A closed-loop control method and system for postoperative dynamic balance training of ankle ligament

Abstract

The application discloses a closed-loop control method and system for postoperative dynamic balance training of ankle ligament, and relates to the technical field of rehabilitation training control, and has the technical scheme as follows: obtaining a postoperative week parameter and calling a corresponding stage parameter set; obtaining flip angle time sequence data and electromyographic signals; calculating a first evaluation index according to the flip angle time sequence data and detecting a second evaluation index according to the electromyographic signals; calculating an integral value accumulated with time for a degree that the flip angle exceeds a safety constraint threshold; generating a down-order when the integral value reaches a preset condition or the first evaluation index or the second evaluation index exceeds an allowed range, and generating an up-order when the down-order is not triggered for continuous multiple rounds; taking a current stage parameter, statistical values of the first evaluation index and the second evaluation index and an integral value statistical total as a state description vector, inputting a prediction model to output a next training cycle stage parameter set; and delivering the order to a training terminal for execution.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation training control technology, and more specifically, to a closed-loop control method and system for dynamic balance training after ankle ligament surgery. Background Technology

[0002] Lateral collateral ligament injury of the ankle joint is one of the most common musculoskeletal injuries in clinical practice. After ligament repair or reconstruction surgery, patients need to perform dynamic balance training on unstable support surfaces to rebuild neuromuscular control. Currently, commonly used dynamic balance training methods include Bosu ball or balance board training, large-scale balance assessment training systems, and wearable ankle exoskeletons. Bosu balls or balance boards force patients to activate the muscles around the ankle joint for balance adjustment by providing an unstable support surface. However, they cannot limit the inversion angle of the ankle joint during training. When the patient's fatigue or inattention leads to an excessive inversion angle, it is easy to exceed the safe range allowed for ligament healing, causing excessive tension on the reconstructed ligament or even re-rupture. Large-scale balance assessment training systems monitor the patient's center of gravity position in real time through sensors and adjust the platform posture accordingly. However, the equipment is bulky and expensive, and is limited to hospital use, making it impossible for patients to perform daily training in a home environment. Wearable ankle exoskeletons mechanically limit the range of motion of the ankle joint through motors or dampers. However, the equipment is heavy and expensive, and excessive mechanical restraint can inhibit natural proprioceptive input, which is not conducive to the reconstruction of proprioceptive function.

[0003] In the practical application of the aforementioned training methods, a prominent issue is the disconnect between safety constraints and the patient's postoperative recovery process. Existing unstable platforms and training systems typically employ fixed safety limits or uniform training parameters, failing to automatically adjust the allowable range of the inversion safety angle according to the patient's postoperative weeks. In the early postoperative period, the patient's peroneal muscle reaction time is significantly prolonged, and the inversion angle easily exceeds the range allowed by ligament healing when subjected to unexpected disturbances, yet the safety constraints are not tightened accordingly. In the later postoperative period, patients require a wider range of inversion angles to stimulate further proprioceptive recovery, but the safety constraints are not relaxed accordingly. In this situation, fixed safety constraints cannot simultaneously address the differentiated needs for training safety and training stimulation intensity at different recovery stages. Furthermore, existing assessment methods struggle to comprehensively reflect the true level of proprioceptive recovery. Commonly used clinical assessment indicators mainly include center of gravity trajectory amplitude and single-leg standing time, which are easily affected by compensatory movements of the lumbar and hip joints. When patients maintain balance through compensatory movements of the trunk and hip joints, the center of gravity trajectory amplitude and standing time may appear normal, but in reality, proprioceptive function around the ankle joint has not been effectively restored. In such cases, the assessment results cannot accurately reflect the patient's rehabilitation progress. In addition, there is no direct correlation between the assessment results and the parameter settings for the next training session. The increase or decrease in training difficulty depends entirely on the subjective experience and judgment of the therapist. Different therapists have different judgment criteria, and the same therapist may have inconsistent judgments at different times, making it difficult to guarantee the standardization and repeatability of the training program.

[0004] Therefore, researching and designing a closed-loop control method and system for dynamic balance training after ankle ligament surgery that can overcome the above-mentioned defects is an urgent problem to be solved. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a closed-loop control method and system for dynamic balance training after ankle ligament surgery. This system automatically adjusts the safety constraint threshold based on the number of weeks post-surgery, achieves multi-dimensional safety monitoring through joint decision-making using integral values ​​and evaluation indicators, and drives personalized evolution of prescription parameters based on historical data. This improves the efficiency and accuracy of proprioceptive reconstruction while ensuring training safety.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution: Firstly, a closed-loop control method for dynamic balance training after ankle ligament surgery is provided, including the following steps: Obtain the postoperative week number parameter of the trainee, and call the stage parameter set corresponding to the postoperative week number parameter. The stage parameter set contains at least one safety constraint threshold that changes with the postoperative week number. Acquire the time-series data of the flip angle of the trainee during the training process and the electromyographic signals of at least one muscle group; The degree of deviation of the flip angle relative to the target angle is calculated based on the flip angle time series data as the first evaluation index, and the activation response speed of the muscle group after the flip perturbation event is detected based on the electromyographic signal as the second evaluation index. The flip angle time series data is compared with the safety constraint threshold, and the integral value of the degree to which the flip angle exceeds the safety constraint threshold is accumulated over time is calculated. When the integral value reaches a preset condition, or when the first evaluation index or the second evaluation index exceeds the corresponding allowable range, a downgrade instruction is generated; when no downgrade is triggered in multiple consecutive training rounds, an upgrade instruction is generated. A state description vector is constructed using the current stage parameters, the statistical values ​​of the first and second evaluation indicators in the historical multi-round training, and the total statistical value of the integral values ​​in the historical multi-round training. The state description vector is input into the pre-trained prediction model, and the stage parameter set for the next training cycle is output. The generated instructions are sent to the training terminal for execution via the communication interface.

[0007] Furthermore, the safety constraint threshold includes an inversion safety angle threshold, which increases with the number of weeks after surgery, and is used to limit the maximum allowable value of the inversion angle during training. The stage parameter set also includes a dual-channel cross-validation threshold, which decreases with the number of weeks after surgery and is used to limit the maximum allowable value of the difference in activation time between the peroneus longus channel and the peroneus brevis channel.

[0008] Furthermore, the calculation of the integral value of the degree to which the flip angle exceeds the safety constraint threshold over time includes: The flip angle at each sampling moment is compared with the inward flip safety angle threshold. When the flip angle is greater than the inward flip safety angle threshold, the instantaneous anomaly degree at that sampling moment is the ratio of the flip angle to the inward flip safety angle threshold minus one. When the flip angle is less than or equal to the inward flip safety angle threshold, the instantaneous anomaly degree at that sampling moment is zero. The instantaneous anomaly degree is accumulated and integrated within the sliding time window to obtain the temporal anomaly degree integral. The preset conditions include: triggering direct order reduction when the time series anomaly integral exceeds a preset integration threshold; and triggering warning order reduction when the time series anomaly integral does not exceed the preset integration threshold, but the time series anomaly integrals of multiple consecutive sliding time windows are all greater than zero.

[0009] Furthermore, in the calculation of the temporal anomaly integral, the inversion safety angle threshold participates in both the calculation of the instantaneous anomaly and the setting of the preset integral threshold: When the inversion safety angle threshold is increased, the instantaneous anomaly corresponding to the same inversion angle decreases, and the preset integral threshold increases proportionally. When the inversion safety angle threshold decreases, the instantaneous anomaly corresponding to the same inversion angle increases, and the preset integral threshold decreases proportionally.

[0010] Furthermore, the step of calculating the deviation of the flip angle relative to the target angle based on the flip angle time-series data as the first evaluation index includes: Calculate the instantaneous deviation between the flip angle and the target angle at each sampling moment; Using the median of all instantaneous deviations within a round as the boundary, instantaneous deviations less than or equal to the median are assigned a first weighting coefficient, while instantaneous deviations greater than the median are assigned a second weighting coefficient greater than the first weighting coefficient. The weighted average value is calculated as the first evaluation metric.

[0011] Furthermore, the step of using the electromyographic signal to detect the activation response speed of the muscle group after the rollover disturbance event as a second evaluation index includes: Candidate activation times of the peroneus longus and peroneus brevis channels after a flip perturbation event were detected separately. Calculate the difference between the candidate activation times of the two channels, and compare the difference with the dual-channel cross-validation threshold corresponding to the current postoperative week number; If the difference is less than or equal to the dual-channel cross-validation threshold, the earlier candidate activation time is used as the activation time. If the difference is greater than the dual-channel cross-validation threshold, then the round data is discarded.

[0012] Furthermore, the prediction model is a Bayesian classification model; The output of the stage parameter set for the next training cycle includes: Enumerate the candidate parameter sets for the next training cycle. Each candidate parameter set includes the adjustment direction and magnitude of the safety constraint threshold, the adjustment direction and magnitude of the perturbation intensity level, and the switching direction of the training mode flag. Calculate the probability that the order reduction instruction will not be triggered during training under each candidate stage parameter set, and select the candidate stage parameter set with the highest probability as the stage parameter set for the next training cycle.

[0013] Furthermore, when the prediction model calculates the probability that the order reduction instruction is not triggered during the training process under each candidate stage parameter set, the total statistical value of the integral value in the state description vector is used as a penalty term in the probability calculation. The higher the total statistical value of the integral value, the higher the probability score will be obtained for the combination of candidates with smaller increases in the safety constraint threshold in the candidate stage parameter set.

[0014] Furthermore, the de-priority instruction includes branches with different priorities: When the integral value of the flip angle exceeding the safety constraint threshold triggers the order reduction condition first, the first priority order reduction is performed, the perturbation intensity level is reduced, and the training mode is switched to open-eye mode. When the first or second evaluation metric exceeds the corresponding allowable range, the second priority is downgraded to reduce the perturbation intensity level or switch the training mode.

[0015] Secondly, a closed-loop control system for dynamic balance training after ankle ligament surgery is provided, including: The parameter calling module is used to obtain the postoperative week number parameter of the trainee and call the stage parameter set corresponding to the postoperative week number parameter. The stage parameter set contains at least one safety constraint threshold that changes with the postoperative week number. The signal acquisition module is used to acquire the time-series data of the flip angle of the trainee during the training process and the electromyographic signals of at least one muscle group; The index calculation module is used to calculate the degree of deviation of the flip angle relative to the target angle based on the flip angle time series data as the first evaluation index, and to detect the activation response speed of the muscle group after the flip perturbation event based on the electromyographic signal as the second evaluation index. The integral calculation module is used to compare the flip angle time series data with the safety constraint threshold and calculate the integral value of the degree to which the flip angle exceeds the safety constraint threshold over time. The instruction decision module is used to generate a downgrade instruction when the integral value reaches a preset condition, or when the first evaluation index or the second evaluation index exceeds the corresponding allowable range; and to generate an upgrade instruction when downgrade is not triggered in multiple consecutive training rounds. The parameter prediction module is used to construct a state description vector using the current stage parameters, the statistical values ​​of the first evaluation index and the second evaluation index in the historical multi-round training, and the total statistical value of the integral value in the historical multi-round training. The state description vector is then input into the pre-trained prediction model, and the stage parameter set for the next training cycle is output. The instruction delivery module is used to send the generated instructions to the training terminal for execution via the communication interface.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention provides a closed-loop control method for dynamic balance training after ankle ligament surgery. The method uses the number of weeks post-surgery as the logical starting point of the entire closed-loop control system, simultaneously driving the dynamic configuration of safety constraint thresholds, the dual-dimensional acquisition of flip angle time-series data and electromyographic signals, the calculation of the integral value of the degree to which the flip angle exceeds the safety constraint threshold over time, the order reduction / upgrade decision jointly triggered by the integral value and evaluation indicators, and the output of the next training cycle stage parameter set based on a prediction model using a state description vector. In this invention, the number of weeks post-surgery determines the safety constraint threshold, which serves as the denominator for the instantaneous anomaly degree in the integral value calculation. The integral value and evaluation indicators jointly determine the order reduction / upgrade, and the order reduction / upgrade results, along with historical data, constitute the state description vector for the next round. The next round stage parameter set is returned as a new source of safety constraint thresholds. This invention ensures that each stage simultaneously provides input to the next stage and is constrained by the previous stage; the absence of any stage leads to a closed-loop break, improving proprioceptive reconstruction efficiency while ensuring training safety.

[0017] 2. This invention introduces an interlocking relationship between the varus safety angle threshold and the preset integral threshold in the integral value calculation. When the varus safety angle threshold increases with the number of weeks post-surgery, if the preset integral threshold remains unchanged, the difficulty of triggering downgrading for the same excessive behavior changes at different post-operative weeks, leading to inconsistent risk assessment sensitivity throughout the rehabilitation period. This invention maintains consistency in risk assessment sensitivity by allowing the preset integral threshold to change in the same direction as the varus safety angle threshold, ensuring that the risk quantification scale and risk judgment threshold maintain a fixed proportional relationship. Simultaneously, the two triggering conditions—direct downgrading and warning downgrading—correspond to acute and chronic risk accumulation modes, respectively. The former intervenes immediately when the integral exceeds the threshold, while the latter provides an early warning when the integral is not exceeded but multiple consecutive time windows are greater than zero.

[0018] 3. This invention employs a dual-channel cross-validation method in the detection of pre-activation delay in the peroneus muscle. Real muscle pre-activation occurs sequentially in the peroneus longus and peroneus brevis muscles, two synergistic muscles, with the difference in activation times between the two channels within a physiologically acceptable range. Motion artifacts, on the other hand, typically manifest as single-channel abrupt changes with no synchronous response in the contralateral channel, resulting in a significant difference between the two channels. Furthermore, the propagation paths of crosstalk in the contralateral muscle groups differ, leading to a larger phase difference in activation times. By calculating the difference in activation times between the two channels and comparing it with the dual-channel cross-validation threshold, real activation can be effectively distinguished from artifacts and crosstalk. The logic behind the decreasing dual-channel cross-validation threshold with postoperative weeks is as follows: in the early postoperative period, nerve conduction velocity is slower, resulting in a larger natural difference in activation times between the two channels, requiring a wider threshold to accommodate physiological delays; in the late postoperative period, nerve conduction function recovers, and the synchronicity of the two channels improves, leading to a corresponding tightening of the threshold to enhance detection accuracy.

[0019] 4. This invention employs a Bayesian classification model and incorporates the total statistical score as a penalty term into probability calculation. The Bayesian classification model integrates information from three dimensions—current stage parameters, recent evaluation index mean, and recent total statistical score—into a unified probabilistic framework for comprehensive evaluation. The logic behind using the total statistical score as a penalty term is as follows: a high total statistical score indicates that the patient's safety margin has been significantly depleted, and choosing a conservative approach can prevent further risk accumulation; a low total statistical score indicates that the patient still has sufficient safety margin, and choosing an aggressive approach can promote training progress. The setting of a lower probability threshold addresses the decision-making problem under incomplete information: when the posterior probability of all candidate approaches is below the threshold, maintaining the status quo and collecting more data is the optimal strategy. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart from Embodiment 1 of the present invention; Figure 2 This is a logic diagram of signal acquisition in Embodiment 1 of the present invention; Figure 3 This is a flowchart of the calculation of the first evaluation index in Embodiment 1 of the present invention; Figure 4 This is a flowchart of the calculation of the second evaluation index in Embodiment 1 of the present invention; Figure 5 This is a system block diagram in Embodiment 2 of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0022] Example 1: A closed-loop control method for dynamic balance training after ankle ligament surgery, such as... Figure 1 As shown, it includes the following steps: S1: Obtain the postoperative week number parameter of the trainee, call the stage parameter set corresponding to the postoperative week number parameter, and the stage parameter set shall contain at least one safety constraint threshold that changes with the postoperative week number. S2: Acquire the time-series data of the flip angle of the trainee during the training process and the electromyographic signals of at least one muscle group; S3: The deviation of the flip angle from the target angle is calculated based on the flip angle time series data as the first evaluation index, and the activation response speed of the muscle group after the flip disturbance event is detected based on the electromyographic signal as the second evaluation index. S4: Compare the flip angle time series data with the safety constraint threshold, and calculate the integral value of the degree to which the flip angle exceeds the safety constraint threshold over time; S5: Generate a downgrade instruction when the integral value reaches the preset condition, or when the first evaluation index or the second evaluation index exceeds the corresponding allowable range; generate an upgrade instruction when no downgrade is triggered in multiple consecutive training rounds. S6: The state description vector is constructed by the current stage parameters, the statistical values ​​of the first and second evaluation indicators in the historical training, and the total statistical value of the integral values ​​in the historical training. The state description vector is input into the pre-trained prediction model, and the stage parameter set of the next training cycle is output. S7: Send the generated instructions to the training terminal for execution via the communication interface.

[0023] In step S1, the postoperative week number parameter is first obtained. The postoperative week number parameter refers to the complete number of weeks that have elapsed since the trainee underwent ankle ligament surgery until the current training day, and is recorded in whole weeks. For example, if the trainee is in the 6th week after surgery, the postoperative week number parameter is 6. This parameter can be obtained either manually entered by the trainee or automatically read by scanning an identification tag.

[0024] Subsequently, based on the postoperative week number parameter, the corresponding stage parameter set is retrieved from the pre-stored database. The stage parameter set is a pre-defined collection of training control parameters, with each postoperative week number corresponding to an independent stage parameter set. Each stage parameter set contains at least one safety constraint threshold that varies with the postoperative week number. This safety constraint threshold limits the maximum permissible ankle rotation angle during training; it monotonically increases with postoperative week number to accommodate the gradual expansion of the safe range of motion during ligament healing.

[0025] Specifically, the safety constraint thresholds include the inversion safety angle threshold. The inversion safety angle threshold is used to limit the maximum allowable value of the flip angle during training, denoted as . ,in This represents the postoperative week number parameter. The varus safety angle threshold increases with postoperative week number; that is, a smaller value is used in the early postoperative period to protect the healing ligaments, and a larger value is used in the later postoperative period to allow for greater training range. In one specific implementation, the varus safety angle threshold is set to 4° at week 6 postoperatively, 8° at week 8, 10° at week 10, and 12° at week 12. The mapping relationship between the varus safety angle threshold and the postoperative week number is stored in the database in the form of a lookup table. The processor directly looks up the corresponding varus safety angle threshold based on the postoperative week number parameter.

[0026] The stage parameter set also includes a dual-channel cross-validation threshold. This threshold limits the maximum permissible difference in activation time between the peroneus longus and peroneus brevis channels, denoted as [insert threshold here]. The dual-channel cross-validation threshold decreases with postoperative weeks; a larger value is used in the early postoperative period to tolerate neuromuscular transmission delay, while a smaller value is used in the later postoperative period to improve detection accuracy. In one specific implementation, the dual-channel cross-validation threshold is set to 50ms at week 6 postoperatively, 40ms at week 8, 30ms at week 10, and 20ms at week 12. The mapping relationship between the dual-channel cross-validation threshold and the postoperative weeks is also stored in the database in the form of a lookup table. The processor directly looks up the corresponding dual-channel cross-validation threshold based on the postoperative week parameter.

[0027] The stage parameter set also includes perturbation intensity level, training mode indicator, upper limit of allowable joint position sense error, and upper limit of allowable pre-activation delay of the peroneus muscle. The perturbation intensity level controls the magnitude of the perturbation applied by the training terminal; a higher level indicates a larger perturbation amplitude and higher training difficulty. The training mode indicator indicates the current training mode, including four modes: open-eye bipedal, open-eye single-foot, closed-eye bipedal, and closed-eye single-foot, each corresponding to different proprioceptive training difficulties. The upper limit of allowable joint position sense error is used to determine the passing standard for the trainee's joint position sense recovery level, denoted as [insert value here]. This value decreases with increasing postoperative weeks. The upper limit of peroneal muscle preactivation delay is used to determine the passing standard for a trainee's peroneal muscle preactivation ability, denoted as [value missing]. This value decreases with the number of weeks after surgery.

[0028] In one specific implementation, the parameter set values ​​for each stage from week 6 to week 12 post-surgery are shown in Table 1.

[0029] Table 1. Parameter values ​​for each stage

[0030] The above parameter mapping relationship is pre-established based on clinical rehabilitation experience data and can be dynamically adjusted according to the actual training effect.

[0031] In step S2, as Figure 2 As shown, the roll angle time series data of the trainee during training is first acquired through a wearable inertial measurement unit (IMU). The wearable IMU employs a nine-axis IMU module, integrating a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, continuously acquiring posture data of the trainee's tibia and foot at a sampling frequency of no less than 100Hz. The processor fuses the tibial posture data with the foot posture data to obtain the relative roll angle time series between the tibia and foot, denoted as... ,in Indicates the sampling time. Flip angle. The positive direction is defined as the inversion direction of the ankle joint, that is, the angle of rotation of the sole of the foot towards the midline of the body. When the trainee's ankle joint is in a neutral position, The value is 0°; when the trainee's ankle joint inverts, The value is positive; when the trainee's ankle joint everts, The value is negative. The flip angle timing data is temporarily stored in the processor memory in the form of a circular buffer with a buffer length of not less than 5 seconds to ensure that a sufficient amount of historical data can be backtracked when a de-order or de-order event occurs.

[0032] Simultaneously, the electromyography (EMG) module acquires the trainee's EMG signals during training. The EMG module uses silver-silver chloride electrodes, applied to the bellies of the peroneus longus and peroneus brevis muscles on the affected lower limb, respectively, with a reference electrode applied to the tibial tuberosity. The EMG module continuously acquires EMG signals from both channels at a sampling frequency of at least 1000 Hz. The EMG signal from the peroneus longus channel is denoted as... The electromyographic signals of the peroneus brevis muscle channel are recorded as follows: The amplitude of the electromyographic (EMG) signal is measured in microvolts. The EMG signal is also temporarily stored in the processor memory in a circular buffer, with a buffer length of at least 5 seconds, synchronized with the buffer for the flip angle timing data.

[0033] This invention also acquires plantar pressure data of the trainee during training through a plantar pressure distribution pad. The plantar pressure distribution pad uses a flexible thin-film array pressure sensor, placed inside the insole of the trainee's affected side, and continuously collects pressure distribution data for various areas of the sole at a sampling frequency of no less than 50Hz. The processor calculates the location of the center of force on the sole based on the plantar pressure distribution data, and records it as... The location of the center of force on the sole of the foot is used to mark the start and end of the standing phase, thereby helping to determine whether the trainee is in a stable single-leg or double-leg standing state, and whether the trainee is in a weight-bearing state when a rollover disturbance occurs.

[0034] It should be noted that all sensor data are timestamped to ensure strict alignment of flip angle timing data, electromyographic signal data, and plantar pressure data on the timeline. The timestamps are in milliseconds and synchronized with the processor system clock. All sensor data is temporarily stored in a circular buffer with a capacity of at least 5000 sampling points, corresponding to a data length of at least 5 seconds. When the buffer is full, the oldest data is overwritten by the newest data to ensure that complete data records for the most recent period are always preserved.

[0035] In step S3, the first evaluation index is calculated. The first evaluation index is joint positional perception error, which is used to quantitatively describe the accuracy with which the trainee reproduces the target flip angle when the eyes are closed.

[0036] like Figure 3 As shown, in the blindfolded reproduction phase, the processor provides the trainee with a target flip angle via voice prompts or screen display, denoted as . After receiving the target angle, the trainee closes their eyes, actively adjusts the affected ankle to a position they believe has reached the target rotation angle, and holds this position for 3 seconds. During this holding period, the processor uses the rotation angle timing data obtained in step S2. Based on this, calculate the measured flip angle corresponding to each sampling number n. From the perspective of the target instantaneous deviation between The calculation formula is: Instantaneous deviation The unit is degrees. The smaller the value, the closer the flip angle of the trainee at the current sampling moment is to the target angle, and the more accurate the joint position sense is.

[0037] After completing the instantaneous deviation calculation, the instantaneous deviation... Weighted processing is performed. The processor calculates all instantaneous deviations within the current round. The median, denoted as The median is the middle value after arranging all instantaneous deviations in ascending order. When the number of sampling points is even, the average of the two middle values ​​is taken. The processor uses the median. To define the boundaries, different weighting coefficients are assigned to instantaneous deviations located on different sides of the median. Specifically, for deviations less than or equal to the median... The instantaneous deviation is assigned a first weighting coefficient; for values ​​greater than the median... The instantaneous deviation is assigned a second weighting coefficient, which is greater than the first weighting coefficient. The calculation formula is as follows: ,in This is the amplification factor, for example, a value of 2.0. The design principle of the weighting factor is as follows: instantaneous deviations less than or equal to the median are common normal fluctuations during training and are given lower weights; instantaneous deviations greater than the median are abnormal fluctuations with larger deviations and are given higher weights, so that large deviations contribute more significantly to the final evaluation results, effectively distinguishing between noise deviations caused by occasional jitters and systematic deviations caused by insufficient proprioception.

[0038] Finally, the weighted average was calculated as the first evaluation indicator. The calculation formula is: ; Where N is the total number of sampling points in the current round. represents the weighting coefficient corresponding to the sampling sequence number n. The unit of the first evaluation index is degrees; the smaller the value, the more accurate the trainee's joint position sense. The blindfolded reproduction operation is repeated three times in each training round, and the average of the three weighted joint position sense errors is taken as the final value of the first evaluation index for this round of training.

[0039] Next, the second evaluation metric was calculated. The second evaluation metric was the peroneal muscle preactivation delay, which is used to quantitatively describe the speed at which the trainee's peroneal muscle produces an active contraction response after being subjected to inversion perturbation.

[0040] like Figure 4 As shown, the timing of the inversion disturbance event is first detected. The processor processes the flip angle timing data. Perform a difference operation to calculate the rate of change of the flip angle between adjacent sampling points, denoted as . When the rate of change of the flip angle When the change per sampling point exceeds 15 degrees per second, that is... ,in The sampling frequency of the flip angle timing data is expressed in Hertz. The processor marks the time corresponding to the current sampling sequence number n as the time when the inversion disturbance event occurs, denoted as n. The moment of the inversion disturbance event is the time reference point for calculating the pre-activation delay.

[0041] After detecting an inversion disturbance event, electromyographic signals in the peroneus longus channel were analyzed. and peroneus brevis muscle channel electromyographic signals Preprocessing is performed separately. The processor first performs bandpass filtering on the electromyography signals from both channels, with a passband range of 20Hz to 450Hz, to remove low-frequency motion artifacts and high-frequency electromagnetic interference. The processor then takes the absolute value of the filtered signal and performs root mean square (RMS) calculation using a sliding window with a length of 50 sampling points. The formula for calculating the RMS of the sliding window is: ; Where M is the length of the sliding window. The amplitude of the electromyographic signal from the peroneus longus channel corresponding to sampling number m after bandpass filtering and absolute value taking is given. The root mean square of the sliding window for the peroneus brevis channel is also given. The same calculation formula is used. The root mean square of the sliding window is used to represent random fluctuations in smooth muscle electromyography (EMG) signals, highlighting changes in signal amplitude caused by muscle activation.

[0042] After preprocessing, baseline parameters at rest are calculated. The processor selects electromyography data within 5 seconds prior to the occurrence of the inversion perturbation event as the resting baseline segment, and calculates the mean root mean square value of the peroneus longus channel sliding window during this time period. and standard deviation And the mean of the root mean square of the peroneus brevis muscle channel sliding window. and standard deviation Baseline parameters are used to set the threshold for determining muscle activation.

[0043] After baseline parameters were calculated, candidate activation times for both channels were detected. For the peroneus longus channel, the processor started from the time of the inversion perturbation event. Start scanning backwards through the root mean square sequence of the sliding window ,when When the threshold of the baseline mean plus three times the baseline standard deviation is first exceeded, the condition is satisfied. The time corresponding to the current sampling sequence number n is recorded as the candidate activation time of the peroneus longus muscle channel, denoted as... For the peroneus brevis muscle passage, the same criteria are applied, and the first time it exceeds... The sampling sequence number is recorded as the candidate activation time of the peroneus brevis muscle channel, denoted as... .

[0044] After candidate activation times are detected, cross-validation is performed on the candidate activation times of the two channels. The processor calculates the difference between the candidate activation times of the two channels. The unit is the number of sampling sequence numbers. The processor will calculate the difference. Convert to time difference The unit is milliseconds. The processor obtains the dual-channel cross-validation threshold corresponding to the current postoperative week number from the stage parameter set called in step S1. The dual-channel cross-validation threshold is used to limit the maximum allowable difference in activation time between the peroneus longus and peroneus brevis channels. If the time difference... Less than or equal to the two-channel cross-validation threshold If the detection result is positive, the event is considered a valid pre-activation event, and the earlier candidate activation time between the two channels is taken as the valid activation time. If the time difference Greater than the two-channel cross-validation threshold If the result is not found to be a motion artifact or electromyographic crosstalk, the data from this round is discarded and marked as an invalid perturbation event. The design principle of dual-channel cross-validation is that: true muscle preactivation will occur simultaneously or sequentially in the two synergistic muscle groups of peroneus longus and peroneus brevis, and the difference in activation time between the two channels is usually within the physiologically permissible range; while motion artifacts usually manifest as a single-channel mutation with no synchronous response in the contralateral channel. The propagation direction of crosstalk in the contralateral muscle group is different from that of the target muscle, resulting in a large difference in activation time between the two channels. Dual-channel cross-validation can effectively filter out the above interference.

[0045] After cross-validation, the second evaluation metric was calculated. The second evaluation metric was the preactivation delay of the peroneus muscle, denoted as [insert metric here]. The unit is milliseconds. The formula for calculating the pre-activation delay is: The smaller the pre-activation delay value, the faster the trainee's peroneal muscle generates an active contraction response after being subjected to inversion perturbation, and the better the integrity of the neuromuscular reflex arc.

[0046] In addition, use the center position of the sole of the foot. Used to assist in determining whether the trainee is in a stable standing position. Before calculating the first and second evaluation metrics, the processor first... The distribution of the data determines whether the trainee is currently in an effective training posture. If... If the fluctuation range exceeds a preset threshold over a continuous period, the trainee is determined to be in an unstable state. The calculation of the indicators for the current round is paused, and the process continues only after the trainee regains their footing. The use of the plantar force center position ensures that the trainee remains in an effective training posture throughout the indicator calculation process, avoiding distortion of evaluation results due to postural instability.

[0047] In step S4, the instantaneous anomaly degree corresponding to each sampling number is first calculated. The flip angle time-series data obtained in step S2 and the inverted safety angle threshold called in step S1 are used as inputs. The flip angle corresponding to each sampling number is... With inversion safety angle threshold Compare. When the angle is flipped... Greater than the inversion safety angle threshold When the sample number n indicates that the flip angle of the trainee at that sampling moment exceeds the safe range allowed for the current postoperative weeks, the processor uses the ratio of the flip angle to the safe inversion angle threshold minus one as the instantaneous anomaly degree corresponding to that sampling number n. The calculation formula is as follows: When the angle is flipped Less than or equal to the inversion safety angle threshold When the flip angle of the trainer at that sampling moment is within a safe range, the processor sets the instantaneous anomaly degree corresponding to the sampling number n to zero. The instantaneous anomaly rate is a dimensionless value that reflects the relative extent to which the flip angle exceeds the safety constraint threshold. For example, when the inward flip safety angle threshold is 8°, if the flip angle reaches 10°, the instantaneous anomaly rate is 10 / 8−1=0.25; if the flip angle reaches 12°, the instantaneous anomaly rate is 12 / 8−1=0.5. The larger the instantaneous anomaly rate, the more severe the relative extent to which the flip angle exceeds the safety constraint threshold.

[0048] After calculating the transient anomaly score, the processor integrates the transient anomalies within a sliding time window to obtain the temporal anomaly integral. The sliding time window is a fixed-length interval of continuous sampling numbers used to limit the time range of the integration calculation. The processor takes the current sampling number n as the endpoint and truncates a sliding time window of length W sampling numbers backward, where W corresponds to a training duration of 30 seconds. Within the sliding time window, the processor calculates the transient anomalies corresponding to all sampling numbers. Perform cumulative integration, where m ranges from n−W+1 to n. The formula for calculating the integral of temporal anomaly degree is: ; in, This represents the integral of the temporal anomaly degree within a sliding time window ending at the sampling number n. Let m be the instantaneous anomaly degree corresponding to the sampling sequence number. The time interval between adjacent sampling points is expressed in seconds and is equal to the reciprocal of the sampling frequency. A larger temporal anomaly integral indicates a higher level of accumulated safety risk to the trainee during the most recent training period.

[0049] Then, based on the integral of the temporal anomaly degree, it is determined whether to trigger direct order reduction. The integral of the temporal anomaly degree of the current sliding time window is calculated. Compared with the preset integration threshold The comparison is performed. The preset integration threshold is a pre-defined constant, such as 2.0 seconds. This is used for integrating time-series anomalies. Exceeding the preset points threshold When the processor triggers a direct reduction instruction, it indicates that the safety risk accumulated by the trainee within the current sliding time window has reached a level requiring immediate intervention. The direct reduction instruction immediately lowers the training difficulty to protect the trainee's ligaments from further damage. The design principle of direct reduction is that a single, brief overshoot of the rotation angle usually does not cause substantial damage to healing ligaments, but continuous or repeated overshoots accumulate risk, and timely intervention is necessary when the accumulated risk exceeds a certain threshold.

[0050] Simultaneously, the processor determines whether to trigger a warning-based downgrade based on the integral of the timing anomaly degree across multiple consecutive sliding time windows. The processor records the integral values ​​of the timing anomaly degree across the most recent multiple consecutive sliding time windows, with the number of consecutive sliding time windows being three. When the timing anomaly degree integral... The score threshold was not exceeded. However, if the temporal anomaly integral is greater than zero for three consecutive sliding time windows, it indicates that although the trainee has not experienced a single severe risk accumulation, it has shown a persistent trend of slight over-limit behavior. In this case, the processor triggers a warning-based reduction instruction. This warning-based reduction instruction intervenes before the risk accumulation reaches a dangerous level, appropriately reducing the training difficulty to prevent further risk accumulation. The design principle of warning-based reduction is that early intervention when an over-limit trend forms but has not yet reached a dangerous level can effectively prevent the risk from developing from quantitative to qualitative change, reflecting a safety philosophy of prevention first.

[0051] In some preferred examples, the inversion safety angle threshold Simultaneously participate in transient anomaly degree Calculation and preset integration threshold The settings establish an interlocking relationship between the two. In the calculation of instantaneous anomaly, the inversion safety angle threshold is used as a denominator: when the inversion safety angle threshold increases, the instantaneous anomaly corresponding to the same inversion angle decreases, making the same degree of over-limit behavior assessed as a lower risk level; when the inversion safety angle threshold decreases, the instantaneous anomaly corresponding to the same inversion angle increases, making the same degree of over-limit behavior assessed as a higher risk level. In the setting of the preset integral threshold, changes in the inversion safety angle threshold simultaneously cause changes in the preset integral threshold in the same direction: when the inversion safety angle threshold increases, the preset integral threshold increases proportionally, increasing the risk accumulation level required to trigger downgrading; when the inversion safety angle threshold decreases, the preset integral threshold decreases proportionally, decreasing the risk accumulation level required to trigger downgrading.

[0052] In the early postoperative period, the varus safety angle threshold is relatively small. At this stage, ligament healing is not yet complete, and even a slight overshoot of the varus angle can cause damage. Therefore, the transient abnormality is highly sensitive to overshooting behavior, and the smaller preset integral threshold makes downsizing easier to trigger. In the late postoperative period, the varus safety angle threshold increases. At this stage, the ligament's load-bearing capacity is enhanced, allowing for a certain degree of overshooting without immediate downsizing. Therefore, the transient abnormality is less sensitive to overshooting behavior, and the larger preset integral threshold makes downsizing more difficult to trigger. This interlocking mechanism automatically matches the strictness of safety constraints with the ligament healing process without manual intervention.

[0053] In step S5, it is first determined whether the triggering conditions for the downgrade instruction are met. The triggering conditions for the downgrade instruction include three parallel conditions; the downgrade instruction is triggered as long as any one of these conditions is met.

[0054] The first condition for triggering a downgrade is when the integral of the timing anomaly degree reaches a preset condition. When step S4 determines that a direct downgrade or a warning downgrade is triggered, the processor generates a downgrade instruction.

[0055] The second downgrade trigger condition is that the first evaluation metric exceeds the corresponding allowable range. The processor will then use the first evaluation metric calculated in step S3. The upper limit of allowable joint position perception error called in step S1 Comparison. When Greater than At that time, the processor generates a downgrade instruction.

[0056] The third downgrade trigger condition is that the second evaluation metric exceeds the corresponding allowable range. The processor will then use the second evaluation metric calculated in step S3. Upper limit of peroneal muscle preactivation delay allowed in step S1 Comparison. When Greater than At that time, the processor generates a downgrade instruction.

[0057] Degradation instructions include branches of different priorities to handle different types of security risks. When multiple degradation trigger conditions are met simultaneously, the processor executes the corresponding degradation operations in priority order.

[0058] The first priority downgrade is triggered by the integral value of the flip angle exceeding the safety constraint threshold. When it is determined in step S4 that a direct downgrade is triggered, the processor executes the first priority downgrade, lowers the perturbation intensity level by two levels, and switches the training mode flag to open mode.

[0059] The second priority demotion is triggered when either the first or second evaluation metric exceeds the corresponding allowable range. When the first evaluation metric... Exceeding the upper limit of allowable joint positional error Or the second evaluation indicator Exceeding the upper limit of peroneal muscle preactivation delay At this time, the processor performs a second priority downgrade, either by lowering the perturbation intensity level by one level or by switching the training mode flag to open-eye mode.

[0060] Simultaneously, it determines whether the triggering condition for an upgrade instruction is met. The triggering condition for an upgrade instruction is that no downgrade instruction has been triggered for multiple consecutive training rounds. The processor maintains a training round counter, recording the number of consecutive training rounds without triggering a downgrade since the last downgrade. When the number of consecutive training rounds without triggering a downgrade reaches a preset value, the processor generates an upgrade instruction, either increasing the perturbation intensity level by one level or switching the training mode flag to a higher difficulty level.

[0061] In some preferred embodiments, the present invention also processes pain signals emitted by the trainee. The trainee can send pain signals to the processor via a handheld button or voice command. When the processor receives a pain signal, it immediately generates a pause command. The pause command is executed as follows: the processor sends a pause command to the training terminal through the communication interface, the training terminal stops all disturbances and returns to a horizontal position, maintains this position for 30 seconds, and simultaneously issues a voice prompt.

[0062] In step S6, the state description vector is first constructed. The state description vector is a multi-dimensional numerical vector used to comprehensively represent the trainer's overall state at the end of the current training cycle, serving as the input for the prediction model to determine the optimal parameter configuration for the next training cycle. The state description vector consists of three parts.

[0063] The first part consists of the current stage parameters. The processor extracts the currently used parameter values ​​from the stage parameter set called in step S1, including the inversion safety angle threshold. Disturbance intensity level Training mode flags Upper limit of joint positional perception error Upper limit of peroneal muscle preactivation delay and dual-channel cross-validation threshold These parameter values ​​directly reflect the difficulty level of the current training cycle.

[0064] The second part contains the statistical values ​​of the first and second evaluation metrics across multiple training iterations. The processor reads the recorded values ​​of the first and second evaluation metrics from the training history database for the most recent specified iterations. The number of recent specified iterations is 3. The processor calculates the average of the first evaluation metric over the most recent 3 training iterations, denoted as . The processor simultaneously calculates the mean of the second evaluation metric over the most recent three training iterations, denoted as... These two statistics reflect the average performance of the trainee's proprioceptive recovery level over a recent period, effectively smoothing out any occasional fluctuations that may occur during a single training session.

[0065] The third part is the total statistical value of the integrals from multiple training rounds. The processor reads the temporal anomaly integral records for each sliding time window within the most recent specified training period from the training history database. The length of the most recent specified training period is 1 week. The processor sums up the temporal anomaly integrals of all sliding time windows within the most recent week to obtain the total statistical value, denoted as . The total score reflects the trainee's accumulated level of safety risk exposure over the past week and is an important indicator for assessing the trainee's tolerance and risk-bearing capacity.

[0066] State description vector The mathematical representation of is: .

[0067] After the state description vector is constructed, the processor will... Input a pre-trained prediction model. This prediction model is a Bayesian classification model, pre-trained based on a large amount of historical training data. It can predict the probability of not triggering a reduction instruction during training under different candidate stage parameter sets, based on the input state description vector. The Bayesian classification model calculates the posterior probability using Bayes' theorem, i.e., given the current state description vector... Under the condition of selecting a certain candidate stage parameter set The probability of not triggering the order reduction instruction during post-training ,in This indicates an event that does not trigger a downgrade instruction during training.

[0068] First, enumerate all possible candidate parameter sets for the next training cycle. Each candidate parameter set... The adjustment plan includes three dimensions.

[0069] The first dimension is the direction and magnitude of the adjustment to the safety constraint threshold. The adjustment of the safety constraint threshold is based on the inward turning safety angle threshold. The candidate adjustment schemes include three options: keeping the current value unchanged, increasing by 2°, and decreasing by 2°.

[0070] The second dimension is the adjustment direction and magnitude of the disturbance intensity level. Disturbance Intensity Level The proposed adjustment options include three choices: keeping the current level unchanged, raising it by one level, or lowering it by one level.

[0071] The third dimension is the direction of the training mode flag switching. Training mode flag The candidate adjustment schemes include two options: keeping the current mode unchanged and switching to the closed-eye direction. The closed-eye direction refers to the progressive direction of switching from open-eye two feet to open-eye one foot, from open-eye one foot to closed-eye two feet, and from closed-eye two feet to closed-eye one foot.

[0072] The total number of combinations of all candidate adjustment schemes is 3×3×2=18, which means that 18 candidate stage parameter sets need to be enumerated.

[0073] For each candidate stage parameter set The processor uses a Bayesian classification model to calculate the probability that a downgrade instruction will not be triggered during training under the parameter set of this candidate stage. In the probability calculation process, the state description vector Total integral values ​​in the statistics It participates in probability calculation as a penalty term. The mechanism of the penalty term is as follows: when the integral value is statistically totaled... A higher score indicates that the trainee has accumulated significant safety risk exposure within the past week. In this case, combinations with smaller increases in the safety constraint threshold within the candidate parameter set will receive higher probability scores, meaning the model tends to choose conservative parameter adjustment schemes to avoid further risk accumulation. When the total integral value is statistically significant... A lower threshold indicates that the trainee has had less exposure to safety risks in the past week. In this case, combinations with larger increases in the safety constraint threshold in the candidate stage parameter set will receive higher probability scores, meaning the model tends to choose aggressive parameter adjustment schemes to promote training progress. The introduction of a penalty term allows the prediction model to consider not only the trainee's average performance level but also the cumulative risk exposure level during training, thereby achieving a more personalized prescription evolution.

[0074] After iterating through all 18 candidate stage parameter sets, select the one that makes... The largest candidate stage parameter set is used as the stage parameter set for the next training cycle. If the maximum probability of all candidate stage parameter sets is still below 0.6, it means that the probability of not triggering a downgrade instruction during training is low regardless of the parameter adjustment scheme chosen in the current state. In this case, the processor adds an extra week of maintenance, neither advancing nor regressing, keeping the current stage parameter set unchanged, collecting more training data, and then making the next round of decision-making.

[0075] In step S7, the instructions generated in step S5 are first classified and encapsulated. Step S5 may generate three types of instructions: de-leveling instructions, rank-up instructions, and pause instructions. Each type of instruction contains different parameter information and needs to be encapsulated according to a unified communication protocol so that the training terminal can correctly parse and execute it.

[0076] The order reduction command includes the following parameters: command type identifier "order reduction command", target value for perturbation intensity level, target value for training mode flag, and order reduction priority flag. The order promotion command includes the following parameters: command type identifier "order promotion command", target value for perturbation intensity level, and target value for training mode flag. The pause command includes the following parameters: command type identifier "pause command", command to return the training terminal to a horizontal position, and a pause duration setting of 30 seconds.

[0077] After instruction encapsulation, the processor establishes a connection with the training terminal through the communication interface. This communication interface is a wireless communication interface, including both Bluetooth and WiFi communication. The processor will first attempt to establish a connection with the training terminal via Bluetooth. If the Bluetooth connection fails, the processor will automatically switch to the WiFi communication interface.

[0078] After the communication connection is established, the processor sends the encapsulated instruction data to the training terminal through the communication interface. The transmission of instruction data uses an acknowledgment mechanism, meaning the training terminal returns an acknowledgment signal to the processor upon receiving the instruction data. If the processor does not receive an acknowledgment signal within a preset time, it determines that the transmission has failed and automatically retransmits the instruction data, up to three times. If no acknowledgment signal is received after three retransmissions, the processor determines that there is a communication failure and prompts the trainee to check the communication connection status via voice announcement.

[0079] After receiving the instruction data, the training terminal executes the corresponding action according to the instruction type.

[0080] When the training terminal receives a reduction order command, it adjusts its perturbation mechanism according to the target perturbation intensity level contained in the command, reducing the perturbation amplitude to the value corresponding to the target level. Simultaneously, the training terminal adjusts the training mode according to the target value of the training mode flag. While executing the reduction order command, the training terminal also provides a voice prompt to the trainee indicating that the training difficulty has been reduced.

[0081] When the training terminal receives an upgrade command, it adjusts its perturbation mechanism according to the target perturbation intensity level contained in the command, increasing the perturbation amplitude to the value corresponding to the target level. Simultaneously, the training terminal adjusts the training mode according to the target training mode flag. While executing the upgrade command, the training terminal also notifies the trainee via voice announcement that the training difficulty has increased.

[0082] When the training terminal receives a pause command, it immediately stops all disturbance mechanisms and returns the training platform to a horizontal position. After returning to a horizontal position, the training terminal locks the platform, maintaining its stability. Simultaneously, the training terminal announces via voice that the trainee can rest and begins a 30-second timer. After the 30-second timer expires, the training terminal announces via voice that the trainee is ready to continue training and unlocks the platform, awaiting the processor's next training command.

[0083] After executing the instructions, the training terminal returns execution result feedback to the processor. This feedback includes an indicator of success or failure, as well as the training terminal's state parameters after execution, including the current perturbation intensity level, training mode flag, and the angle position of the training platform. Upon receiving the execution result feedback, the processor records the result in the training history database.

[0084] If the training terminal returns a failure response, the processor will handle the situation accordingly based on the cause of the failure. If the failure is due to a hardware malfunction in the training terminal, the processor will prompt the trainee to stop training and contact maintenance personnel via voice announcement. If the failure is due to abnormal instruction parameters, the processor will repackage the instruction and reissue it.

[0085] It should be noted that this embodiment describes the processor as the execution subject, but the scope of protection of the present invention is not limited thereto. The processor may be an embedded controller built into the training terminal, a microprocessor built into a wearable device, a central processing unit in a general-purpose smartphone or tablet, a computing chip in an edge computing node, a virtual computing instance in a cloud server, or a distributed combination of the above-mentioned computing devices.

[0086] For example, when a cloud server is used as the execution entity, the training terminal and wearable sensors upload flip angle time-series data, electromyographic signal data, and plantar pressure data to the cloud server in real time via a communication network. The cloud server then executes all the computational logic in steps S1 to S6 and generates instructions, which are then sent to the training terminal for execution. When an edge computing node is used as the execution entity, the edge computing node is responsible for executing the computational tasks with high real-time requirements in steps S2 to S5, while the cloud server is responsible for executing the management and prediction tasks with lower real-time requirements in steps S1 and S6. When the training terminal's built-in controller or the wearable device's built-in processor is used as the execution entity, all computational logic is completed locally without relying on external computing devices. The communication interface can be a Bluetooth interface, a WiFi interface, a mobile communication interface, a wired Ethernet interface, or a near-field communication interface.

[0087] Example 2: A closed-loop control system for dynamic balance training after ankle ligament surgery. This system is used to implement the closed-loop control method for dynamic balance training after ankle ligament surgery described in Example 1, such as... Figure 5 As shown, it includes a parameter calling module, a signal acquisition module, an indicator calculation module, an integral calculation module, an instruction decision module, a parameter prediction module, and an instruction issuance module.

[0088] The system includes the following modules: a parameter retrieval module for acquiring the trainee's post-operative week number parameter and retrieving the corresponding stage parameter set, which includes at least one safety constraint threshold that varies with the post-operative week number; a signal acquisition module for acquiring the trainee's rotation angle time-series data and electromyographic signals of at least one muscle group during training; an index calculation module for calculating the deviation of the rotation angle from the target angle based on the rotation angle time-series data as a first evaluation index and detecting the activation response speed of the muscle group after the rotation disturbance event based on the electromyographic signals as a second evaluation index; and an integration calculation module for comparing the rotation angle time-series data with the safety constraint threshold and calculating the deviation of the rotation angle from the target angle. The system comprises: a safety constraint threshold accumulated over time as an integral value; an instruction decision module, used to generate a downgrade instruction when the integral value reaches a preset condition, or when the first or second evaluation metric exceeds the corresponding allowable range; and an upgrade instruction when no downgrade is triggered in multiple consecutive training rounds; a parameter prediction module, used to construct a state description vector using the current stage parameters, the statistical values ​​of the first and second evaluation metric in multiple historical training rounds, and the total statistical value of the integral value in multiple historical training rounds, inputting the state description vector into the pre-trained prediction model, and outputting the stage parameter set for the next training cycle; and an instruction delivery module, used to deliver the generated instructions to the training terminal for execution via a communication interface.

[0089] Working Principle: This invention uses the postoperative week number parameter as the logical starting point of the entire closed-loop control system. This parameter simultaneously drives the dynamic configuration of the safety constraint threshold, the dual-dimensional acquisition of flip angle time-series data and electromyographic signals, the calculation of the integral value of the degree to which the flip angle exceeds the safety constraint threshold over time, the order reduction / upgrade decision jointly triggered by the integral value and evaluation index, and the output of the next training cycle stage parameter set based on the state description vector. In this invention, the postoperative week number parameter determines the safety constraint threshold, which serves as the denominator of the instantaneous anomaly degree in the integral value calculation. The integral value and evaluation index jointly determine the order reduction / upgrade, and the order reduction / upgrade results, along with historical data, constitute the state description vector for the next round. The next round stage parameter set is returned as a new source of the safety constraint threshold. This invention ensures that each step simultaneously provides input to the next step and is constrained by the previous step. The absence of any step leads to a closed-loop break, improving proprioceptive reconstruction efficiency while ensuring training safety.

[0090] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0091] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0092] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0094] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A closed-loop control method for dynamic balance training after ankle ligament surgery, characterized in that, Includes the following steps: Obtain the postoperative week number parameter of the trainee, and call the stage parameter set corresponding to the postoperative week number parameter. The stage parameter set contains at least one safety constraint threshold that changes with the postoperative week number. Acquire the time-series data of the flip angle of the trainee during the training process and the electromyographic signals of at least one muscle group; The degree of deviation of the flip angle relative to the target angle is calculated based on the flip angle time series data as the first evaluation index, and the activation response speed of the muscle group after the flip perturbation event is detected based on the electromyographic signal as the second evaluation index. The flip angle time series data is compared with the safety constraint threshold, and the integral value of the degree to which the flip angle exceeds the safety constraint threshold over time is calculated. When the integral value reaches a preset condition, or when the first evaluation index or the second evaluation index exceeds the corresponding allowable range, a downgrade instruction is generated. When no order reduction is triggered in multiple consecutive training rounds, an order increase instruction is generated. A state description vector is constructed using the current stage parameters, the statistical values ​​of the first and second evaluation indicators in the historical multi-round training, and the total statistical value of the integral values ​​in the historical multi-round training. The state description vector is input into the pre-trained prediction model, and the stage parameter set for the next training cycle is output. The generated instructions are sent to the training terminal for execution via the communication interface.

2. The closed-loop control method for dynamic balance training after ankle ligament surgery according to claim 1, characterized in that, The safety constraint threshold includes an inversion safety angle threshold, which increases with the number of weeks after surgery and is used to limit the maximum allowable value of the inversion angle during training. The stage parameter set also includes a dual-channel cross-validation threshold, which decreases with the number of weeks after surgery and is used to limit the maximum allowable value of the difference in activation time between the peroneus longus channel and the peroneus brevis channel.

3. The closed-loop control method for dynamic balance training after ankle ligament surgery according to claim 1, characterized in that, The integral value of the degree to which the flip angle exceeds the safety constraint threshold accumulated over time includes: The flip angle at each sampling moment is compared with the inward flip safety angle threshold. When the flip angle is greater than the inward flip safety angle threshold, the instantaneous anomaly degree at that sampling moment is the ratio of the flip angle to the inward flip safety angle threshold minus one. When the flip angle is less than or equal to the inward flip safety angle threshold, the instantaneous anomaly degree at that sampling moment is zero. The instantaneous anomaly degree is accumulated and integrated within the sliding time window to obtain the temporal anomaly degree integral. The preset conditions include: triggering direct order reduction when the time series anomaly integral exceeds a preset integration threshold; and triggering warning order reduction when the time series anomaly integral does not exceed the preset integration threshold, but the time series anomaly integrals of multiple consecutive sliding time windows are all greater than zero.

4. The closed-loop control method for dynamic balance training after ankle ligament surgery according to claim 3, characterized in that, In the calculation of the temporal anomaly integral, the inversion safety angle threshold is involved in both the calculation of the instantaneous anomaly and the setting of the preset integral threshold. When the inversion safety angle threshold is increased, the instantaneous anomaly corresponding to the same inversion angle decreases, and the preset integral threshold increases proportionally. When the inversion safety angle threshold decreases, the instantaneous anomaly corresponding to the same inversion angle increases, and the preset integral threshold decreases proportionally.

5. The closed-loop control method for dynamic balance training after ankle ligament surgery according to claim 1, characterized in that, The step of calculating the deviation of the flip angle relative to the target angle based on the flip angle time-series data as the first evaluation index includes: Calculate the instantaneous deviation between the flip angle and the target angle at each sampling moment; Using the median of all instantaneous deviations within a round as the boundary, instantaneous deviations less than or equal to the median are assigned a first weighting coefficient, while instantaneous deviations greater than the median are assigned a second weighting coefficient greater than the first weighting coefficient. The weighted average value is calculated as the first evaluation metric.

6. The closed-loop control method for dynamic balance training after ankle ligament surgery according to claim 1, characterized in that, The second evaluation index, which uses the speed of muscle group activation response after a rollover disturbance event detected by the electromyographic signal, includes: Candidate activation times of the peroneus longus and peroneus brevis channels after a flip perturbation event were detected separately. Calculate the difference between the candidate activation times of the two channels, and compare the difference with the dual-channel cross-validation threshold corresponding to the current postoperative week number; If the difference is less than or equal to the dual-channel cross-validation threshold, the earlier candidate activation time is used as the activation time. If the difference is greater than the dual-channel cross-validation threshold, then the round data is discarded.

7. The closed-loop control method for dynamic balance training after ankle ligament surgery according to claim 1, characterized in that, The prediction model is a Bayesian classification model; The output of the stage parameter set for the next training cycle includes: Enumerate the candidate parameter sets for the next training cycle. Each candidate parameter set includes the adjustment direction and magnitude of the safety constraint threshold, the adjustment direction and magnitude of the perturbation intensity level, and the switching direction of the training mode flag. Calculate the probability that the order reduction instruction will not be triggered during training under each candidate stage parameter set, and select the candidate stage parameter set with the highest probability as the stage parameter set for the next training cycle.

8. The closed-loop control method for dynamic balance training after ankle ligament surgery according to claim 7, characterized in that, When the prediction model calculates the probability that the order reduction instruction is not triggered during the training process under each candidate stage parameter set, the total statistical value of the integral value in the state description vector is used as a penalty term in the probability calculation. The higher the total statistical value of the integral value, the higher the probability score will be obtained for the combination of candidates with smaller increases in the safety constraint threshold in the parameter set.

9. The closed-loop control method for dynamic balance training after ankle ligament surgery according to claim 1, characterized in that, The demotion instruction includes branches with different priorities: When the integral value of the flip angle exceeding the safety constraint threshold triggers the order reduction condition first, the first priority order reduction is performed, the perturbation intensity level is reduced, and the training mode is switched to open-eye mode. When the first or second evaluation metric exceeds the corresponding allowable range, the second priority is downgraded to reduce the perturbation intensity level or switch the training mode.

10. A closed-loop control system for dynamic balance training after ankle ligament surgery, characterized in that, include: The parameter calling module is used to obtain the postoperative week number parameter of the trainee and call the stage parameter set corresponding to the postoperative week number parameter. The stage parameter set contains at least one safety constraint threshold that changes with the postoperative week number. The signal acquisition module is used to acquire the time-series data of the flip angle of the trainee during the training process and the electromyographic signals of at least one muscle group; The index calculation module is used to calculate the degree of deviation of the flip angle relative to the target angle based on the flip angle time series data as the first evaluation index, and to detect the activation response speed of the muscle group after the flip perturbation event based on the electromyographic signal as the second evaluation index. The integral calculation module is used to compare the flip angle time series data with the safety constraint threshold and calculate the integral value of the degree to which the flip angle exceeds the safety constraint threshold over time. The instruction decision module is used to generate a downgrade instruction when the integral value reaches a preset condition, or when the first evaluation index or the second evaluation index exceeds the corresponding allowable range. When no order reduction is triggered in multiple consecutive training rounds, an order increase instruction is generated. The parameter prediction module is used to construct a state description vector using the current stage parameters, the statistical values ​​of the first evaluation index and the second evaluation index in the historical multi-round training, and the total statistical value of the integral value in the historical multi-round training. The state description vector is then input into the pre-trained prediction model, and the stage parameter set for the next training cycle is output. The instruction delivery module is used to send the generated instructions to the training terminal for execution via the communication interface.

Images