Internet of things based digital twin rehabilitation robot state monitoring and control system

Abstract

The application belongs to the technical field of digital twinning, and discloses a digital twinning rehabilitation robot state monitoring and control system based on an Internet of Things; multi-dimensional state data such as joint angles, torque outputs, electromyographic signals and environmental temperatures of a rehabilitation robot are acquired in real time through an Internet of Things sensor network, a digital twinning model is constructed, and hierarchical updating is carried out based on an adaptive updating frequency adjustment mechanism. The system dynamically adjusts the updating cycle of a high-frequency updating layer and a low-frequency updating layer according to the state parameter change rate and the deviation trend of the predicted response vector, and realizes accurate synchronization of the digital twinning model and the physical system. A state prediction module predicts future response behavior, an error compensation module generates an error compensation vector to adjust model updating and control instructions, and a control module generates optimized control instructions that meet multiple constraints. The application significantly improves the control accuracy and safety of the rehabilitation robot, supports personalized and accurate rehabilitation, and can effectively respond to rapid changes and irregular movements.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital twinning, more particularly, the present application relates to a digital twinning rehabilitation robot state monitoring and control system based on an internet of things. BACKGROUND

[0002] As a product of the combination of modern rehabilitation medicine and advanced engineering technology, rehabilitation robots have played an increasingly important role in the fields of neurological rehabilitation, motor function recovery and daily life ability training in recent years.

[0003] However, the existing rehabilitation robot system still has problems in the application of digital twinning technology, that is, the digital twinning model update frequency and the physical system response speed do not match, resulting in serious prediction control deviation. The traditional system usually adopts a fixed frequency update strategy, which cannot flexibly respond to the complex dynamic changes in the rehabilitation scene. For example, when the patient performs upper limb rehabilitation training, the physical system response speed will suddenly change from slow control movement to rapid stretching movement, while the digital twinning model is still updated at a fixed frequency, causing digital-physical synchronization lag. This lag is particularly evident in high-intensity rehabilitation training, with a model prediction value and actual state deviation of up to 15%-30%, causing the control command to deviate from the actual needs of the patient. More seriously, in the presence of irregular jitter in the rehabilitation scene, fixed frequency updates cannot capture such high frequency and random state changes, resulting in a control system "blind area". At the same time, in the relatively static phase (such as isometric contraction training), the system still maintains high frequency updates, not only wasting computing resources, but also possibly introducing unnecessary control fluctuations by mistaking transmission noise as effective signal changes. Although the existing compensation method such as Kalman filtering can partially alleviate the problem, the linear assumption-based method is difficult to cope with the highly nonlinear and time-varying characteristics in the rehabilitation scene. In addition, multi-dimensional state data (joint angle, torque, electromyographic signal) have different variation characteristics and sampling requirements, and a uniform update frequency cannot take into account the characteristics of each dimension, resulting in either insufficient electromyographic signal sampling or over-sampling of joint angle resources. This mismatch between update frequency and response speed seriously restricts the control accuracy, safety and rehabilitation effect of the rehabilitation robot, which is particularly detrimental to neurological rehabilitation that requires accurate force feedback, and may even cause safety risks during high-speed motion training.

[0004] In view of this, the present application proposes a digital twinning rehabilitation robot state monitoring and control system based on an internet of things to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a digital twinning rehabilitation robot state monitoring and control system based on an internet of things, comprising:

[0006] An acquisition module is configured to acquire multi-dimensional state data of a physical system of a rehabilitation robot in real time through an Internet of Things (IoT) sensor network, the multi-dimensional state data including joint angle, torque output, electromyography (EMG) signal, and environmental temperature.

[0007] A digital twin construction module is configured to construct a digital twin model of the rehabilitation robot according to the multi-dimensional state data, and perform hierarchical updating on the digital twin model based on an adaptive update frequency adjustment mechanism, the hierarchical updating including a high-frequency update layer and a low-frequency update layer.

[0008] A state prediction module is configured to predict a response behavior of the physical system of the rehabilitation robot in a future preset time period according to a current state and historical state data of the digital twin model, and obtain a predicted response vector.

[0009] An error compensation module is configured to generate an error compensation vector according to a dynamic error between the predicted response vector and the multi-dimensional state data acquired in real time, and adjust an update frequency and a control instruction of the digital twin model based on the error compensation vector.

[0010] A control module is configured to generate an optimized control instruction according to the adjusted digital twin model, and issue the optimized control instruction to the physical system of the rehabilitation robot through an IoT communication network, so as to realize state monitoring and real-time control.

[0011] Preferably, the adaptive update frequency adjustment mechanism is implemented by the following method:

[0012] A change rate vector of the multi-dimensional state data in a preset analysis period is calculated, the change rate vector being composed of joint angle change rate, torque output change rate, EMG signal change rate, and environmental temperature change rate.

[0013] An update requirement level of the digital twin model is determined according to a comparison result of a module length of the change rate vector and a preset change threshold, the update requirement level including a high requirement level and a low requirement level.

[0014] When the update requirement level is the high requirement level, an update cycle proportion of the high-frequency update layer is increased, and an update time interval of the high-frequency update layer is shortened; when the update requirement level is the low requirement level, an update cycle proportion of the low-frequency update layer is increased, and an update time interval of the low-frequency update layer is lengthened.

[0015] A deviation trend value between the predicted response vector and the multi-dimensional state data acquired in real time is calculated, the deviation trend value being obtained by weighted average of deviation vectors at continuous multiple time points.

[0016] According to the comparison result of the deviation trend value and the preset deviation threshold, the update time interval of the high-frequency update layer and the low-frequency update layer is further fine-tuned to realize dynamic matching of the digital twin model update frequency and the physical system response speed.

[0017] Preferably, the error compensation vector generation method comprises:

[0018] An error vector between the predicted response vector and the real-time acquired multi-dimensional state data is obtained, the error vector comprising joint angle deviation, torque output deviation, electromyographic signal deviation and environmental temperature deviation;

[0019] The error vector is time-decomposed to obtain high-frequency and low-frequency components of the error vector;

[0020] The error compensation vector is generated according to the weight proportion of the high-frequency component and the low-frequency component, wherein the weight proportion is dynamically adjusted by analyzing the periodic characteristics of the change rate vector of the multi-dimensional state data;

[0021] The error compensation vector is superimposed on the predicted response vector to obtain a corrected predicted response vector, and the update frequency and control instruction of the digital twin model are adjusted based on the corrected predicted response vector.

[0022] Preferably, the hierarchical update of the digital twin model comprises:

[0023] The digital twin model is divided into a high-frequency update layer and a low-frequency update layer;

[0024] The real-time streaming update strategy is adopted for the high-frequency update layer, the multi-dimensional state data is preprocessed by an edge computing node of an Internet of Things sensor network, and the preprocessing result is directly mapped to the high-frequency update layer;

[0025] The batch update strategy is adopted for the low-frequency update layer, the multi-dimensional state data is globally analyzed by an Internet of Things cloud server, and the analysis result is periodically mapped to the low-frequency update layer;

[0026] According to the update results of the high-frequency update layer and the low-frequency update layer, the global state of the digital twin model is dynamically adjusted to realize the synchronization of the digital twin model and the physical system.

[0027] Preferably, the method for generating the optimized control instruction comprises:

[0028] According to the adjusted digital twin model, a target state vector of the rehabilitation robot physical system in a future preset time period is extracted, the target state vector comprising a target joint angle, a target torque output and a target electromyographic signal;

[0029] calculating a state difference vector between the target state vector and the current multi-dimensional state data, and generating an initial control instruction according to a length and a direction of the state difference vector;

[0030] performing constraint optimization on the initial control instruction, the constraint optimization including joint angle range constraints, torque output safety constraints, and electromyographic signal stability constraints, to obtain the optimized control instruction;

[0031] issuing the optimized control instruction to an actuator of a rehabilitation robot physical system through an Internet of Things communication network, and monitoring execution results in real time to feed back to the digital twin model.

[0032] Preferably, the method for calculating the rate of change vector comprises:

[0033] performing time series differentiation on each state parameter of the multi-dimensional state data within the preset analysis period to obtain a rate of change sequence of the corresponding state parameter;

[0034] performing sliding window smoothing processing on the rate of change sequence to obtain a smoothed rate of change sequence, wherein the window length of the sliding window is adaptively adjusted according to the sampling frequency of the multi-dimensional state data;

[0035] taking the smoothed rate of change value of each state parameter in the smoothed rate of change sequence as a component to form the rate of change vector;

[0036] calculating the length of the rate of change vector, and determining the update requirement level according to a comparison result of the length and the preset change threshold.

[0037] Preferably, the time series decomposition method of the deviation vector comprises:

[0038] performing wavelet transform on the deviation vector to obtain a high-frequency component and a low-frequency component of the deviation vector;

[0039] calculating an energy ratio of the high-frequency component and the low-frequency component, and dynamically adjusting a weight ratio of the high-frequency component and the low-frequency component according to the energy ratio;

[0040] weighting and fusing the high-frequency component and the low-frequency component according to the weight ratio to obtain the error compensation vector.

[0041] Preferably, the implementation method of the real-time streaming update strategy comprises:

[0042] performing local feature extraction on the multi-dimensional state data through an edge computing node of an Internet of Things sensor network, the local features including an instantaneous rate of change of joint angle and a peak value feature of torque output;

[0043] mapping the local features to the high-frequency update layer, and dynamically adjusting an update time interval of the high-frequency update layer according to a variation trend of the local features;

[0044] performing consistency verification on an update result of the high-frequency update layer, the consistency verification including verifying whether a state parameter of the high-frequency update layer satisfies a kinematic constraint of a physical system;

[0045] when the consistency verification fails, performing global correction on the high-frequency update layer through the Internet of Things cloud server to ensure accuracy of the digital twin model.

[0046] Preferably, the system further includes a distributed data collaboration module for:

[0047] distributing the multi-dimensional state data to a plurality of distributed computing nodes through an Internet of Things communication network, the distributed computing nodes including edge computing nodes and cloud computing nodes;

[0048] performing real-time preprocessing on the multi-dimensional state data on the edge computing nodes to obtain local state feature vectors, and uploading the local state feature vectors to the cloud computing nodes;

[0049] performing global fusion on the local state feature vectors on the cloud computing nodes to obtain global state feature vectors, and feeding back the global state feature vectors to the digital twin model;

[0050] dynamically adjusting allocation of computing resources of the edge computing nodes and the cloud computing nodes according to a difference between the global state feature vectors and the local state feature vectors to improve update efficiency of the digital twin model.

[0051] Preferably, the implementation method of the constraint optimization includes:

[0052] constructing a constraint optimization objective function, the objective function taking minimization of a length of the state difference vector as an optimization objective, and taking the joint angle range constraint, the torque output safety constraint, and the electromyographic signal stability constraint as constraint conditions;

[0053] solving the constraint optimization objective function by using a Lagrange multiplier method to obtain a candidate solution set of the optimization control instruction;

[0054] performing stability evaluation on the candidate solution set, the stability evaluation including calculating an execution stability index of the control instruction corresponding to the candidate solution set in the digital twin model;

[0055] Select the candidate solution with the highest stability index as the optimal control instruction, and issue it to the rehabilitation robot physical system through the Internet of Things communication network.

[0056] The technical effects and advantages of the digital twin rehabilitation robot state monitoring and control system based on the Internet of Things are as follows:

[0057] The present application realizes precise synchronization of model updating and physical changes during rehabilitation, greatly improves control accuracy, and effectively eliminates the control lag and misalignment problems of traditional systems in rapid state changes. In the rehabilitation of irregularly dithering patients, the system can accurately capture and respond to these high-frequency random changes, filling the "blind area" of traditional control, significantly enhancing the safety of rehabilitation training, and effectively reducing the risk of accidental injury. The system intelligently allocates computing resources, reducing unnecessary high-frequency calculations when the patient's state is relatively stable, greatly improving computing efficiency, reducing energy consumption, and prolonging device endurance time, making portable rehabilitation training possible. The present application can dynamically adjust the training intensity and mode according to the real-time state of the patient, improving rehabilitation efficiency and patient training compliance. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 The present application provides a digital twin rehabilitation robot state monitoring and control system based on the Internet of Things. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0060] The present application provides a digital twin rehabilitation robot state monitoring and control system based on the Internet of Things. The execution subject of the digital twin rehabilitation robot state monitoring and control system based on the Internet of Things includes but is not limited to the rehabilitation robot control platform, the Internet of Things edge computing node, the Internet of Things cloud server, the data analysis engine, the intelligent control decision system, etc. which can be regarded as the general computing node of the present application, and the data processing platform includes but is not limited to at least one of the multi-dimensional state data acquisition system, the digital twin model construction system, the state prediction and error compensation system.

[0061] Please refer to Figure 1 The present application provides a digital twin rehabilitation robot state monitoring and control system based on the Internet of Things, which includes an acquisition module, a digital twin construction module, a state prediction module, an error compensation module and a control module.

[0062] The application realizes accurate perception of the state of the rehabilitation robot by real-time acquisition of multi-dimensional state data of the physical system of the rehabilitation robot through the Internet of Things sensor network, constructs a digital twin model of the rehabilitation robot to provide a virtual mapping basis for subsequent control, performs hierarchical updating based on an adaptive update frequency adjustment mechanism to keep the model always synchronized with the physical system, predicts the response behavior of the physical system of the rehabilitation robot in a future preset time period to improve the foresight of control, considers the coupling relationship between multi-dimensional states in the prediction response vector to ensure the systematicness of prediction, generates an error compensation vector through dynamic error analysis to solve the deviation problem between the model and the actual system, time decomposition of the error compensation vector enhances the adaptability of the system to disturbances of different time scales, adjustment of the update frequency and control instructions of the digital twin model based on the error compensation vector optimizes the system resource utilization efficiency, generation of optimized control instructions according to the adjusted digital twin model improves the accuracy of control, and constraint processing of the optimized control instructions ensures the safety and reliability of the operation of the rehabilitation robot.

[0063] In the embodiment of the application, the detailed implementation steps of the digital twin rehabilitation robot state monitoring and control system based on the Internet of Things include:

[0064] Firstly, the acquisition module acquires multi-dimensional state data of the physical system of the rehabilitation robot in real time through the Internet of Things sensor network, and the multi-dimensional state data includes joint angle, torque output, electromyographic signal and environmental temperature; then the digital twin construction module constructs a digital twin model of the rehabilitation robot according to the multi-dimensional state data, and performs hierarchical updating on the digital twin model based on an adaptive update frequency adjustment mechanism, which includes a high-frequency updating layer and a low-frequency updating layer; next, the state prediction module predicts the response behavior of the physical system of the rehabilitation robot in a future preset time period according to the current state and historical state data of the digital twin model, and obtains a prediction response vector; then the error compensation module generates an error compensation vector according to the dynamic error between the prediction response vector and the real-time acquired multi-dimensional state data, and adjusts the update frequency and control instructions of the digital twin model based on the error compensation vector; finally, the control module generates optimized control instructions according to the adjusted digital twin model, and issues them to the physical system of the rehabilitation robot through the Internet of Things communication network to realize state monitoring and real-time control; wherein the adaptive update frequency adjustment mechanism dynamically adjusts the update period of the high-frequency updating layer and the low-frequency updating layer by analyzing the change rate of the multi-dimensional state data and the deviation trend of the prediction response vector, to solve the problem of mismatch between the update frequency of the digital twin model and the response speed of the physical system.

[0065] In this embodiment, the acquisition module constructs a comprehensive sensing layer through an Internet of Things sensor network composed of angle sensors, torque sensors deployed at each joint of the rehabilitation robot, electromyographic signal sensors in contact with the patient's muscles, and environmental temperature sensors. All sensors use low-power wireless communication protocols (such as Bluetooth 5.0 BLE, ZigBee, or Wi-Fi 6) to achieve efficient data transmission. Each sensor node integrates a local preprocessing algorithm that can perform preliminary filtering, denoising, and compression on raw data, effectively reducing transmission bandwidth requirements. The acquisition module collects multi-dimensional state data at a pre-set sampling frequency (such as 100 Hz for joint angle, 50 Hz for torque output, 1000 Hz for electromyographic signal, and 1 Hz for environmental temperature). Data fusion and timestamp synchronization are performed through edge computing nodes to ensure consistency in timing across dimensions. The acquisition module also includes an anomaly detection component that can quickly identify sensor failures or data anomalies and take appropriate emergency handling strategies (such as data estimation, redundant sensor switching, etc.) to ensure data stream continuity and reliability.

[0066] The digital twin construction module constructs a digital twin model of the rehabilitation robot based on multi-dimensional state data using a hybrid modeling method that combines deep learning and physical models. The model includes multiple sub-models such as geometric models, kinematic models, dynamic models, and thermodynamic models, which comprehensively represent the physical characteristics of the rehabilitation robot. The digital twin model uses a hierarchical structure to divide state parameters with different change frequencies into high-frequency update layers and low-frequency update layers. The high-frequency update layer mainly includes rapidly changing state parameters such as joint angles and torque outputs, and uses a real-time streaming update strategy. The low-frequency update layer mainly includes slowly changing state parameters such as environmental temperature, and uses a batch update strategy. Based on an adaptive update frequency adjustment mechanism, the system can dynamically adjust the update frequency based on the change rate and prediction bias of the state parameters, achieving efficient allocation of computing resources.

[0067] The state prediction module uses a hybrid prediction method combining long short-term memory networks (LSTM) and physical constraints based on the constructed digital twin model to analyze historical state data and the current state of the rehabilitation robot and predict the system's response behavior within a pre-set time period (such as 100ms-1s). The prediction process considers the coupling relationships between multi-dimensional states and physical constraints such as joint kinematic constraints and torque balance conditions, ensuring that the prediction results conform to physical laws. The prediction results form a prediction response vector, which includes the predicted value sequences of joint angles, torque outputs, electromyographic signals, and environmental temperatures at future time points.

[0068] The error compensation module calculates the dynamic error by comparing the predicted response vector with the actually obtained multi-dimensional state data, and extracts the high-frequency component and the low-frequency component of the error through time decomposition. Based on these components, an error compensation vector is generated to adjust the update frequency of the digital twin model and the control instructions. The error compensation mechanism can effectively reduce the influence of model prediction deviation on control performance and improve the robustness of the system.

[0069] Based on the adjusted digital twin model, the control module generates optimized control instructions that meet multiple constraint conditions using model predictive control (MPC) combined with reinforcement learning. These constraints include joint angle range constraints, torque output safety constraints, and electromyographic signal stability constraints, ensuring the safety and effectiveness of the control instructions. The optimized control instructions are issued to the actuators of the physical system of the rehabilitation robot through the Internet of Things communication network, achieving precise state control. The entire control process forms a closed loop, and the system continuously learns and optimizes to continuously improve control performance.

[0070] In the embodiment of the application, the implementation method of the adaptive update frequency adjustment mechanism includes: calculating the change rate vector of the multi-dimensional state data within a preset analysis period, the change rate vector being composed of joint angle change rate, torque output change rate, electromyographic signal change rate, and environmental temperature change rate; determining the update requirement level of the digital twin model according to the comparison result of the module length of the change rate vector and the preset change threshold, the update requirement level including a high requirement level and a low requirement level; when the update requirement level is the high requirement level, increasing the update cycle proportion of the high-frequency update layer and shortening the update time interval of the high-frequency update layer; when the update requirement level is the low requirement level, increasing the update cycle proportion of the low-frequency update layer and extending the update time interval of the low-frequency update layer; calculating the deviation trend value between the predicted response vector and the real-time obtained multi-dimensional state data, the deviation trend value being obtained by weighted average of deviation vectors at consecutive multiple time points; further fine-tuning the update time interval of the high-frequency update layer and the low-frequency update layer according to the comparison result of the deviation trend value and the preset deviation threshold, to realize dynamic matching of the update frequency of the digital twin model and the response speed of the physical system.

[0071] In this embodiment, first, the multi-dimensional state data within a preset analysis period (such as 100ms-1s) is obtained from the Internet of Things sensor network, including time series data of joint angle, torque output, electromyographic signal, and environmental temperature. The system calculates the time derivative of the state data in each dimension to obtain the joint angle change rate (° / s), torque output change rate (Nm / s), electromyographic signal change rate (mV / s), and environmental temperature change rate (°C / s), forming a change rate vector. To improve calculation stability, the system uses finite difference method or Kalman filter method for change rate estimation, effectively suppressing noise influence.

[0072] The system compares the magnitude of the rate of change vector (i.e., the square root of the sum of the squares of all rate of change components) with a preset change threshold to determine the update requirement level of the digital twin model. The preset change threshold is set based on historical data statistics and expert knowledge, such as the typical range of the rate of change of each state parameter during normal rehabilitation training. When the magnitude of the rate of change vector exceeds the high requirement threshold (e.g., 1.5 times the standard rate of change), the system determines a high requirement level; when the magnitude of the rate of change vector is below the low requirement threshold (e.g., 0.5 times the standard rate of change), the system determines a low requirement level; and when it is between the two, the system maintains the current update frequency unchanged.

[0073] For the high requirement level, the system increases the update cycle proportion of the high-frequency update layer (e.g., from 50% to 70%), which means more computing resources are allocated to the high-frequency update layer, and the update time interval of the high-frequency update layer is shortened (e.g., from 20ms to 10ms), ensuring timely tracking of rapidly changing states. For the low requirement level, the system increases the update cycle proportion of the low-frequency update layer (e.g., from 50% to 70%), which means more computing resources are allocated to the low-frequency update layer, and the update time interval of the high-frequency update layer is extended (e.g., from 20ms to 40ms), avoiding unnecessary frequent updates and improving system efficiency.

[0074] The system evaluates the time trend of prediction accuracy by calculating the deviation vector sequence between the predicted response vector and the actual multi-dimensional state data obtained. The deviation vectors of consecutive time points (e.g., the last 10 sampling points) are exponentially weighted and averaged, giving higher weight to recent deviations, to obtain the deviation trend value. The deviation trend value reflects the dynamic change trend of prediction accuracy and is an important reference for adjusting the update frequency.

[0075] According to the comparison result of the deviation trend value and the preset deviation threshold, the system further fine-tunes the update time interval of the high-frequency update layer and the low-frequency update layer. When the deviation trend value rises (prediction accuracy decreases), the system appropriately shortens the update time interval and increases the update frequency; when the deviation trend value decreases (prediction accuracy improves), the system appropriately extends the update time interval and reduces the update frequency. The fine-tuning process uses a proportional-integral control algorithm to ensure the smoothness and stability of the update frequency adjustment.

[0076] Through the above adaptive update frequency adjustment mechanism, the system realizes the dynamic matching of the digital twin model update frequency and the physical system response speed, ensuring the real-time and accuracy of the model, and optimizing the allocation efficiency of computing resources.

[0077] In the embodiment of the present application, the method for generating an error compensation vector includes: obtaining a deviation vector between a predicted response vector and multi-dimensional state data obtained in real time, the deviation vector including joint angle deviation, torque output deviation, electromyographic signal deviation, and environmental temperature deviation; performing time series decomposition on the deviation vector to obtain a high-frequency component and a low-frequency component of the deviation vector, wherein the high-frequency component is used to represent instantaneous fluctuations in the response speed of the physical system, and the low-frequency component is used to represent long-term trends in the response speed of the physical system; generating an error compensation vector according to the weight proportion of the high-frequency component and the low-frequency component, wherein the weight proportion is dynamically adjusted by analyzing the periodic characteristics of the change rate vector of the multi-dimensional state data; and superimposing the error compensation vector on the predicted response vector to obtain a corrected predicted response vector, and adjusting the update frequency and control instructions of the digital twin model based on the corrected predicted response vector.

[0078] In the embodiment, first, the difference between the predicted response vector and the multi-dimensional state data obtained in real time is calculated to form a deviation vector. The system synchronously records the prediction time, the actual measurement time, and the time when the deviation is calculated to ensure the accuracy of time alignment. The deviation vector includes components such as joint angle deviation (°), torque output deviation (Nm), electromyographic signal deviation (mV), and environmental temperature deviation (°C). Due to the different physical units and orders of magnitude of each component, the system uses normalization processing to make each component comparable.

[0079] The system uses wavelet transform method to perform time series decomposition on the normalized deviation vector to extract high-frequency components and low-frequency components. Appropriate wavelet basis functions (such as Daubechies wavelet, Morlet wavelet, etc.) and decomposition levels are selected to ensure that the decomposition results can accurately represent the deviation characteristics of different time scales. The high-frequency component mainly reflects the instantaneous fluctuations in the response of the physical system, such as rapid changes caused by patient muscle tremors, environmental disturbances, etc.; the low-frequency component mainly reflects the long-term trends in the response of the physical system, such as slow changes caused by fatigue accumulation, system parameter drift, etc.

[0080] The system analyzes the periodic characteristics of the change rate vector of the multi-dimensional state data, such as identifying the periodic pattern through Fourier transform or autocorrelation analysis, to dynamically adjust the weight proportion of the high-frequency component and the low-frequency component. When the state data presents obvious periodic changes, the system increases the weight of the low-frequency component to better track the basic trend of the system; when the state data presents non-periodic rapid changes, the system increases the weight of the high-frequency component to respond to the instantaneous changes of the system more quickly. The weight distribution is realized by a fuzzy logic controller, which adaptively adjusts the weight value according to the periodic intensity index.

[0081] According to the adjusted weight ratio, the system fuses the high-frequency component and the low-frequency component to generate an error compensation vector. The error compensation vector retains the key features of the original deviation, and at the same time optimizes the compensation effect through weight adjustment. The system superimposes the error compensation vector to the predicted response vector to obtain a modified predicted response vector, which more accurately represents the future state of the physical system.

[0082] Based on the modified predicted response vector, the system further adjusts the update frequency and control instruction of the digital twin model. When the modified predicted response vector shows that the system state changes accelerate, the update frequency is increased; when the modified predicted response vector shows that the system state changes slow down, the update frequency is reduced. At the same time, the modified predicted response vector is input into the control algorithm to generate more accurate control instructions, improving the control performance. The entire error compensation process forms a closed loop, continuously optimizing the prediction accuracy and control effect.

[0083] In the embodiment of the application, the digital twin model is updated in layers, including: dividing the digital twin model into a high-frequency update layer and a low-frequency update layer, wherein the high-frequency update layer is used to update state parameters in the multi-dimensional state data whose change rate is greater than a preset change rate threshold, and the low-frequency update layer is used to update state parameters in the multi-dimensional state data whose change rate is less than or equal to the preset change rate threshold; the real-time streaming update strategy is adopted for the high-frequency update layer, the multi-dimensional state data is preprocessed by the edge computing node of the Internet of Things sensor network, and the preprocessing result is directly mapped to the high-frequency update layer; the batch update strategy is adopted for the low-frequency update layer, the multi-dimensional state data is globally analyzed by the Internet of Things cloud server, and the analysis result is periodically mapped to the low-frequency update layer; according to the update results of the high-frequency update layer and the low-frequency update layer, the global state of the digital twin model is dynamically adjusted to realize the synchronization of the digital twin model and the physical system.

[0084] In this embodiment, first, according to the change characteristics of the state parameters, the digital twin model is divided into a high-frequency update layer and a low-frequency update layer. The system sets a suitable preset change rate threshold for each parameter by long-term monitoring of the change rate statistical characteristics of each state parameter. Typically, the change rate of joint angle and torque output is usually large, which is divided into the high-frequency update layer; and the change rate of the environment temperature is usually small, which is divided into the low-frequency update layer. The change rate of the electromyographic signal has a bimodal characteristic, and the system dynamically determines its belonging level according to its current change rate.

[0085] The system adopts a real-time streaming update strategy for the high-frequency update layer, fully utilizing the edge computing capabilities of the Internet of Things. The edge computing nodes are equipped with dedicated data processing chips (such as FPGA or edge AI accelerators) for real-time preprocessing of sensor data, including digital filtering, outlier detection, data compression, and other operations. The preprocessing results are directly mapped to the high-frequency update layer of the digital twin model through high-speed communication links (such as 5G or dedicated low-latency networks), achieving millisecond-level state synchronization. The system sets independent update frequencies for each high-frequency update parameter, differentiating processing according to their importance and variation characteristics, and optimizing the allocation of computing resources.

[0086] The system adopts a batch update strategy for the low-frequency update layer, utilizing the powerful computing capabilities of cloud servers for in-depth analysis. The cloud server collects multi-dimensional state data over a period of time (such as 10 seconds or 1 minute) and performs trend analysis, pattern recognition, and correlation mining, among other advanced analysis operations. The analysis results are periodically mapped to the low-frequency update layer of the digital twin model at a lower frequency (such as 1 Hz or 0.1 Hz). Batch updating not only reduces communication overhead but also allows the execution of more computationally complex analysis algorithms, providing a more in-depth understanding of the system.

[0087] Based on the update results of the high-frequency update layer and the low-frequency update layer, the system dynamically adjusts the global state of the digital twin model. The real-time data of the high-frequency update layer provides immediate state information, while the analysis results of the low-frequency update layer provide trend and pattern information. The system uses a multi-scale fusion algorithm to integrate information from both levels, forming a complete representation of the system state. Global state adjustment uses an incremental update method, updating only the changed parts to reduce computational redundancy.

[0088] Through this hierarchical update mechanism, the system achieves efficient utilization of computing resources and accurate tracking of state changes, ensuring high synchronization between the digital twin model and the physical system. Compared with traditional unified update strategies, hierarchical updating can significantly reduce computational burden while maintaining or improving model accuracy, making it particularly suitable for edge device application scenarios with limited computing resources.

[0089] In the embodiment of the present application, the method for generating the optimized control instruction comprises: extracting a target state vector of the physical system of the rehabilitation robot in a future preset time period according to the adjusted digital twin model, the target state vector including a target joint angle, a target torque output and a target electromyographic signal; calculating a state difference vector between the target state vector and the current multi-dimensional state data, and generating an initial control instruction according to the length and direction of the state difference vector; performing constraint optimization on the initial control instruction, the constraint optimization including joint angle range constraint, torque output safety constraint and electromyographic signal stability constraint, to obtain the optimized control instruction; and issuing the optimized control instruction to the actuator of the physical system of the rehabilitation robot through the Internet of Things communication network, and monitoring the execution result in real time to feed back to the digital twin model.

[0090] In the embodiment, first, based on the adjusted digital twin model, the system determines the target state that the rehabilitation robot should reach in a future preset time period (such as 0.5 seconds to 5 seconds) according to the rehabilitation training plan or the therapist's instruction. The target state vector includes the target joint angle (representing the expected motion position), the target torque output (representing the expected force assistance or impedance), and the target electromyographic signal (representing the expected muscle activation level). The system dynamically adjusts the target state vector considering the patient's rehabilitation progress, muscle state and fatigue level, to achieve personalized rehabilitation training.

[0091] The system calculates the state difference vector between the target state vector and the current multi-dimensional state data, i.e., the difference between the target state and the current state. The length of the state difference vector represents the overall deviation of the current state from the target state, and the direction represents the relative relationship of the deviation of each state parameter. The system generates an initial control instruction using the model predictive control (MPC) algorithm according to the length and direction of the state difference vector. The MPC algorithm predicts the influence of the control instruction on the system state by solving the optimization problem in a finite time domain, and selects the control sequence that can minimize the state difference.

[0092] The system performs constraint optimization on the initial control instruction to ensure the safety and effectiveness of the control. The joint angle range constraint ensures that the motion does not exceed the patient's range of motion, preventing injury caused by excessive stretching or flexion; the torque output safety constraint limits the size of the torque applied to the patient's joints, preventing discomfort or injury caused by excessive torque; and the electromyographic signal stability constraint ensures smooth changes in muscle activation level during control, avoiding sudden muscle contraction or relaxation. The constraint optimization uses sequential quadratic programming (SQP) or interior point method (IPM) to solve, minimizing control energy consumption and state deviation under the premise of meeting the constraint conditions.

[0093] The control instruction optimized by the constraint becomes the final optimized control instruction, which is issued to the actuators of the physical system of the rehabilitation robot through the Internet of Things communication network. The communication network adopts a deterministic low-delay transmission protocol to ensure timely delivery of the control instruction. The actuators include servo motors, solenoid valves or other execution elements, which adjust the physical state of the rehabilitation robot according to the control instruction. The system monitors the execution result in real time, collects feedback data such as the actual executed joint angle and the actual output torque during the execution process, and feeds these data back to the digital twin model for updating the model state and optimizing subsequent control.

[0094] Through this digital twin-based optimization control method, the system realizes accurate control and safety guarantee of the rehabilitation robot, and improves the effect of rehabilitation training and patient experience. Compared with traditional control methods, this method can better adapt to individual differences and state changes of patients, and provide more personalized and adaptive rehabilitation assistance.

[0095] In the embodiment of the application, the method for calculating the rate of change vector comprises: performing time series difference on each state parameter of the multi-dimensional state data in a preset analysis period to obtain a rate of change sequence corresponding to the state parameter; performing sliding window smoothing processing on the rate of change sequence to obtain a smoothed rate of change sequence, wherein the window length of the sliding window is adaptively adjusted according to the sampling frequency of the multi-dimensional state data; taking the smoothed rate of change value of each state parameter in the smoothed rate of change sequence as a component to form the rate of change vector; calculating the module length of the rate of change vector, and determining the update requirement level according to the comparison result of the module length and the preset change threshold.

[0096] In this embodiment, the system first performs time series difference processing on the multi-dimensional state data collected in the preset analysis period (such as 100 ms to 1 s). For each state parameter (joint angle, torque output, electromyographic signal and environmental temperature), the system calculates the difference between adjacent time points and divides by the time interval to obtain the instantaneous rate of change. To process parameters with different sampling frequencies, the system uses an interpolation method to unify the time reference, ensuring the accuracy of the difference calculation. For each state parameter, the system generates a complete rate of change time sequence to describe the change speed of the parameter over time.

[0097] The system performs sliding window smoothing processing on the calculated rate of change sequence to reduce the influence of random noise and measurement error. The length of the sliding window is adaptively adjusted according to the sampling frequency of the state parameter. The higher the sampling frequency, the longer the window length, ensuring the consistency of the smoothing effect. For example, for an electromyographic signal with a sampling frequency of 1000Hz, a window length of 50-100 sampling points may be used; while for an environmental temperature with a sampling frequency of 1Hz, a window length of only 3-5 sampling points may be used. The smoothing processing uses a weighted moving average method, with higher weights given to samples near the center of the window and lower weights given to samples at the edges of the window, forming a smoothed rate of change sequence.

[0098] The system takes the latest smoothed rate of change value of each state parameter in the smoothed rate of change sequence as a component to form a rate of change vector. The rate of change vector is a multi-dimensional vector with the same number of dimensions as the number of state parameters, and each component represents the current rate of change of the corresponding state parameter. Due to the different physical units and orders of magnitude of different state parameters, the system performs normalization processing to make the components comparable. The normalization method is based on historical data statistics to map the rate of change of each parameter to a unified dimensionless scale.

[0099] The system calculates the length of the rate of change vector, which is the square root of the sum of the squares of the components. The length is a scalar value representing the speed of the overall system state change. The system compares the calculated length with a pre-set change threshold to determine the update requirement level of the digital twin model. The pre-set change threshold includes a high demand threshold and a low demand threshold, which are set based on system performance requirements and resource constraints. When the length is greater than the high demand threshold, the system determines that it is a high demand level and needs more frequent updates; when the length is less than the low demand threshold, the system determines that it is a low demand level and can reduce the update frequency; when the length is between the two thresholds, the system maintains the current update frequency unchanged.

[0100] Through this change rate-based update requirement evaluation method, the system can adaptively adjust the update strategy of the digital twin model according to the dynamic characteristics of the state change, achieving efficient utilization of computing resources and dynamic balance of model accuracy. Compared with fixed update frequency, this method significantly improves the adaptability and resource utilization efficiency of the system.

[0101] In the embodiment of the application, the time sequence decomposition method of the deviation vector comprises: performing wavelet transform on the deviation vector to obtain a high-frequency component and a low-frequency component of the deviation vector, wherein the high-frequency component corresponds to a high-frequency subband of the wavelet transform, and the low-frequency component corresponds to a low-frequency subband of the wavelet transform; calculating an energy ratio of the high-frequency component and the low-frequency component, and dynamically adjusting a weight ratio of the high-frequency component and the low-frequency component according to the energy ratio; and performing weighted fusion on the high-frequency component and the low-frequency component according to the weight ratio to obtain an error compensation vector.

[0102] In this embodiment, the system first performs wavelet transform on the calculated deviation vector (the difference between the predicted response vector and the actual multi-dimensional state data). The system selects a wavelet basis function suitable for time series data analysis, such as Daubechies wavelet, Haar wavelet or Symlet wavelet, and determines the optimal selection according to the data characteristics. Wavelet transform performs multi-resolution decomposition on the deviation vector in the time-frequency domain, generating a series of wavelet coefficients. These coefficients correspond to signal components of different time scales and frequency ranges.

[0103] The system extracts high-frequency components and low-frequency components from the wavelet coefficients. High-frequency components correspond to high-frequency subbands (detail coefficients) of wavelet transform, mainly containing fast-changing components in the deviation vector, such as noise, transient interference and short-term fluctuations. Low-frequency components correspond to low-frequency subbands (approximation coefficients) of wavelet transform, mainly containing slow-changing components in the deviation vector, such as trends, periodic patterns and long-term drifts. By adjusting the decomposition level, the system can control the degree of separation of high-frequency and low-frequency components, adapting to the needs of different application scenarios.

[0104] The system calculates the energy ratio of high-frequency components and low-frequency components, quantifying the relative importance of the two components in the deviation vector. Energy calculation is based on the sum of squares of wavelet coefficients, calculating the energy values of high-frequency subbands and low-frequency subbands respectively, and then calculating the ratio of the two. The energy ratio reflects the relative strength of transient fluctuations and long-term trends in the deviation vector, and is an important basis for weight adjustment.

[0105] The system dynamically adjusts the weight ratio of high-frequency components and low-frequency components according to the energy ratio. When the energy ratio is large (high-frequency component energy dominates), it indicates that the deviation is mainly due to transient fluctuations, and the system increases the weight of high-frequency components; when the energy ratio is small (low-frequency component energy dominates), it indicates that the deviation is mainly due to long-term trends, and the system increases the weight of low-frequency components. Weight adjustment uses a nonlinear mapping function to map the energy ratio to a weight value in the [0,1] interval, ensuring the smoothness and stability of the adjustment.

[0106] The system generates an error compensation vector by weighting and fusing high-frequency components and low-frequency components according to the adjusted weight ratio. In the fusion process, the system appropriately limits the amplitude of high-frequency components to prevent system oscillation caused by excessive compensation; and performs trend prediction on low-frequency components to improve the forward-looking nature of long-term compensation. The final error compensation vector can respond to both fast changes and long-term trends in the system, providing comprehensive and balanced error compensation.

[0107] Through this wavelet transform-based time series decomposition method, the system can accurately identify and separate error components of different time scales, implement targeted compensation strategies, and significantly improve the prediction accuracy of the digital twin model and the robustness of the control system. Compared with traditional time or frequency domain analysis methods, the wavelet transform method provides joint analysis capability in the time-frequency domain, which is more suitable for processing non-stationary and multi-scale error signals.

[0108] In the embodiment of the application, the implementation method of the real-time streaming update strategy comprises: performing local feature extraction on multi-dimensional state data through an edge computing node of an Internet of Things sensor network, the local features including an instantaneous change rate of joint angle and a peak value feature of torque output; mapping the local features to a high-frequency update layer, and dynamically adjusting an update time interval of the high-frequency update layer according to a change trend of the local features; performing consistency checking on an update result of the high-frequency update layer, the consistency checking including verifying whether state parameters of the high-frequency update layer satisfy kinematic constraints of a physical system; when the consistency checking fails, performing global correction on the high-frequency update layer through an Internet of Things cloud server, so as to ensure the accuracy of the digital twin model.

[0109] In the embodiment, first, the system uses the Internet of Things sensor network deployed at key positions of the rehabilitation robot to collect multi-dimensional state data in real time. Each sensor node is connected to a nearby edge computing node, forming a distributed data processing network. The edge computing node is equipped with a special processor (such as an ARM Cortex-M series or a special DSP chip) and has real-time data processing capability.

[0110] The edge computing node extracts local features from the received multi-dimensional state data, focusing on state parameters that change rapidly. For joint angle data, the instantaneous change rate (angular velocity) is calculated to identify the acceleration and deceleration phases of the motion. For torque output data, peak features such as maximum value, minimum value, and peak duration are extracted to reflect the key characteristics of force changes. Feature extraction uses lightweight algorithms such as sliding window analysis and threshold detection to ensure that the processing speed meets real-time requirements.

[0111] The system maps the extracted local features directly to the high-frequency update layer of the digital twin model through a high-speed communication link (such as 5G, Wi-Fi 6, or a special low-latency network). The mapping process uses an incremental update strategy, only transmitting features that have changed significantly, reducing communication overhead. The system dynamically adjusts the update time interval of the high-frequency update layer according to the change trend of the local features. When the features change dramatically (such as sudden increase in joint angle change rate or torque peak), the system shortens the update time interval and increases the update frequency. When the features change smoothly, the system lengthens the update time interval and reduces the update frequency. The adjustment process uses an adaptive control algorithm to ensure that the update frequency matches the state changes.

[0112] The system performs consistency check on the update result of the high-frequency update layer, verifies whether the updated state parameters meet the basic constraints of the physical system. Kinematic constraints include joint motion range limits, joint speed limits, coordination relationships between adjacent joints, etc. The verification process uses geometric and kinematic models to quickly identify state parameters that violate physical laws. When the consistency check passes, the system accepts the update result and continues to the next round of update; when the consistency check fails, the system marks the abnormal parameters and triggers the global correction process.

[0113] When the consistency check of the high-frequency update layer fails, the system sends a correction request to the Internet of Things cloud server through a secure communication channel, including abnormal parameter information and related context data. The cloud server has powerful computing power and global data access authority, and can perform more complex analysis and correction algorithms. The cloud server performs global correction on the abnormal parameters, including historical data comparison, physical model reconstruction, multi-source data fusion, etc. The correction result is returned to the edge computing node through a secure channel, updating the corresponding parameters of the high-frequency update layer to ensure the accuracy of the digital twin model.

[0114] Through this edge-cloud collaborative real-time streaming update strategy, the system realizes fast tracking and accurate characterization of high-frequency changing states, fully utilizing the low-latency advantage of edge computing and combining the high-computing power support of cloud computing, forming an efficient and reliable distributed update mechanism. Compared with traditional centralized update strategies, this method significantly reduces the end-to-end delay, improves the real-time performance and robustness of the system.

[0115] In the embodiment of the application, the Internet of Things-based digital twin rehabilitation robot state monitoring and control system further comprises a distributed data collaboration module, configured to: distribute multi-dimensional state data to a plurality of distributed computing nodes through an Internet of Things communication network, the distributed computing nodes comprising edge computing nodes and cloud computing nodes; perform real-time preprocessing on the multi-dimensional state data on the edge computing nodes to obtain local state feature vectors, and upload the local state feature vectors to the cloud computing nodes; perform global fusion on the local state feature vectors on the cloud computing nodes to obtain global state feature vectors, and feed back the global state feature vectors to the digital twin model; and dynamically adjust the allocation of computing resources of the edge computing nodes and the cloud computing nodes according to the difference between the global state feature vectors and the local state feature vectors, so as to improve the update efficiency of the digital twin model.

[0116] In this embodiment, the distributed data collaboration module first establishes a unified data distribution framework, effectively distributing the multi-dimensional state data collected by the Internet of Things sensor network to each computing node in the distributed computing architecture. The data distribution adopts a publish-subscribe mode, selectively pushing relevant data according to the functions and needs of the computing nodes, avoiding unnecessary data transmission. The system selects appropriate communication protocols according to data type, priority, and timeliness, such as using deterministic low-latency protocols (such as TSN) for high-priority real-time data and using efficient protocols (such as MQTT) for low-priority non-real-time data.

[0117] The edge computing node is deployed on the rehabilitation robot body or the device interacting with the patient, and has low delay and real-time processing capability. The edge node performs real-time preprocessing on the received multi-dimensional state data, including data cleaning (removing noise and outliers), feature extraction (calculating statistical features and time series features), and data compression (dimension reduction and encoding). The preprocessing algorithm is optimized to adapt to the resource constraints of edge computing, such as limited memory and limited computing power. The preprocessing result forms a local state feature vector, which contains key features and local patterns of each state parameter, and the data volume is significantly smaller than the original multi-dimensional state data.

[0118] The edge computing node uploads the local state feature vector to the cloud computing node through a secure channel. The uploading process adopts incremental transmission and priority scheduling mechanism to ensure timely transmission of important features. The cloud computing node is deployed on a high-performance computing platform and has powerful computing and storage resources. The cloud node collects local state feature vectors from multiple edge nodes for global fusion analysis. The fusion methods include feature alignment (time synchronization and space registration), multi-source fusion (weighted average and Kalman filtering), and pattern mining (clustering analysis and association rules). The fusion result forms a global state feature vector, which contains system-level state features and global patterns, providing a more comprehensive system understanding.

[0119] The cloud computing node feeds back the global state feature vector to the digital twin model through a high-speed communication link. The feedback mechanism supports two modes of regular push and event triggering, and can be flexibly selected according to model update needs. The digital twin model updates relevant parameters and structures according to the received global state feature vector, achieving synchronization with the physical system. The update process considers the confidence and importance of features, and uses weighted fusion method to ensure the accuracy and stability of the update.

[0120] The system continuously monitors the difference between the global state feature vector and the local state feature vector, evaluates the effect and efficiency of distributed computing. The difference analysis includes statistical difference (mean square error, KL divergence) and semantic difference (consistency of important features). According to the difference analysis result, the system dynamically adjusts the allocation of computing resources of the edge computing node and the cloud computing node. When the difference is small, increase the processing proportion of the edge node, reduce the dependence on the cloud, and reduce the communication overhead; when the difference is large, increase the processing proportion of the cloud node, and provide more comprehensive analysis support. The resource adjustment adopts a reinforcement learning algorithm, which continuously optimizes the allocation strategy through practice to maximize the overall performance of the system.

[0121] Through this edge-cloud collaborative distributed data processing architecture, the system realizes efficient utilization of computing resources and maximization of data value extraction, significantly improving the update efficiency and accuracy of the digital twin model.

[0122] In the embodiment of the application, the implementation method of constraint optimization comprises: constructing a constraint optimization objective function, the objective function taking the minimization of the modulus of the state difference vector as the optimization objective, and taking the joint angle range constraint, the torque output safety constraint and the electromyographic signal stability constraint as the constraint conditions; solving the constraint optimization objective function by using the Lagrange multiplier method to obtain a candidate solution set of the optimization control instruction; performing stability evaluation on the candidate solution set, the stability evaluation comprising calculating an execution stability index of the control instruction corresponding to the candidate solution set in the digital twin model; selecting the candidate solution with the highest stability index as the optimization control instruction, and issuing the optimization control instruction to the rehabilitation robot physical system through the Internet of Things communication network.

[0123] In the embodiment, first, the system constructs a constraint optimization objective function based on rehabilitation training requirements and patient safety considerations. The core of the objective function is to minimize the modulus of the state difference vector, that is, to make the actual state of the rehabilitation robot as close to the target state as possible. The state difference vector includes the difference between the target joint angle and the current angle, the difference between the target torque output and the current torque, and the difference between the target electromyographic signal and the current electromyographic signal. Each component can be assigned different weights to reflect the importance of different state parameters. The objective function adopts a quadratic form to ensure smoothness and numerical stability.

[0124] The system sets multiple constraint conditions to ensure the safety and effectiveness of control. The joint angle range constraint defines the upper and lower limits of the movement of each joint to prevent excessive stretching or flexing; the torque output safety constraint limits the maximum torque value that can be applied to each joint to prevent excessive force from causing discomfort or injury; and the electromyographic signal stability constraint limits the rate of change of the electromyographic signal to ensure smooth transition of muscle activation levels. The constraint conditions are formalized in the form of inequality or equality mathematical expressions and incorporated into the solution framework of the optimization problem.

[0125] The system solves the constructed constraint optimization problem using the Lagrange multiplier method. First, a Lagrange function is constructed, combining the objective function and the constraint conditions (through the Lagrange multiplier) into a unified function. Then, the KKT (Karush-Kuhn-Tucker) condition is applied to establish a set of necessary condition equations for solving the optimal solution. The system uses iterative algorithms (such as gradient descent, Newton's method, or interior point method) to solve the KKT equation set, obtaining a set of candidate solutions that satisfy the constraint conditions. To deal with complex non-convex constraints, the system uses a multi-start strategy to start optimization from different initial points, increasing the probability of finding a global optimal solution. Finally, a candidate solution set containing multiple local optimal solutions is formed.

[0126] The system conducts a comprehensive stability evaluation of the candidate solution set, with evaluation criteria including the execution stability and robustness of the control instructions. The system uses a digital twin model as a virtual test platform to simulate and test each candidate solution. During the testing process, the system introduces parameter perturbations and external disturbances to simulate uncertainties in the real environment. The system records performance indicators such as state deviation, energy consumption, and overshoot under different perturbation conditions, and comprehensively calculates the execution stability index. The stability index reflects the resistance of the control instructions to perturbations and the convergence of the system state.

[0127] The system selects the solution with the highest stability index from the candidate solution set as the final optimized control instruction. The selection process considers the balance between stability and control performance, prioritizing solutions with better control performance while ensuring sufficient stability. The final optimized control instruction is issued to the physical system of the rehabilitation robot through the Internet of Things communication network. The issuance process includes instruction encryption, integrity verification, and confirmation mechanisms to ensure the secure transmission and accurate execution of the instructions. The system continuously monitors the execution effect of the control instructions and collects feedback data for subsequent control optimization and model updating.

[0128] Through this constraint optimization-based control method, the system achieves precise control and safety assurance of the rehabilitation robot, effectively balancing multiple goals such as rehabilitation effectiveness, patient safety, and system efficiency. Compared with traditional PID control or fuzzy control, this method can systematically handle multi-dimensional constraints and optimization objectives, providing more intelligent and personalized rehabilitation training support.

[0129] The digital twin rehabilitation robot state monitoring and control system based on the Internet of Things provided by the embodiment of the application realizes efficient synchronization of the physical system and the digital model by acquiring multi-dimensional state data in real time through the Internet of Things sensor network, constructing a digital twin model, and using an adaptive update frequency adjustment mechanism for hierarchical updating. The system can accurately predict future response behavior, generate an error compensation vector to adjust the model update and control strategy, and generate safe and effective control instructions through constraint optimization.

[0130] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art will appreciate that modifications, substitutions and changes can be made to the technical solutions described in the foregoing embodiments, or equivalent replacements can be made to some of the technical features. Any modifications, equivalent replacements, improvements, and the like made within the spirit and principle of the present application shall be included in the scope of protection of the present application.

[0131] It should be noted that the preset parameters and the threshold values in the present application are set by those skilled in the art according to actual conditions.

[0132] Although the embodiments of the present application have been shown and described, those skilled in the art can understand that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and the spirit of the present application, and the scope of the present application is defined by the claims and their equivalents.

Claims

1. The digital twin rehabilitation robot state monitoring and control system based on the Internet of Things is characterized by: include: An acquisition module is used to acquire multi-dimensional state data of the physical system of the rehabilitation robot in real time through an Internet of Things sensor network. The multi-dimensional state data includes joint angles, torque output, electromyographic signals, and ambient temperature. A digital twin construction module, configured to construct a digital twin model of the rehabilitation robot based on the multidimensional state data, and perform hierarchical updates on the digital twin model based on an adaptive update frequency adjustment mechanism, wherein the hierarchical updates include a high-frequency update layer and a low-frequency update layer; A state prediction module is used to predict the response behavior of the physical system of the rehabilitation robot within a preset time period in the future based on the current state and historical state data of the digital twin model, and obtain a predicted response vector; an error compensation module, configured to generate an error compensation vector according to a dynamic error between the predicted response vector and the multidimensional state data acquired in real time, and adjust the update frequency and control instructions of the digital twin model based on the error compensation vector; A control module, configured to generate optimized control instructions based on the adjusted digital twin model and send them to the physical system of the rehabilitation robot via an Internet of Things communication network; The method for implementing the adaptive update frequency adjustment mechanism includes: Calculating a change rate vector of the multidimensional state data within a preset analysis period, wherein the change rate vector is composed of a joint angle change rate, a torque output change rate, an electromyographic signal change rate, and an ambient temperature change rate; Determining an update requirement level of the digital twin model based on a comparison result of the modulus of the change rate vector and a preset change threshold, where the update requirement level includes a high requirement level and a low requirement level; When the update demand level is high, the update cycle ratio of the high-frequency update layer is increased, and the update time interval of the high-frequency update layer is shortened; when the update demand level is low, the update cycle ratio of the low-frequency update layer is increased, and the update time interval of the low-frequency update layer is extended; Calculating a deviation trend value between the predicted response vector and the multi-dimensional state data acquired in real time, wherein the deviation trend value is obtained by weighted averaging the deviation vectors at a plurality of consecutive moments; Further fine-tuning the update time intervals of the high-frequency update layer and the low-frequency update layer according to a comparison result of the deviation trend value and a preset deviation threshold; The method for generating the error compensation vector includes: Obtaining a deviation vector between the predicted response vector and the multi-dimensional state data obtained in real time, wherein the deviation vector includes a joint angle deviation, a torque output deviation, an electromyographic signal deviation, and an ambient temperature deviation; Performing time series decomposition on the deviation vector to obtain a high-frequency component and a low-frequency component of the deviation vector; generating the error compensation vector according to a weight ratio of the high-frequency component to the low-frequency component, wherein the weight ratio is dynamically adjusted by analyzing a periodic characteristic of a rate of change vector of the multi-dimensional state data; Superimposing the error compensation vector onto the predicted response vector to obtain a modified predicted response vector, and adjusting the update frequency and control instructions of the digital twin model based on the modified predicted response vector; The hierarchical updating of the digital twin model includes: Dividing the digital twin model into a high-frequency update layer and a low-frequency update layer; A real-time streaming update strategy is adopted for the high-frequency update layer, the multi-dimensional state data is preprocessed by the edge computing nodes of the Internet of Things sensor network, and the preprocessing results are directly mapped to the high-frequency update layer; A batch update strategy is adopted for the low-frequency update layer, a global analysis of the multi-dimensional status data is performed through the IoT cloud server, and the analysis results are periodically mapped to the low-frequency update layer; According to the update results of the high-frequency update layer and the low-frequency update layer, the global state of the digital twin model is dynamically adjusted.

2. The state monitoring and control system of the digital twin rehabilitation robot based on the Internet of Things according to claim 1 is characterized in that: The method for generating the optimization control instruction comprises: Extracting a target state vector of the physical system of the rehabilitation robot within a preset future time period based on the adjusted digital twin model, wherein the target state vector includes a target joint angle, a target torque output, and a target electromyographic signal; Calculating a state difference vector between the target state vector and the current multi-dimensional state data, and generating an initial control instruction according to the module length and direction of the state difference vector; performing constraint optimization on the initial control instruction, wherein the constraint optimization includes joint angle range constraint, torque output safety constraint, and electromyographic signal stability constraint, to obtain the optimized control instruction; The optimization control instructions are sent to the actuator of the physical system of the rehabilitation robot through the Internet of Things communication network, and the execution results are monitored in real time to feed back to the digital twin model.

3. The state monitoring and control system of the digital twin rehabilitation robot based on the Internet of Things according to claim 1 is characterized in that: The method for calculating the change rate vector includes: Performing time series difference on each state parameter of the multidimensional state data within the preset analysis period to obtain a change rate sequence of the corresponding state parameter; Performing sliding window smoothing on the change rate sequence to obtain a smoothed change rate sequence, wherein the window length of the sliding window is adaptively adjusted according to the sampling frequency of the multidimensional state data; The smoothed change rate value of each state parameter in the smoothed change rate sequence is used as a component to form the change rate vector; The modulus of the change rate vector is calculated, and the update requirement level is determined based on a comparison result of the modulus with the preset change threshold.

4. The state monitoring and control system of the digital twin rehabilitation robot based on the Internet of Things according to claim 1 is characterized in that: The time series decomposition method of the deviation vector includes: Performing wavelet transform on the deviation vector to obtain a high-frequency component and a low-frequency component of the deviation vector; Calculating an energy ratio of the high-frequency component to the low-frequency component, and dynamically adjusting a weight ratio of the high-frequency component to the low-frequency component according to the energy ratio; The high-frequency component and the low-frequency component are weightedly fused according to the weight ratio to obtain the error compensation vector.

5. The state monitoring and control system of the digital twin rehabilitation robot based on the Internet of Things according to claim 1 is characterized in that: The method for implementing the real-time streaming update strategy includes: Extracting local features of the multi-dimensional state data through an edge computing node of an Internet of Things sensor network, wherein the local features include the instantaneous rate of change of the joint angle and the peak characteristics of the torque output; Mapping the local features to the high-frequency update layer, and dynamically adjusting the update time interval of the high-frequency update layer according to the change trend of the local features; Performing a consistency check on an update result of the high-frequency update layer, wherein the consistency check includes verifying whether state parameters of the high-frequency update layer satisfy kinematic constraints of a physical system; When the consistency check fails, the high-frequency update layer is globally corrected by the IoT cloud server.

6. The state monitoring and control system of the digital twin rehabilitation robot based on the Internet of Things according to claim 1 is characterized in that: The system also includes a distributed data collaboration module for: Distributing the multidimensional state data to a plurality of distributed computing nodes via an Internet of Things communication network, wherein the distributed computing nodes include edge computing nodes and cloud computing nodes; Preprocessing the multidimensional state data in real time on the edge computing node to obtain a local state feature vector, and uploading the local state feature vector to the cloud computing node; Performing global fusion on the local state feature vectors on the cloud computing node to obtain a global state feature vector, and feeding the global state feature vector back to the digital twin model; Dynamically adjust the computing resource allocation of the edge computing node and the cloud computing node according to the difference between the global state feature vector and the local state feature vector.

7. The state monitoring and control system of the digital twin rehabilitation robot based on the Internet of Things according to claim 2 is characterized in that: The implementation method of the constrained optimization includes: Constructing a constrained optimization objective function, wherein the objective function takes minimizing the modulus of the state difference vector as an optimization goal, and takes the joint angle range constraint, the torque output safety constraint, and the myoelectric signal stability constraint as constraint conditions; Solving the constraint optimization objective function using a Lagrange multiplier method to obtain a candidate solution set of the optimization control instruction; Performing a stability evaluation on the candidate solution set, the stability evaluation comprising calculating an execution stability index of a control instruction corresponding to the candidate solution set in the digital twin model; The candidate solution with the highest stability index is selected as the optimization control instruction, and is sent to the physical system of the rehabilitation robot through the Internet of Things communication network.

Images