-result. If it is adjustment, then ( The tuning action weakens the parameters of the desensitization guidance strategy. The tactile instruction tuning formula is:

[0162] ;

[0163] ;

[0164] In the formula (Intensity tuning factor) (Rhythm tuning coefficient), which means that the stimulus intensity is halved and the rhythm is slowed down.

[0165] The pressure command tuning formula is ,in This enhances joint stability by increasing support pressure.

[0166] If it is to suspend execution ( Aborting the action maintains the current state, and the tactile actuator retains the vibration parameters from the previous moment. , The pressure regulating actuator performs a pressure-maintaining action, closing the air pump / solenoid valve. This helps to avoid sudden changes in condition that could affect joint stability.

[0167] If it is to appease and suspend execution ( If the desensitization process is terminated, soothing feedback will be initiated simultaneously. The default soothing method is rhythmic vibration, which will transmit the soothing instructions. The formula for converting this into actuator actions is:

[0168] ;

[0169] ;

[0170] Among the constraints , This ensures the vibration is gentle and non-irritating. For example, , ,but The actuator vibrates at an amplitude of 0.41 mm and a frequency of 0.77 Hz.

[0171] When the user's psychological stress level During periods of extreme stress, a gentle thermal stimulation is activated for soothing (thermal stimulation actuator). The temperature mapping formula is as follows: ,constraint Heating will be automatically cut off if the temperature exceeds the specified range (hardware protection); the actual temperature will be collected every 50ms during execution. If the temperature deviation exceeds 0.3℃, adjust the heating power promptly to avoid the risk of burns.

[0172] It is important to emphasize that, to ensure absolute safety during execution, an execution status monitoring and anomaly handling mechanism can be set up. The operating current of the tactile actuator and air pump is monitored in real time using a current sensor. If the current exceeds the rated range (e.g., the rated current of the tactile actuator is 50mA, and it exceeds 80mA), it is determined that the actuator is faulty, and the actuator operation is immediately stopped and the fault information is recorded. Simultaneously, multiple safety threshold protections are set, and pressure-related issues are also addressed. (Exceeding the safety limit), immediately activate the solenoid valve to release gas to 15 kPa; regarding temperature... (If the safety limit is exceeded), immediately cut off the thermal stimulation power supply. All abnormal situations will send an abnormal signal to the central processing unit.

[0173] In one embodiment of the present invention, the present invention further includes S7: storing user response data and adjustment result data in each adjustment process to a data storage module; the user response data is the biomechanical data newly collected and processed after executing step S6, and the difference data obtained by comparing it with the corresponding biomechanical data before executing S6; the adjustment result data includes at least the type and parameters of the adjustment instruction that has been executed.

[0174] User response data is the core feedback data stored in S7, used to quantify the actual impact of the adjustment action on the user's motor state and muscle function. It is defined as the difference in the user's biomechanical data before and after performing the S6 adjustment action. This data has a clearly defined time window constraint: biomechanical data before adjustment. The average biomechanical data collected and processed in step S1 within 100ms before the adjustment action is performed is taken to avoid errors caused by transient signal fluctuations; the biomechanical data after adjustment... The data is taken from the average of the biomechanical data collected and processed in step S1 within 500ms after the adjustment action is performed, ensuring that the user feedback reaches a stable state before data extraction. User response data. Presented in vector form, it encompasses changes in three core dimensions: posture, position, and muscle activation. Its mathematical calculation model is as follows: Among them, the attitude difference vector (Unit: rad) reflects changes in joint roll, pitch, and yaw angles after adjustment, and can verify whether tactile guidance prompts the joint to return to a safe posture; position difference vector (Unit: m), representing changes in the three-dimensional spatial position of a joint, used to determine whether the center of gravity has been adjusted according to the guiding direction; muscle activation difference vector. (Dimensionless, range [0,1]), reflecting changes in the activation intensity of the target muscle group, is a key indicator for assessing whether desensitization guidance enhances the muscle's autonomous activation ability. It should be noted that if a certain difference value exceeds a reasonable range (e.g., ...), If any of the above conditions are met, the data is considered abnormal and will be temporarily stored but not used in subsequent optimization calculations to ensure the validity of the training samples.

[0175] The adjustment result data is the execution-end data stored in S7, used to fully record the decision-making basis and actual execution parameters of this adjustment, clarifying "what specific actions were performed in this adjustment," and providing accurate input parameter samples for subsequent optimization. This data is entirely based on the post-arbitration adjustment instruction in step S5. The actual actions generated by the executor in step S6 are defined in structured vector form as follows:

[0176] ;

[0177] Among the types of arbitration results For the enumerated value (R1 / R) 21 / R 22 / R 23 The decision type of this adjustment is directly taken from the arbitration result of step S5, and is marked as per the actual output support pressure (allowing guidance / tuning / suspension / appeasement). (Unit: kPa) The final stable pressure (average within 1 second after adjustment) collected by the pressure sensor in step S6 is recorded, along with the actual support pressure achieved by the actuator (including minor execution errors); the actual tactile stimulation intensity is also recorded. (Dimensionless, range [0,1]) is derived from the driving current of the tactile actuator in step S6, avoiding intensity deviations caused by actuator malfunctions; actual tactile stimulation rhythm (Unit: Hz) The actual frequency of tactile feedback is accurately reflected by calculating the PWM signal period of the tactile actuator in step S6; the actual soothing thermal stimulation temperature. (Unit: °C) Only when the arbitration result is R 23 Effective when (soothing + stopping), the value is taken from the final temperature collected by the thermal stimulation sensor in step S6; in non-soothing mode, this value is null; adjust the duration. (Unit: s) is the time it takes for the actuator to reach a steady state in terms of pressure, touch, or temperature from startup. It is calculated by the system clock and reflects the smoothness of the adjustment action; adjustment timestamp. Recording the system time at the start of the adjustment process is the core identifier for linking user response data and adjustment result data. Depending on the actual application requirements, additional optional data items such as environmental parameters (e.g., temperature, humidity) and sensor status (signal quality score) can be added to further enhance data completeness.

[0178] To adapt to the machine learning algorithm call in step S8, this invention uses JSON structured format to store data. This format has clear fields, can be directly parsed, and supports rapid extraction of target features.

[0179] In one embodiment of the present invention, the present invention further includes S8: using the user response data and adjustment result data stored in step S7, the generation logic of the personalized dynamic safety boundary model and the desensitization guidance strategy is periodically optimized and updated through a machine learning algorithm; wherein, the optimization and update of the generation logic includes adjusting the adjustment step size of the intensity and frequency parameters of the guiding haptic feedback command and the target support pressure parameter.

[0180] The raw data stored in step S7 needs to be filtered and feature-engineered to transform it into training samples that can be directly used by machine learning algorithms, ensuring optimization quality. First, sample filtering is performed on the dataset accumulated and stored in S7. (in To remove invalid data from the cumulative number of stored data entries, valid samples must simultaneously meet the following constraints: absolute value of attitude difference. Absolute value of muscle activation difference (Dimensionless) Actual output pressure Actual tactile stimulation intensity (Dimensionless) and without actuator fault markers, the effective sample set was obtained after screening. ,in To be a valid sample size, it must satisfy the following conditions: Only then can the optimization process be triggered.

[0181] After sample selection, feature engineering is performed to extract the input and label features required for training. Input features From the adjustment results data Extracted from, including the arbitration result Target pressure Stimulation intensity Stimulation frequency Rehabilitation trend factors And the historical psychological stress level of step S3 is linked by timestamps. Tag features From user response data Extracted from [the data], it serves as a quantitative indicator characterizing the regulatory effect, and its calculation model is as follows:

[0182] ;

[0183] In the formula For the first Sample adjustment effect index ( The target muscle group number, For the first (mean muscle activation increased in the sample) (Attitude adjustment weights) (Muscle activation weight) (Maximum pose difference threshold) (Maximum activation difference threshold), final A higher value indicates a better adjustment effect.

[0184] For the Gaussian Mixture Model (GMM) in step S2, the incremental EM algorithm is used for parameter optimization to ensure that the safety boundary dynamically adapts to the user's motion characteristics and rehabilitation process. The core of model optimization is to incrementally update the GMM weights based on biomechanical feature data from valid samples. mean Covariance matrix At the same time, optimize the security probability threshold. .

[0185] The incremental update formula for the GMM model parameters is as follows:

[0186] The weight update formula is:

[0187] ;

[0188] The formula for updating the mean is:

[0189] ;

[0190] The formula for updating the covariance matrix is:

[0191] ;

[0192] In the formula , This is the vector of component mean values ​​before and after the update. The number of valid samples before optimization. For the newly added The sample belongs to the first The posterior probability of the component (calculated from the current GMM model). For the newly added The biomechanical feature vectors of the samples (with the same feature dimensions as in step S2) are updated incrementally, which ensures the model's adaptability to new data and avoids the resource consumption caused by full retraining.

[0193] Safety probability threshold The optimization and adjustment effects are linked to ensure the dynamic rationality of the safety boundary. The optimization formula is as follows: ,in (Adjust the threshold step size to avoid adjusting too quickly). This represents the average adjustment effect of the effective samples. A larger value indicates a better guidance effect within the current safety boundary. γ can be appropriately reduced to expand the safety range and encourage users to gradually improve their motor skills.

[0194] For the desensitization guidance strategy in step S4, we focused on precise parameter tuning of core parameters, including the tactile feedback intensity adjustment coefficient, the tactile feedback frequency correlation coefficient, and the target support pressure adjustment step size, to ensure that the strategy is more in line with the user's response habits and improve the effectiveness of guidance.

[0195] Haptic feedback intensity adjustment coefficient The optimized model for (the relationship between stimulus intensity and muscle activation deviation) is as follows:

[0196] ;

[0197] in , The intensity adjustment coefficient before / after the update, η=0.05 (learning rate, controlling the adjustment range), Y target =0.7 (target adjustment effect) For the first Target muscle activation of the sample No. The actual muscle activation level of the sample. If the average modulatory effect is lower than the target and the muscle activation level deviation is large, the level will be increased. The response of the intensity to the activation bias is strengthened; conversely, it is reduced. Avoid excessive stimulation.

[0198] tactile feedback frequency Optimization is achieved by adjusting its correlation with the recovery trend factor. The correlation coefficient is achieved, and the optimized formula is as follows: ,in ( This is the frequency adjustment coefficient (default value is 1). If the average adjustment effect is greater than 0.5 (good effect), it will be increased. To enhance guidance, the frequency should be increased during the initial stages of rehabilitation; if the effect is less than 0.5, the frequency should be reduced. This makes the frequency changes smoother.

[0199] Target support and resistance adjustment step size The optimization model is as follows:

[0200] ;

[0201] in , Adjust the step size for pressure before / after the update. (Initial step size) For the first The actual pressure change of the sample (Actual pressure change). This represents the maximum pressure difference. If the pressure change is small and the adjustment effect is good, the step size will be increased to accelerate pressure adaptation; if the pressure change is large but the effect is poor, the step size will be decreased to avoid discomfort caused by sudden pressure changes.

[0202] The optimized parameters will replace the original parameters in step S4. The new strategy generation logic is: stimulus intensity Stimulation frequency Target pressure (Pressure adjustment rate correlation) ).

[0203] The optimized model and strategy need to undergo dual verification through offline validation and online trial operation to ensure safety and effectiveness before formal deployment. During the offline validation phase, the most recent 10% of valid samples will be used for testing. If the optimization results are not satisfactory, the average effect will be adjusted. Increase ≥10%, and the proportion of abnormal movement within the safety boundary ( If the number of samples (in this case / total number of samples) is ≤5%, the validation is considered successful. During the online trial operation phase, the validated model and strategy will run with "mixed weights". The weight of the optimized parameters will be 0.3, and the weight of the original parameters will be 0.7. 500 samples will be collected continuously. If no safety anomalies occur (such as stress exceeding the threshold, sudden increase in user stress level, etc.), the optimized parameters will be completely switched to, and the optimization will be implemented.

[0204] According to the sports protective bandage system of this invention, the invention acquires user movement and physiological data in real time through multimodal sensing and signal processing, constructs and continuously updates a personalized dynamic safety boundary model, solving the problem that general protection cannot adapt to individual differences; by calculating the level of psychological stress, psychological safety is incorporated into decision-making, avoiding secondary risks caused by panic; based on real-time data and rehabilitation trends, a desensitization guidance strategy is generated, and through the linkage of tactile and pressure regulation, muscle exertion and center of gravity adjustment are actively guided while ensuring joint safety, effectively overcoming the "disuse" problem caused by traditional protection; finally, through a safety guidance arbitration mechanism, intelligent decision-making is carried out in coordination of physical safety boundaries and psychological state, achieving a dynamic and personalized balance between immediate protection and long-term active rehabilitation.

[0205] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A motion guard bandage system comprising a multi-modal sensor array, a central processing unit and an array of actuators, using a smart pressure-aware motion guard bandage linkage adjustment method, characterized in that, The method includes the following steps: S1: The user's biomechanical and physiological signals are collected in real time through the multimodal sensor array, and the central processing unit processes the collected signals to obtain processed biomechanical and physiological data; wherein, the processed biomechanical data includes at least the real-time motion trajectory and posture data of the joints and the activation feature data of the target muscle group. S2: Based on processed biomechanical data from historical records, a personalized dynamic safety boundary model representing the safe range of motion of a user's joints is constructed and continuously updated through machine learning algorithms. S3: Based on the processed physiological data acquired in real time, a psychological stress level that quantifies the user's panic or tension is calculated. S4: Based on the current processed biomechanical data and historical rehabilitation trends, generate a desensitization guidance strategy aimed at guiding users to actively shift their center of gravity or adjust their posture. S5: Execute safety guidance arbitration, specifically: compare the predicted user movement state corresponding to the desensitization guidance strategy with the real-time safety range defined by the personalized dynamic safety boundary model, and simultaneously combine the psychological stress level, generating a post-arbitration adjustment instruction based on preset collaborative arbitration logic; wherein, the collaborative arbitration logic is configured as follows: when the predicted user movement state is within the real-time safety range and the psychological stress level is lower than a first threshold, output a first arbitration result allowing the execution of the desensitization guidance strategy; when the predicted user movement state exceeds the real-time safety range or the psychological stress level is higher than a second threshold, output a second arbitration result to tune, terminate, or convert the desensitization guidance strategy to a soothing mode; S6: Based on the arbitrated adjustment command generated in step S5, drive the actuator array to perform the corresponding adjustment operation; Specifically, the personalized dynamic security boundary model constructed and continuously updated in step S2 includes: By analyzing the historical sequence of the real-time motion trajectory and posture data of the joint, a multidimensional parameter space boundary is constructed using clustering or probability distribution methods to characterize the safe range of motion of the joint under different activity modes. This serves as the personalized dynamic safety boundary model, and the threshold of the multidimensional parameter space boundary is periodically updated using newly acquired real-time motion trajectory and posture data of the joint. Step S4 generates the desensitization guidance strategy, specifically as follows: Based on the real-time motion trajectory and posture data of the joint and the activation characteristic data of the target muscle group, combined with historical rehabilitation trend data, guiding tactile feedback instructions and / or target support pressure parameters are generated to guide the user to perform center of gravity shift or posture adjustment. The guiding tactile feedback instructions include information on the site of action, stimulation mode and intensity. Step S6 specifically includes: When the adjustment instruction after arbitration includes the first arbitration result, the distributed tactile actuator group in the actuator array is driven to generate directional tactile stimulation at the position on the user's body surface corresponding to the action site information in the guided tactile feedback instruction, and / or the pressure regulating actuator in the actuator array is driven to adjust the bandage pressure to the target support pressure parameter. In step S5, when the second arbitration result is output, the collaborative arbitration logic is further configured as follows: Generate soothing tactile feedback instructions to reduce user tension, including rhythmic vibration or gentle thermal stimulation patterns. When the adjustment command after arbitration includes the second arbitration result, the distributed tactile actuator group is driven to generate tactile stimulation corresponding to the soothing tactile feedback command, and / or the pressure regulating actuator is driven to adjust the bandage pressure to a preset safe support pressure.

2. The sports protection bandage system of claim 1, wherein Step S1 specifically includes: S11: Acquire multi-axis kinematic signals of the joint through the inertial measurement unit, acquire electromyographic signals of the target muscle group through the electromyographic sensor module, and acquire skin conductance signals of the user through the skin conductance sensor. S12: The central processing unit processes the multi-axis kinematic signals to obtain the real-time motion trajectory and posture data of the joint, processes the electromyographic signals to obtain the activation characteristic data of the target muscle group, and processes the skin conductance signals to obtain the physiological data.

3. The sports protective bandage system according to claim 1, characterized in that, The psychological stress level is calculated in step S3 as follows: By analyzing the amplitude change rate and spectral characteristics of the skin conductance signal in the physiological data, a quantitative value of psychological stress level reflecting the degree of excitation of the user's autonomic nervous system is calculated.

4. The sports protective bandage system according to claim 1, characterized in that, The method further includes: S7: Store the user response data and adjustment result data of each adjustment process into the data storage module; the user response data is the biomechanical data newly collected and processed after executing step S6, and the difference data obtained by comparing it with the corresponding biomechanical data before executing S6; the adjustment result data includes at least the type and parameters of the adjustment command executed.

5. The sports protective bandage system according to claim 4, characterized in that, Also includes: S8: Using the user response data and adjustment result data stored in step S7, the generation logic of the personalized dynamic safety boundary model and the desensitization guidance strategy is periodically optimized and updated using a machine learning algorithm; wherein, the optimization and update of the generation logic includes adjusting the intensity and frequency parameters of the guiding haptic feedback instruction and the adjustment step size of the target support pressure parameter.

Images