Ankle joint rehabilitation brace system based on dynamic stability perception and closed-loop regulation

Abstract

The present application relates to the technical field of medical rehabilitation and intelligent wearable device, and particularly relates to an ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop regulation. The system comprises an individualized and modularized brace body, a distributed ankle-foot monitoring system, an individualized evaluation and decision module, and a closed-loop regulation and doctor-patient interaction module. The system synchronously collects multi-dimensional features such as kinematics, dynamics and electromyography in real time through multi-modal sensors embedded in the brace, extracts gait feature vectors through edge computing, and uploads them to the cloud for re-injury risk prediction and constraint tension recommendation based on the individualized model of LSTM and attention mechanism. When a high-risk varus event is identified, the system drives the micro actuator to actively increase the tension of the lateral constraint belt within milliseconds of delay, realizes immediate mechanical intervention, and automatically releases the tension to restore comfort after the risk is removed. The system also forms a closed-loop of whole-cycle rehabilitation management through data interconnection between the patient APP and the doctor platform.

Description

Technical Field

[0001] This invention belongs to the field of medical rehabilitation and smart wearable device technology, specifically relating to an intelligent ankle joint rehabilitation brace system that integrates ankle and foot posture monitoring, load assessment, muscle function analysis, individualized rehabilitation suggestions and active intervention for conservative treatment and postoperative rehabilitation of the ankle joint. Background Technology

[0002] The ankle joint is one of the most important weight-bearing and movement joints in the human body. Injuries such as lateral collateral ligament injury, medial deltoid ligament injury, distal tibiofibular syndesmosis injury, chronic ankle instability, osteochondral injury of the talus, postoperative functional impairment after ankle fracture surgery, and gait abnormalities after Achilles tendon repair surgery are common in both athletes and the general population. These conditions often lead to increased pain, swelling, recurrent sprains, and re-injury risk, seriously affecting patients' daily life and return to sports.

[0003] Ankle braces, as important auxiliary devices for conservative treatment and postoperative rehabilitation, traditionally provide external protection for the ankle joint mainly through elastic compression, semi-rigid support plates, cross straps, and inversion / variversion restriction structures. For short-term fixation after acute injury, early postoperative protection, and general sports protection, traditional ankle braces have advantages such as relatively simple structure, ease of wear, and low cost, and are therefore widely used in clinical and sports settings.

[0004] However, ankle rehabilitation is not simply about fixation and support; it is a dynamic process involving local tissue healing, restoration of joint range of motion, gait reconstruction, improvement of balance, synergistic muscle control, and prevention of re-injury. Currently, in the application of ankle braces, clinical assessments of brace efficacy and rehabilitation progress still primarily rely on physical examinations, imaging evaluations, and patient subjective descriptions during outpatient follow-ups. This approach fails to accurately reflect the patient's gait rhythm, ankle inversion / eversion posture, plantar load distribution, and high-risk landing behaviors in home or real-world environments, resulting in a lack of continuous and objective rehabilitation monitoring capabilities. While some existing smart braces have attempted to integrate pedometers or single inertial sensors, they often only provide step counts, activity time, or rough posture changes, making it difficult to comprehensively assess multidimensional characteristics such as ankle kinematics, plantar dynamics, stabilizing muscle recruitment status, and local tolerance. The monitoring indicators remain limited.

[0005] Meanwhile, existing devices mostly remain at the level of data recording and simple threshold reminders. They cannot combine patient age, weight, disease type, surgical procedure, postoperative time, pain feedback and rehabilitation stage to dynamically model and individualize the gait quality, re-injury risk, training load and brace support parameters. They lack the ability to make truly individualized decisions for the ankle rehabilitation process.

[0006] Furthermore, even if problems such as abnormal landing, excessive lateral load, or gait asymmetry can be identified, existing devices still struggle to provide timely support enhancement and tension adjustment through the brace itself, or to update training prescriptions and intervene through mobile terminals and doctor platforms. The lack of an effective closed-loop control and remote doctor-patient collaboration mechanism means that the brace remains a passive device that can be worn and recorded but is difficult to intervene in.

[0007] Therefore, there is an urgent need to construct an intelligent rehabilitation brace system for the entire process of conservative treatment and postoperative rehabilitation of the ankle joint. Based on traditional ankle support, this system should further integrate ankle and foot posture monitoring, plantar load detection, muscle function assessment, individualized analysis, and closed-loop control technology to achieve continuous perception of ankle joint kinematics, dynamics, muscle synergy, and local tolerance. It should also dynamically optimize the brace support strength, restraint method, training tasks, and gait suggestions based on the patient's individual characteristics and real-time rehabilitation status, ultimately forming a closed-loop rehabilitation management model of perception, analysis, decision-making, and execution. Summary of the Invention

[0008] This invention addresses the shortcomings of existing ankle braces in rehabilitation management. These braces only provide passive support, have limited monitoring indicators, and lack individualized dynamic decision-making capabilities for the rehabilitation process. Consequently, they cannot effectively suppress abnormal postures during high-dynamic activities or dynamically optimize brace parameters based on individual patient characteristics and real-time rehabilitation status. To solve these problems, this invention provides an ankle rehabilitation brace system based on dynamic stability sensing and closed-loop control. This system deeply integrates a distributed multimodal sensor network, edge computing unit, and microelectromechanical actuator into a personalized brace body, and collaborates with a cloud-deployed individualized assessment and decision-making model and a doctor-patient interaction terminal to construct a complete closed-loop rehabilitation management architecture.

[0009] This invention provides an ankle rehabilitation brace system based on dynamic stability sensing and closed-loop control. The system includes an individually molded modular brace body, a distributed ankle and foot monitoring system, an individualized assessment and decision-making module, and a closed-loop adjustment and doctor-patient interaction module. The individually molded modular brace body serves as the system's basic physical platform and mechanical interface. The distributed ankle and foot monitoring system is embedded in a pre-set mounting position within the brace body to collect multi-source heterogeneous sensor data. The individualized assessment and decision-making module runs on an edge computing unit and a cloud server, and is used to fuse and analyze sensor data and patient-specific characteristic information to generate control commands and rehabilitation suggestions. The closed-loop adjustment and doctor-patient interaction module drives the actuators on the brace body to generate mechanical responses according to the control commands and provides information feedback to the patient and doctor.

[0010] Furthermore, the individually molded modular brace body includes a distal calf support sleeve, a dorsal and heel support shell, a medial malleolar support plate, a lateral malleolar support plate, a cross restraint strap assembly, a replaceable plantar module, and a sensor and actuation module mounting position. The distal calf support sleeve and the dorsal and heel support shell are integrally molded using a non-uniform lattice-filled structure generated through a topology optimization algorithm based on three-dimensional scan data of the distal two-thirds of the patient's calf and the ankle and foot shape, employing selective laser sintering or fused deposition modeling. The porosity of the lattice-filled structure is 45% to 55% in non-load-bearing areas and 30% to 40% in load-bearing areas such as the anterior ridge of the calf sleeve, the heel wrapping area, and the support plate connecting base. The shell wall thickness in the load-bearing areas is 3.0 mm to 4.0 mm, and in the non-load-bearing areas it is 1.8 mm to 2.5 mm, thus meeting the bending and torsional stiffness requirements of different areas while keeping the overall weight within the range of 230 g to 420 g. The inner wall of the distal calf support sleeve is fitted with a removable flexible pad. This flexible pad is made of medical-grade silicone or thermoplastic polyurethane foam. The pad thickness is 7mm to 10mm in the areas corresponding to the medial malleolus, lateral malleolus, and the Achilles tendon groove, and 2mm to 3mm in other areas. The bottom of the heel support shell is embedded with a shock-absorbing module composed of a shear-thickening colloid and a honeycomb elastomer. When the impact velocity exceeds 2.5m / s, the shear-thickening colloid undergoes a shear-thickening effect, causing the energy storage modulus to jump from 0.8MPa to 1.2MPa to 5.0MPa to 8.0MPa to absorb impact energy.

[0011] Furthermore, the medial malleolar support plate and the lateral malleolar support plate are detachably fixed to the distal lower leg support sleeve and the connecting base on the dorsum of the foot and heel support shell via dovetail groove guide rails. The lateral malleolar support plate is made of carbon fiber reinforced polyetheretherketone composite material with a flexural modulus of 15 GPa to 20 GPa, and the medial malleolar support plate is made of glass fiber reinforced nylon with a flexural modulus of 5 GPa to 8 GPa, thereby forming an asymmetric support characteristic of high rigidity constraint on the outside and flexible deformation on the inside. The distal extension of the lateral malleolar support plate covers the base region of the fifth metatarsal bone, and a flexible thin-film pressure sensor is attached to its inner wall to detect the interface contact pressure between the lateral malleolus and the lateral malleolar support plate. The lateral malleolus support plate and the medial malleolus support plate each have three preset limit plate slots, corresponding to limit plates with plantar flexion / dorsiflexion limit angle differences of 10°, 20° and 30° respectively. The range of motion of the brace body in the sagittal plane can be adjusted by replacing the limit plates of different angle specifications.

[0012] Furthermore, the cross restraint strap assembly includes a first restraint strap and a second restraint strap; the first restraint strap and the second restraint strap are cross-arranged and fixed to the instep.

[0013] Furthermore, the micro-winding drive structure comprises a worm gear reducer motor, a ratchet one-way locker, and a winding reel. The output shaft of the worm gear reducer motor drives a worm gear pair with a reduction ratio of 64:1. The worm gear and the winding reel are coaxially connected. The winding reel has a diameter of 8mm, and its circumferential surface is provided with a spiral groove to accommodate the dynamic tension adjustment belt. The tension sensor is a strain gauge S-type load cell with a range of 0N to 50N, connected in series between the second end of the dynamic tension adjustment belt and the intersection of the first constraint belt. When the worm gear reducer motor receives a PWM drive signal from the individualized evaluation and decision module, the micro-winding drive structure dynamically adjusts the effective tension of the dynamic tension adjustment belt from a base value of 3N to 5N to a target value of 15N to 25N within a response time of 0.15s to 0.8s. This actively enhances the outer constraint strength when a high-risk inward rollover event is detected. The ratchet one-way locker maintains the current tension after the motor stops to prevent loosening. The miniature winding drive structure and tension sensor are both encapsulated in a housing with an IP54 protection rating. The housing is fixed to the mounting position on the rear outer side of the distal support sleeve of the lower leg by screws.

[0014] The replaceable foot module includes a base layer, a sensing layer, and an encapsulation layer. The base layer is made of ethylene-vinyl acetate copolymer with a Shore hardness of 60A to 70A, and its upper surface has positioning bosses that mate with the bottom recesses of the instep and heel support shell to form a detachable fit. The sensing layer is embedded inside the base layer and consists of a sensing array of 8 to 12 flexible capacitive pressure sensors arranged in a matrix. Each sensor has an effective area of ​​10mm × 10mm, a range of 0kPa to 500kPa, and a resolution of 2kPa to 5kPa. The sensor array is arranged to correspond to the first metatarsal head region, the second to third metatarsal head regions, the fourth to fifth metatarsal head regions, the lateral midfoot region, the medial midfoot region, the medial heel region, the lateral heel region, and the central heel region. Each sensor consists of two polyimide films printed with silver nanowire electrodes on top and bottom, and a microstructured ionogel dielectric layer sandwiched between them. The surface of the microstructured ionogel dielectric layer has a microcone array with a height of 50 μm to 80 μm and a spacing of 100 μm to 150 μm to improve capacitive response sensitivity. The encapsulation layer is a 0.3 mm thick thermoplastic polyurethane film, which is hot-pressed onto the sensing layer to provide a waterproof seal.

[0015] Furthermore, the distributed ankle-foot monitoring system includes a kinematic sensing unit, a dynamics and contact sensing unit, a muscle function sensing unit, a physiological and local microenvironment sensing unit, and an edge computing architecture. The kinematic sensing unit comprises a first nine-axis inertial measurement unit, a second nine-axis inertial measurement unit, and a third nine-axis inertial measurement unit. The first nine-axis inertial measurement unit is embedded in a mounting groove 5 cm distal to the tibial tuberosity at the anterior midline of the distal lower leg support sleeve. The second nine-axis inertial measurement unit is embedded in the corresponding region of the tarsal sinus of the foot dorsum support shell. The third nine-axis inertial measurement unit is embedded in the posterior edge of the lateral malleolar support plate. Each nine-axis inertial measurement unit integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. The sampling rate is uniformly set to 100 Hz, and noise is denoised using a fourth-order Butterworth low-pass filter with a cutoff frequency of 20 Hz. The Euler angles calculated by the first and second nine-axis inertial measurement units are used to obtain the real-time ankle plantar flexion / dorsiflexion angle and inversion / eversion angle through differential calculation. The third nine-axis inertial measurement unit serves as a reference benchmark to compensate for crosstalk of the overall trunk movement on the ankle joint angle calculation.

[0016] The dynamics and contact sensing unit includes a plantar pressure sensing array composed of the flexible capacitive pressure sensors, a thin-film pressure sensor integrated into the inner wall of the lateral malleolus support plate, a strain detection module arranged on the medial malleolus support plate, and a tension sensor. The strain detection module consists of a half-bridge circuit composed of two metal foil strain gauges arranged at a 45° angle, attached to the midpoint of the long axis of the medial malleolus support plate. It is used to monitor the bending strain generated by the medial malleolus support plate under the tendency of foot eversion and to estimate the eversion torque accordingly. The analog signals from the tension sensor, the plantar pressure sensing array, the thin-film pressure sensor, and the strain detection module are all connected to a 24-bit analog-to-digital converter (ADC), which synchronously samples the above channels at a sampling rate of 200Hz.

[0017] The muscle function sensing unit consists of a first surface electromyography (EMG) acquisition channel and a second surface EMG acquisition channel. The Ag / AgCl gel electrode pair of the first EMG acquisition channel is attached to a region 4cm to 6cm below the head of the fibula along the direction of the peroneus longus muscle myofibrils, with a center-to-center distance of 20mm. The electrode pair of the second EMG acquisition channel is attached to a region 3cm to 5cm lateral to the tibial tuberosity along the direction of the tibialis anterior muscle belly fibers, with a center-to-center distance of 20mm. The EMG signals from both channels are pre-amplified and bandpass filtered, with a passband frequency of 20Hz to 500Hz. After analog-to-digital conversion at a sampling rate of 1500Hz, the edge computing unit calculates the root mean square value of the sliding window with a length of 100ms and the integrated EMG value within the pre-activation time window in real time. The pre-activation time window is defined as the time interval 50ms before the initial contact of the foot with the ground.

[0018] The physiological and local microenvironment sensing unit includes a digital temperature sensor and a capacitive humidity sensor attached to the inner side of the flexible pad. The digital temperature sensor is a silicon-based bandgap temperature sensor with a resolution of 0.0625°C, and the capacitive humidity sensor uses a polyimide moisture-sensitive dielectric layer to detect the relative humidity inside the brace cavity.

[0019] Specifically, the edge computing architecture includes a microcontroller, flash memory, and a Bluetooth Low Energy and Wi-Fi dual-mode wireless communication module integrated into the mounting compartment at the rear of the distal calf support sleeve. The microcontroller is based on an ARM Cortex-M4 core with a clock speed of 168MHz and a flash memory capacity of 16MB. The firmware on the microcontroller executes a hierarchical sampling and feature extraction strategy: data from the first, second, and third nine-axis inertial measurement units are continuously acquired at a duty cycle of 100Hz; the plantar pressure sensor array is continuously acquired at a duty cycle of 50Hz; the surface electromyography (EMG) signals are acquired at a low duty cycle of 100Hz during static standing detected by the inertial measurement units, and switch to full-bandwidth acquisition at 1500Hz during motion determined by the resultant acceleration vector of the inertial measurement units exceeding the 1.2g threshold. The microcontroller performs the following edge computing tasks on-chip on the raw sensor data: S1. Using an extended Kalman filter algorithm, the attitude quaternion data of the nine-axis inertial measurement unit is fused with the plantar pressure center trajectory to estimate the ankle coronal plane moment in real time; S2. Using a gait event detection algorithm based on continuous wavelet transform, four gait phases are identified: heel strike, full foot support, heel lift-off, and toe lift-off; S3. For each gait cycle, a feature vector is extracted, which includes the maximum plantar flexion angle, maximum inversion angle, peak inversion velocity, peak lateral plantar pressure, root mean square value of peroneus longus preactivation, and values ​​of peroneus longus and tibialis anterior muscle. The system incorporates muscle co-contraction index, gait cycle duration, proportion of dual-support phase, and temperature and humidity change rates. S4. The individualized assessment and decision-making module uploads the feature vector to the cloud server via Bluetooth Low Energy or Wi-Fi. A real-time anomaly alarm is triggered only when any component of the feature vector exceeds a preset warning threshold, and the original data fragment is packaged and uploaded for further cloud verification. The preset warning threshold includes one or more combinations of conditions such as a maximum inversion angle exceeding 8°, a peak inversion angle velocity exceeding 350° / s, a peak lateral plantar pressure exceeding 180 kPa, and a peroneus longus pre-activation root mean square value less than 15% of the maximum voluntary contraction value. The hierarchical sampling and feature extraction strategy enables the system to achieve at least 12 hours of continuous operation with a 600mAh lithium polymer battery.

[0020] Furthermore, the individualized assessment and decision-making module is deployed on a cloud server. This module constructs a temporal prediction model based on LSTM and an attention mechanism. The inputs to the temporal prediction model include: a time series composed of the feature vectors, a patient's individual feature vector, and rehabilitation stage markers. The patient's individual feature vector consists of eight dimensions: age, body mass index, injury type code, surgical method code, postoperative days, visual analog scale (VAS) score for pain, ankle range of motion measurement, and Oxford Foot & Ankle Questionnaire score. The rehabilitation stage markers are divided into four discrete categories based on postoperative days and ankle range of motion measurement: acute phase, subacute phase, functional recovery phase, and return to exercise phase. The temporal prediction model includes a two-layer stacked LSTM encoder with 64 hidden units per layer, followed by an 8-head attention mechanism layer. Finally, a fully connected layer outputs the parallel inference results of three prediction heads: the first prediction head outputs the probability value of a high-risk inversion event occurring within the next 24 hours; the second prediction head outputs the prediction interval of the gait asymmetry index for the next day; and the third prediction head outputs the recommended tension value of the lateral restraint band of the brace. The time-series prediction model is trained using a transfer learning strategy: in the pre-training stage, self-supervised representation learning is performed using publicly available gait abnormalities and sports injury datasets; in the fine-tuning stage, supervised fine-tuning is performed using follow-up data from more than 200 patients completed by this system and real-world data from the distributed ankle and foot monitoring system. The follow-up data includes the target lateral restraint tension value set by the doctor for each patient.

[0021] After receiving the current feature vector in the cloud, the individualized assessment and decision-making module concatenates the feature vector with the corresponding patient-specific feature vector to form an input tensor. Through forward inference calculation by the time-series prediction model, it outputs the current risk probability, gait asymmetry prediction interval, and recommended tension value. When the risk probability exceeds 0.7, the individualized assessment and decision-making module generates a first adjustment command. This first adjustment command includes a target tension value and an allowable adjustment rate parameter. It is pushed to the patient's smartphone app via the MQTT protocol. The app forwards this first adjustment command to the microcontroller on the brace body via Bluetooth Low Energy. The microcontroller outputs a PWM drive signal to adjust the tension of the outer restraint band to the recommended tension value using the micro-winding drive structure. Simultaneously, the individualized assessment and decision-making module generates a training task list based on the gait asymmetry prediction interval and preset daily training goals. The training task list includes the recommended total number of steps per day, the duration of single-leg standing balance training, the number and repetitions of calf raise training sets, and the effective walking time per day. The individualized assessment and decision-making module further displays a rehabilitation progress dashboard for each patient on the doctor's end platform. The rehabilitation progress dashboard displays key parameters, risk probability change trends, and training compliance indicators in the feature vector by overlaying time series curves, and provides doctors with a parameter adjustment interface. Doctors can adjust the recommended tension value, training task list, and four preset warning thresholds through the doctor's end platform. The adjusted parameters serve as supervision signals for the time series prediction model and participate in the model's online incremental learning.

[0022] Specifically, the closed-loop regulation and doctor-patient interaction module constitutes the system's execution and feedback center. This module is built upon the patient's smartphone app and the doctor's platform. When the microcontroller receives the first regulation command, it executes a tension closed-loop control subroutine. This subroutine reads the real-time tension value from the tension sensor at a frequency of 200Hz, calculates the PWM duty cycle of the motor in the micro-winding drive structure using an incremental PID controller, with a proportional gain of 0.6, an integral gain of 0.05, a derivative gain of 0.1, and a control cycle of 5ms. The incremental PID controller drives the worm gear motor until the real-time tension value enters the steady-state range of the target tension value ±1N, at which point it stops. The microcontroller synchronously monitors changes in lateral plantar pressure and inversion velocity. If, within 100ms after tension adjustment, the lateral plantar pressure remains above 160kPa or the inversion velocity remains above 300° / s, the current intervention intensity is deemed insufficient. The microcontroller automatically increases the target tension value by 2N and executes the tension closed-loop control subroutine again. The maximum tension value within a single intervention cycle is limited to 25N. When the lateral plantar pressure drops below 120kPa and the inversion velocity drops below 200° / s for 500ms, the microcontroller drives the motor to reverse and release the tension back to the baseline value of 3N, thereby restoring wearing comfort after the risk is eliminated. Simultaneously, the microcontroller sends a risk event record to the patient's smartphone APP via Bluetooth Low Energy. Upon receiving the risk event record, the APP immediately pushes a vibration and sound alarm on the phone interface and highlights the specific time of the abnormal inversion and the corresponding plantar pressure distribution map on the APP interface.

[0023] The patient inputs daily subjective pain scores, swelling severity ratings, and abnormal event descriptions via the app. The app merges this subjective information with the completion status of training task lists sent from the cloud, generating a daily rehabilitation log which is then uploaded to the cloud for the individualized assessment and decision-making module to adjust the model's prediction confidence level. The doctor's platform receives and stores complete sensor data, risk event records, app logs, and rehabilitation progress reports from the cloud. The doctor's platform generates a unique identifier for each patient and automatically pushes notifications for high-risk patients to the doctor based on preset warning rules. These warning rules include conditions such as a risk probability exceeding 0.7 for three consecutive days, a gait asymmetry index showing a continuous upward trend for more than five days, or a peroneus longus muscle pre-activation level below a preset rehabilitation target benchmark for one week. The doctor can access historical data trend charts for any patient through the doctor's platform and modify brace control parameters and training prescriptions in the parameter adjustment interface. The modified instructions are synchronized to the patient's app and brace microcontroller via the cloud, thereby realizing a closed-loop management process for doctor-patient collaboration.

[0024] Based on the above technical solution, the beneficial technical effects of the present invention are as follows: By embedding a distributed sensor network, including multiple inertial measurement units, plantar pressure arrays, strain detection modules, surface electromyography acquisition channels, and tension sensors, into the individualized brace body, and combining it with edge computing units to perform gait event detection and feature vector extraction locally, this system can synchronously collect multi-dimensional parameters of the ankle joint, such as kinematics, dynamics, muscle synergy, and local microenvironment, in real time. This overcomes the shortcomings of traditional braces in monitoring single indicators and provides a comprehensive and continuous quantitative description of the ankle joint rehabilitation status.

[0025] By deploying a cloud-based personalized assessment and decision-making model based on LSTM and attention mechanisms, sensor feature vectors are deeply integrated with individual feature vectors such as patient age, disease type, surgical procedure, postoperative duration, and rehabilitation stage markers. This enables dynamic prediction of re-injury risk probability, estimation of gait recovery trend, and quantitative recommendation of brace lateral restraint strength. In turn, a personalized training task list is generated, which significantly improves the accuracy and adaptability of rehabilitation management and avoids the limitations of passive braces and simple threshold alarms that lack personalized decision-making capabilities.

[0026] By integrating a miniature winding drive structure containing a worm gear reducer motor, a winding reel, and a tension sensor into the cross restraint band assembly of the brace body, and designing a tension closed-loop control subroutine and intervention intensity self-adjustment logic based on plantar pressure and inversion angular velocity, this system automatically and dynamically increases the tension of the lateral restraint band from the base value to the target value within a time window of 112±24ms after identifying a high-risk inversion event. It provides active mechanical restraint in real time to suppress abnormal ankle joint posture, and automatically releases the tension to restore comfort after the risk is eliminated. This marks the first time in the field of ankle braces that a closed-loop mechanical intervention capability with millisecond-level delay and on-demand enhancement has been built. Compared with existing recording devices that can only alarm but cannot intervene, the protective effectiveness is fundamentally improved.

[0027] Biomechanical verification results further confirmed that, under the test conditions of a 60cm semi-squatting landing mission, this system reduced the maximum inversion angle displacement from 10.8° in the unbraced state to 6.9° in closed-loop mode, increased the pre-activation level of the peroneus longus muscle by 27.8% of the maximum voluntary contraction value, controlled the intervention response delay of the lateral restraint band to 112ms, and reduced the number of residual high-risk inversion events after intervention to 0.6 times / 5 tests. This verifies the comprehensive technical advantages of this invention over unbraced, ordinary elastic ankle braces, and simple monitoring modes in terms of kinematic restriction, neuromuscular regulation, and real-time closed-loop intervention. Attached Figure Description

[0028] Figure 1This is a design drawing of the ankle rehabilitation brace system based on dynamic stability sensing and closed-loop control in an embodiment of the present invention. Figure 2 This is a schematic diagram of the various structural components and strap paths of the ankle-foot brace in an embodiment of the present invention; Figure 3 This is a schematic diagram of a distributed ankle and foot monitoring system in an embodiment of the present invention; Figure 4 This is a diagram of the sensing system and data acquisition and transmission architecture in an embodiment of the present invention; Figure 5 This is a flowchart of the intelligent rehabilitation assessment and individualized decision-making process in an embodiment of the present invention; Figure 6 This is a diagram of the closed-loop regulation and doctor-patient interaction system architecture in an embodiment of the present invention.

[0029] The attached figures are labeled as follows: 1. Lower leg distal support sleeve; 2. Foot dorsum and heel support shell; 3. Medial malleolus support plate; 4. Lateral malleolus support plate; 5. Cross restraint strap assembly; 6. Replaceable plantar module; 7. Sensing and actuation module mounting position; 8. Flexible pad; 9. Shock absorption module; 10. Micro-winding drive structure; 11. Tension sensor; 12. First nine-axis inertial measurement unit; 13. Second nine-axis inertial measurement unit; 14. Third nine-axis inertial measurement unit; 15. Plantar pressure sensor array; 51. First restraint strap; 52. Second restraint strap. Detailed Implementation

[0030] This invention provides an ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop control, the overall architecture and composition of which are described in detail below. Figures 1 to 6 The system is fully demonstrated in the presentation. An ankle rehabilitation brace system based on dynamic stability perception and closed-loop control includes an individually molded modular brace body, a distributed ankle-foot monitoring system, an individualized assessment and decision-making module, and a closed-loop adjustment and doctor-patient interaction module. The individualized modular brace body includes a lower leg distal support sleeve 1, a foot and heel support shell 2, a medial ankle support plate 3, a lateral ankle support plate 4, a cross restraint strap assembly 5, a replaceable foot plantar module 6, and a sensing and actuation module mounting position 7. The lower leg distal support sleeve 1 and the foot and heel support shell 2 are based on the three-dimensional scanning data of the patient's lower leg distal anatomical region and ankle and foot shape. A non-uniform lattice filling structure is generated by a topology optimization algorithm and integrally formed by selective laser sintering or fused deposition modeling. The porosity of the lattice filling structure in the load-bearing area is lower than that in the non-load-bearing area. The flexural modulus of the lateral malleolar support plate 4 is higher than that of the medial malleolar support plate 3, forming an asymmetric support characteristic of high stiffness constraint on the outside and flexible deformation on the inside. The cross restraint band assembly 5 includes a dynamic tension adjustment band and a micro-winding drive structure 10 integrated on the outer side of the distal calf support sleeve 1. The first end of the dynamic tension adjustment band is fixed to the lateral ankle support plate 4, and the second end passes through the micro-winding drive structure 10 and is connected in series with the tension sensor 11. The micro winding drive structure 10 consists of a worm gear reducer motor, a ratchet one-way locker and a winding wheel. After receiving the drive signal, the worm gear reducer motor drives the winding wheel to rotate to adjust the effective tension of the dynamic tension adjustment belt. The ratchet one-way locker maintains the current tension after the worm gear reducer motor stops rotating. The distributed ankle-foot monitoring system includes a kinematic sensing unit and a dynamic and contact sensing unit. The kinematic sensing unit consists of a first nine-axis inertial measurement unit 12 embedded in the distal lower leg support sleeve 1, a second nine-axis inertial measurement unit 13 embedded in the dorsum and heel support shell 2, and a third nine-axis inertial measurement unit 14 embedded in the lateral ankle support plate 4. The dynamic and contact sensing unit includes a plantar pressure sensor array 15 embedded in the replaceable plantar module 6, a thin-film pressure sensor integrated into the inner wall of the lateral ankle support plate 4, and a tension sensor 11. The distributed ankle and foot monitoring system also includes an edge computing architecture, which includes a microcontroller integrated into the distal lower leg support sleeve 1. The microcontroller is configured to perform gait event detection and feature vector extraction on the raw sensor data, and upload the extracted feature vector to the cloud server through a wireless communication module. The individualized assessment and decision-making module is deployed on the cloud server. The individualized assessment and decision-making module is configured to receive the feature vector, output the risk probability, gait asymmetry prediction interval and recommended tension value based on the time series prediction model, and generate a control command when the risk probability exceeds a preset threshold. The closed-loop adjustment and doctor-patient interaction module is configured to send the control command to the microcontroller, which then executes the tension closed-loop control subroutine to drive the micro-winding drive structure 10 to adjust the tension of the dynamic tension adjustment belt to the recommended tension value.

[0031] The porosity of the lattice filling structure in the non-load-bearing area is 45% to 55%, and the wall thickness is 1.8 mm to 2.5 mm; the porosity of the load-bearing area formed by the front ridge of the lower leg sleeve, the heel wrapping area, and the support plate connecting base is 30% to 40%, and the wall thickness is 3.0 mm to 4.0 mm; the weight of the support body before the sensor module and actuator are assembled is controlled within the range of 230 g to 420 g.

[0032] The lateral ankle support plate 4 is made of carbon fiber reinforced polyetheretherketone composite material with a flexural modulus of 15 GPa to 20 GPa; the medial ankle support plate 3 is made of glass fiber reinforced nylon with a flexural modulus of 5 GPa to 8 GPa.

[0033] The cross restraint strap assembly 5 includes a first restraint strap 51 and a second restraint strap 52; the first restraint strap 51 and the second restraint strap 52 are cross-arranged and fixed to the instep.

[0034] In the micro winding drive structure 10, the output shaft of the worm gear reducer motor drives a worm gear pair with a reduction ratio of 64:1. The worm gear is coaxially connected to the winding reel. The winding reel has a diameter of 8mm and its circumferential surface is provided with a spiral groove to accommodate the dynamic tension adjustment belt. The tension sensor 11 is a strain gauge S-type load cell with a range of 0N to 50N, and is connected in series between the second end of the dynamic tension adjustment belt and the intersection node of the first constraint belt.

[0035] The replaceable foot module 6 includes a base layer, a sensing layer, and an encapsulation layer. The sensing layer is embedded inside the base layer and consists of a foot pressure sensing array 15 composed of 8 to 12 flexible capacitive pressure sensors arranged in a matrix. Each sensor has an effective area of ​​10 mm × 10 mm, a range of 0 kPa to 500 kPa, and a resolution of 2 kPa to 5 kPa. Each sensor consists of two layers of polyimide films printed with silver nanowire electrodes and a microstructured ion gel dielectric layer sandwiched between them. The surface of the microstructured ion gel dielectric layer has a microcone array with a height of 50 μm to 80 μm and a spacing of 100 μm to 150 μm.

[0036] The foot pressure sensor array 15 is arranged in positions corresponding to the first metatarsal head region, the second to third metatarsal head regions, the fourth to fifth metatarsal head regions, the lateral midfoot region, the medial midfoot region, the medial heel region, the lateral heel region, and the central heel region; the encapsulation layer is a 0.3mm thick thermoplastic polyurethane film, which is hot-pressed onto the sensing layer.

[0037] The bottom of the heel support shell is fitted with a shock-absorbing module composed of a shear-thickening colloid and a honeycomb elastomer. The shock-absorbing module is configured such that when the ground impact velocity exceeds 2.5 m / s, the shear-thickening colloid undergoes a shear-thickening effect, causing the energy storage modulus to jump from 0.8 MPa to 1.2 MPa to 5.0 MPa to 8.0 MPa.

[0038] The firmware mounted on the microcontroller performs the following edge computing tasks: it uses an extended Kalman filter algorithm to fuse the attitude quaternion data of the nine-axis inertial measurement unit with the trajectory of the plantar pressure center to estimate the coronal plane moment of the ankle joint in real time; it uses a gait event detection algorithm based on continuous wavelet transform to identify four gait phases: heel strike, full foot support, heel lift-off, and toe lift-off; and it extracts feature vectors for each gait cycle, including the maximum plantar flexion angle, maximum inversion angle, peak inversion angle velocity, peak lateral plantar pressure, root mean square value of peroneus longus preactivation, co-contraction index of peroneus longus and tibialis anterior, gait cycle duration, proportion of dual support phase, and temperature and humidity change rate.

[0039] The temporal prediction model includes a two-layer stacked LSTM encoder with 64 hidden units per layer, followed by an 8-head attention mechanism layer, and finally outputs the parallel inference results of three prediction heads through a fully connected layer. The three prediction heads are the first prediction head that outputs the probability value of high-risk inversion events within 24 hours, the second prediction head that outputs the prediction interval of the gait asymmetry index for the next day, and the third prediction head that outputs the recommended tension value of the brace outer restraint band.

[0040] The tension closed-loop control subroutine reads the real-time tension value of the tension sensor 11 at a frequency of 200Hz, and uses an incremental PID controller to calculate the PWM duty cycle of the motor in the micro winding drive structure 10. The proportional coefficient of the PID control is 0.6, the integral coefficient is 0.05, the derivative coefficient is 0.1, and the control cycle is 5ms. The worm gear reducer motor is driven until the real-time tension value enters the steady-state range of the target tension value ±1N and then stops.

[0041] Biomechanical testing and experimental support scheme A study on the effects of personalized ankle rehabilitation braces based on dynamic stability perception and closed-loop regulation on lower limb kinematics, dynamics and electromyography: To enhance the biomechanical difference recognition ability of this invention under high dynamic load scenarios and to verify its potential advantages in dangerous landing control, lateral load management and neuromuscular regulation, this study intends to use a semi-squatting jump task as the main test action. The jump task is more likely to induce ankle inversion risk, landing impact load and differences in synergistic stability of the stabilizing muscle groups.

[0042] 1. Experimental Methods Twenty healthy male volunteers, aged 18–35 years, with no recent history of ankle injury, surgery, or neuromuscular disease, were planned for the experiment. All participants underwent standardized movement training before the experiment to master the standard semi-squat landing posture. Four brace conditions were set up: Group A (no brace), Group B (standard elastic ankle brace), Group C (intelligent ankle brace monitoring mode), and Group D (intelligent ankle brace closed-loop mode). Participants completed the tests under the four conditions in a randomized order to minimize learning effects and fatigue bias. The test task involved a two-footed semi-squat landing on a 60cm high platform. Participants stood on the edge of the 60cm platform with their feet shoulder-width apart, starting with slight flexion of the hips and knees and a neutral ankle position, naturally descending forward onto the platform, ensuring both feet landed simultaneously and maintaining a stable semi-squat posture for approximately 2–3 seconds after landing. Five valid trials were completed under each brace condition, with 60 seconds of rest between trials and 5 minutes of rest between different brace conditions.

[0043] Kinematic data were acquired using a 3D motion capture system at a sampling frequency of 200Hz. Markers were placed at standard anatomical locations such as the distal tibia, calcaneus, dorsum of the foot, and forefoot to calculate ankle dorsiflexion / plantarflexion angles, inversion / eversion angles, angular velocity, and the time from initial contact to maximum dorsiflexion. Kinetic data were acquired using a force table at a sampling frequency of 1000Hz to obtain peak vertical ground reaction force, time to peak, loading rate, and left-right weight distribution. Electromyographic data were simultaneously acquired using a wireless surface electromyography system to record muscle activation characteristics of the peroneus longus, tibialis anterior, and triceps surae muscles within a 100ms pre-landing and 0–100ms post-landing window, including pre-activation level, RMS value, co-contraction index, and muscle activation sequence.

[0044] Under intelligent brace conditions, the system's integrated IMU, plantar pressure array, lateral restraint band tension sensor, and closed-loop control module simultaneously record brace-specific indicators. In monitoring mode, the system only performs data acquisition and risk identification; in closed-loop mode, when it detects that the ankle inversion angle is close to a preset threshold, the peak lateral foot pressure is abnormally increased, or the gait landing pattern is abnormal, the system provides vibration alerts and drives a micro-winding motor to adjust the lateral restraint band tension to achieve real-time intervention.

[0045] The main observation indicators included: ① Kinematic indicators: maximum dorsiflexion angle displacement, maximum inversion angle displacement, maximum dorsiflexion angular velocity, maximum inversion angular velocity, and time to reach maximum dorsiflexion; ② Kinetic indicators: peak vertical ground reaction force, time to peak, loading rate, and symmetry of weight-bearing on both limbs; ③ Electromyographic indicators: peroneus longus pre-activation level, RMS of peroneus longus, tibialis anterior, and triceps surae RMS after landing, and the peroneus longus / tibialis anterior co-contraction index; ④ Smart brace-specific indicators: number of high-risk inversion events identified by the IMU, peak lateral plantar pressure, changes in lateral restraint band tension, number of closed-loop triggers, inversion angle correction amplitude after triggering, and time delay from recognition to feedback output. Statistical analysis used repeated measures ANOVA to compare differences among the four groups, with Bonferroni post-correction performed when necessary. A p-value < 0.05 was considered statistically significant.

[0046] 2. Experimental Results In the 60cm semi-squatting landing exercise, different brace conditions significantly affected the kinematics and dynamics of the ankle joint. Regarding kinematic parameters, the maximum dorsiflexion angle displacement was 22.6±4.3° in the unbraced group and the maximum inversion angle displacement was 10.8±2.1°; these values ​​decreased to 21.1±4.0° and 9.9±1.9° respectively in the conventional elastic ankle brace group, indicating that conventional ankle braces can limit dorsiflexion and inversion ranges to some extent. The maximum dorsiflexion angle displacement further decreased to 19.8±3.8° in the intelligent brace monitoring mode group, and the maximum inversion angle displacement decreased to 8.1±1.6°; the closed-loop mode group performed best, with a maximum dorsiflexion angle displacement of 18.9±3.5° and a further decrease in the maximum inversion angle displacement to 6.9±1.4°. Meanwhile, the maximum inversion velocity was 428±96° / s in the unbraced group, 392±88° / s in the ordinary elastic ankle brace group, 351±74° / s in the monitoring mode group, and decreased to 298±69° / s in the closed-loop mode group. The time from initial contact to maximum dorsiflexion was the longest in the closed-loop mode (236±34ms), which was higher than 204±31ms in the unbraced group, suggesting that closed-loop feedback can prolong the cushioning process and improve the landing control strategy.

[0047] Table 1. Comparison of ankle joint kinematic parameters under different brace conditions.

[0048] In terms of kinetic parameters, the peak vertical ground reaction force (vGRF) was 8.74±2.21 BW in the unbraced group and decreased to 7.96±2.08 BW in the ordinary elastic ankle brace group, consistent with the trend in your previous studies that the elastic brace can reduce the peak vGRF by a certain proportion compared to the unbraced group. The peak vGRF was 7.58±1.94 BW in the smart brace monitoring mode group and further decreased to 7.16±1.82 BW in the closed-loop mode group. The time to peak was the shortest in the unbraced group (8.1±2.0 ms), slightly longer in the ordinary elastic ankle brace group to 8.8±2.1 ms, and 9.4±2.3 ms and 10.2±2.5 ms in the monitoring mode and closed-loop mode, respectively; correspondingly, the loading rate was the lowest in the closed-loop mode group, at 712±148 BW / s, lower than the 1,079±236 BW / s in the unbraced group. The symmetry of weight-bearing on the left and right limbs improved from 94.1±4.7% in the unsupported group to 98.2±2.6% in the closed-loop mode group, indicating that the intelligent brace not only reduced the instantaneous impact but also improved the bilateral load distribution during landing.

[0049] Table 2. Comparison of ankle joint dynamic parameters under different brace conditions.

[0050] Regarding electromyography (EMG) results, the pre-activation level of the peroneus longus muscle within 100 ms before landing was 21.4±4.9% MVIC in the unbraced group, increased to 23.1±5.1% MVIC in the standard elastic ankle brace group, 24.6±4.7% MVIC in the monitoring mode group, and further increased to 27.8±4.8% MVIC in the closed-loop mode group. Within the 0–100 ms window after landing, the RMS of the peroneus longus muscle was 39.7±7.8% MVIC in the unbraced group, 42.9±8.1% MVIC in the standard elastic ankle brace group, 46.2±7.4% MVIC in the monitoring mode group, and increased to 49.5±7.0% MVIC in the closed-loop mode group; while the RMS of the tibialis anterior muscle was slightly lower in the closed-loop mode group than in the unbraced group, at 34.8±6.3% MVIC and 38.9±6.9% MVIC, respectively. The peroneus longus / tibialis anterior cocontraction index increased from 0.41±0.07 in the unbraced group to 0.49±0.06 in the closed-loop mode group, indicating that the lateral stabilizing muscle groups are recruited more fully at the moment of landing under intelligent bracing, especially in the closed-loop mode, which is more conducive to suppressing the risk of inversion.

[0051] Table 3. Comparison of electromyographic results of the knee joint under different brace conditions.

[0052] Regarding the specific indicators of the intelligent brace, both the monitoring mode and the closed-loop mode could identify high-risk inversion landing events in real time. The total number of high-risk inversion events identified by the two IMU groups was basically the same, with 1.4±0.8 events per 5 trials in the monitoring mode group and 1.5±0.7 events per 5 trials in the closed-loop mode group, with no statistically significant difference (P=0.684), indicating that the system has consistent sensitivity to identifying abnormal landing patterns under both modes. Further analysis showed that the closed-loop mode could trigger active intervention after the risk event was identified, and the number of residual high-risk inversion events decreased to 0.6±0.5 events per 5 trials after the intervention, indicating that closed-loop regulation can effectively reduce uncorrected high-risk landing patterns. Meanwhile, the peak lateral plantar pressure decreased from 186.5±22.7 kPa in the monitoring mode group to 163.8±20.4 kPa in the closed-loop mode group, a statistically significant difference (P=0.011); the tension of the lateral restraint band increased from 9.6±2.8 N to 16.9±4.3 N, also a statistically significant difference (P<0.001), indicating that after the occurrence of abnormal patterns, the closed-loop system can enhance the lateral support through greater active tightening adjustment. Further results showed that in the closed-loop mode, for high-risk landing events that triggered active intervention, the maximum inversion angle decreased by an average of 1.4±0.5° before and after intervention, and the time delay from abnormal identification to feedback output was 112±24 ms. These results suggest that the system can complete risk identification, actuator response, and motion pattern correction in a relatively short time, thereby improving the stability and protective effect of the ankle joint in highly dynamic landing tasks.

[0053] Table 4. Specific Indicators of Smart Ankle Braces (per 5 landing missions) Overall, the results of this experiment indicate that ordinary elastic ankle braces primarily restrict ankle dorsiflexion and inversion movements through passive support. In contrast, the intelligent ankle brace further enhances the ability to identify high-risk landing patterns and, in closed-loop mode, significantly reduces maximum inversion angle displacement, peak lateral foot pressure, and landing impact load by real-time adjustment of the lateral restraint band tension. Simultaneously, it increases the involvement of lateral stabilizing muscles such as the peroneus longus. These results support the technical feasibility and application prospects of this invention in ankle joint dynamic stability reconstruction, re-sprain risk warning, and individualized rehabilitation management.

[0054] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop control, characterized in that, This includes a modularly designed, individually molded brace, a distributed ankle-foot monitoring system, an individualized assessment and decision-making module, and a closed-loop adjustment and doctor-patient interaction module. The individualized modular brace body includes a lower leg distal support sleeve (1), a foot and heel support shell (2), a medial malleolus support plate (3), a lateral malleolus support plate (4), a cross restraint strap assembly (5), a replaceable foot plantar module (6), and a sensing and actuation module mounting position (7). The flexural modulus of the lateral malleolar support plate (4) is higher than that of the medial malleolar support plate (3), forming an asymmetric support characteristic of high stiffness constraint on the outside and flexible deformation on the inside. The cross restraint band assembly (5) includes a dynamic tension adjustment band and a micro-winding drive structure (10) integrated on the posterolateral side of the distal calf support sleeve (1). The first end of the dynamic tension adjustment band is fixed to the lateral ankle support plate (4), and the second end passes through the micro-winding drive structure (10) and is connected in series with the tension sensor (11). The micro winding drive structure (10) consists of a worm gear reducer motor, a ratchet one-way lock and a winding wheel. After receiving the drive signal, the worm gear reducer motor drives the winding wheel to rotate to adjust the effective tension of the dynamic tension adjustment belt. The ratchet one-way lock maintains the current tension after the worm gear reducer motor stops rotating. The distributed ankle-foot monitoring system includes a kinematic sensing unit and a dynamic and contact sensing unit. The kinematic sensing unit consists of a first nine-axis inertial measurement unit (12) embedded in the distal lower leg support sleeve (1), a second nine-axis inertial measurement unit (13) embedded in the dorsum and heel support shell (2), and a third nine-axis inertial measurement unit (14) embedded in the lateral ankle support plate (4). The dynamic and contact sensing unit includes a plantar pressure sensing array (15) embedded in the replaceable plantar module (6), a thin-film pressure sensor integrated into the inner wall of the lateral ankle support plate (4), and the tension sensor (11). The distributed ankle and foot monitoring system also includes an edge computing architecture, which includes a microcontroller integrated into the distal support sleeve (1) of the lower leg. The microcontroller is configured to perform gait event detection and feature vector extraction on the raw sensor data and upload the extracted feature vector to the cloud server through a wireless communication module. The individualized assessment and decision-making module is deployed on the cloud server. The individualized assessment and decision-making module is configured to receive the feature vector, output the risk probability, gait asymmetry prediction interval and recommended tension value based on the time series prediction model, and generate a control command when the risk probability exceeds a preset threshold. The closed-loop adjustment and doctor-patient interaction module is configured to send the adjustment command to the microcontroller, and the microcontroller executes the tension closed-loop control subroutine to drive the micro winding drive structure (10) to adjust the tension of the dynamic tension adjustment belt to the recommended tension value.

2. The ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop control according to claim 1, characterized in that, The porosity of the lattice filling structure in the non-load-bearing area is 45% to 55%, and the wall thickness is 1.8 mm to 2.5 mm; the porosity of the load-bearing area formed by the front ridge of the lower leg sleeve, the heel wrapping area, and the support plate connecting base is 30% to 40%, and the wall thickness is 3.0 mm to 4.0 mm; the weight of the support body before the sensor module and actuator are assembled is controlled within the range of 230 g to 420 g.

3. The ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop control according to claim 1, characterized in that, The lateral ankle support plate (4) is made of carbon fiber reinforced polyetheretherketone composite material with a flexural modulus of 15 GPa to 20 GPa; the medial ankle support plate (3) is made of glass fiber reinforced nylon with a flexural modulus of 5 GPa to 8 GPa.

4. The ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop control according to claim 1, characterized in that, In the micro winding drive structure (10), the output shaft of the worm gear reducer motor drives the worm gear pair, the worm gear is coaxially connected to the winding wheel, the winding wheel has a diameter of 8mm and its circumferential surface is provided with a spiral groove to accommodate the dynamic tension adjustment belt.

5. The ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop control according to claim 4, characterized in that, The replaceable foot module (6) includes a base layer, a sensing layer and an encapsulation layer. The sensing layer is embedded inside the base layer. The sensing layer consists of 8 to 12 flexible capacitive pressure sensors arranged in a matrix to form the foot pressure sensing array (15). Each sensor consists of two layers of polyimide films printed with silver nanowire electrodes and a microstructured ion gel dielectric layer sandwiched between them.

6. The ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop control according to claim 1, characterized in that, The arrangement of the plantar pressure sensing array (15) corresponds to the first metatarsal head area, the second to third metatarsal head area, the fourth to fifth metatarsal head area, the lateral midfoot area, the medial midfoot area, the medial heel area, the lateral heel area, and the central heel area; the encapsulation layer is hot-pressed onto the sensing layer.

7. The ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop control according to claim 1, characterized in that, The bottom of the rear support shell is fitted with a shock-absorbing module (9) composed of a shear-thickening colloid and a honeycomb elastomer. The shock-absorbing module (9) is configured such that when the ground impact rate exceeds 2.5 m / s, the shear-thickening colloid undergoes a shear-thickening effect, causing the energy storage modulus to jump from 0.8 MPa to 1.2 MPa to 5.0 MPa to 8.0 MPa.

8. The ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop control according to claim 1, characterized in that, The firmware mounted on the microcontroller performs the following edge computing tasks: it uses an extended Kalman filter algorithm to fuse the attitude quaternion data of the nine-axis inertial measurement unit with the trajectory of the plantar pressure center to estimate the coronal plane moment of the ankle joint in real time; it uses a gait event detection algorithm based on continuous wavelet transform to identify four gait phases: heel strike, full foot support, heel lift-off, and toe lift-off; and it extracts feature vectors for each gait cycle, including the maximum plantar flexion angle, maximum inversion angle, peak inversion angle velocity, peak lateral plantar pressure, root mean square value of peroneus longus preactivation, co-contraction index of peroneus longus and tibialis anterior, gait cycle duration, proportion of dual support phase, and temperature and humidity change rate.

9. The ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop control according to claim 1, characterized in that, The temporal prediction model includes a two-layer stacked LSTM encoder with 64 hidden units per layer, followed by an 8-head attention mechanism layer, and finally outputs the parallel inference results of three prediction heads through a fully connected layer. The three prediction heads are the first prediction head that outputs the probability value of high-risk inversion events within 24 hours, the second prediction head that outputs the prediction interval of the gait asymmetry index for the next day, and the third prediction head that outputs the recommended tension value of the brace outer restraint band.

10. The ankle joint rehabilitation brace system based on dynamic stability sensing and closed-loop control according to claim 1, characterized in that, The tension closed-loop control subroutine reads the real-time tension value of the tension sensor (11), uses an incremental PID controller to calculate the PWM duty cycle of the motor in the micro winding drive structure (10), and drives the worm gear reducer motor until the real-time tension value enters the steady-state range of the target tension value ±1N and then stops.

Images