Intelligent rehabilitation nursing management system based on data analysis

Abstract

The invention relates to the technical field of data processing, in particular to an intelligent rehabilitation nursing management system based on data analysis, and the system comprises a motion state collection module which collects original electromyographic signals and joint torque and angle values, a physiological feature analysis module which extracts a frequency domain median frequency point sequence and constructs an electromyographic fatigue gradient; the motion compensation calculation module calculates the tail end driving torque and generates a motion compensation coefficient, and the self-adaptive control execution module calculates the resistance increment based on the fatigue gradient and calculates virtual wall parameters to construct a control strategy when the compensation coefficient exceeds the standard. According to the method, the fatigue gradient is constructed in combination with frequency domain analysis, the compensation ratio is calculated by utilizing the dynamic model, the resistance is adjusted according to the compensation coefficient and the fatigue state, and the virtual wall constraint is constructed, so that quantitative analysis of the physiological features and the motion state is realized, and the optimal strategy is adaptively matched under the condition that the safety is ensured; and the training intelligence level and accuracy are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to an intelligent rehabilitation nursing management system based on data analysis. BACKGROUND

[0002] The technical field of data processing refers to the technical category of using computer hardware and software systems to collect, store, retrieve, convert and operate various types of digital information. Among them, the traditional intelligent rehabilitation nursing management system refers to a local area network information input platform built based on a client and server architecture (C / S architecture). Its physical components usually include a medical workstation desktop computer, a barcode scanner, a network switch and a central database server. Medical staff fills in the basic identity information of patients, daily physical data and rehabilitation training projects one by one into the fixed form interface of the workstation software through physical keyboard and mouse. The data is transmitted via network lines and stored in the form of structured table in the mechanical hard disk of the server. The system simply archives and retrieves single data fields according to the preset logical rules.

[0003] The existing system highly depends on manual mechanical input through physical peripherals, resulting in tedious and inefficient data collection and being easily affected by human interference. Its storage mode limited to fixed forms can only simply archive static results, and cannot real-time perceive dynamic physiological changes and subtle action postures in patient training. This offline management makes it difficult for the system to capture muscle fatigue degree and compensation movement trend, causing the training scheme to be out of touch with the actual function of patients, reducing the treatment pertinence and effectiveness, and failing to intervene in time when the patient's action deforms, thus burying the safety hazard of secondary injury. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art and propose an intelligent rehabilitation nursing management system based on data analysis.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme, an intelligent rehabilitation nursing management system based on data analysis comprises:

[0006] A motion state acquisition module activates a multi-channel sensor to collect raw electromyographic voltage signals of target muscle groups and compensatory muscle groups, reads real-time joint torque values output by a rehabilitation robot joint torque sensor, and acquires real-time angle values fed back by a position encoder;

[0007] A physiological feature analysis module performs fast Fourier transform on the raw electromyographic voltage signals to calculate power spectral density distribution, extracts a frequency domain median frequency point sequence from the power spectral density distribution, calls a linear regression model to calculate a time evolution slope value of the frequency domain median frequency point sequence, and constructs an electromyographic fatigue gradient;

[0008] A motion compensation calculation module inputs the real-time joint torque value and the real-time angle value into a rigid body dynamics model to calculate an end driving torque, decomposes the end driving torque into a target torque component and a compensation torque component, calculates a proportion value of the compensation torque component in the end driving torque, and generates a motion compensation coefficient;

[0009] An adaptive control execution module calculates a resistance increment value and generates a motor current control instruction based on the electromyographic fatigue gradient, calculates a spatial coordinate constraint parameter of a reverse virtual wall when the motion compensation coefficient exceeds a preset threshold, and constructs a rehabilitation control strategy according to the motor current control instruction and the spatial coordinate constraint parameter.

[0010] As a further scheme of the present application, the specific function of the motion state acquisition module is implemented as:

[0011] A signal sensing sub-module activates a multi-channel electromyographic electrode patch attached to the skin surface of a patient in response to a system start instruction, synchronously captures weak bioelectric signals of target muscle groups and compensatory muscle groups at a preset high-frequency sampling rate, performs preamplification and filtering processing on the weak bioelectric signals, and generates the original electromyographic voltage signals;

[0012] A kinematics reading sub-module accesses a joint driver interface of a rehabilitation robot arm in real time through a communication bus, reads torque feedback data of a plurality of joint torque sensors under high dynamic motion, simultaneously acquires absolute angle position data output by a joint position encoder, performs time stamp alignment and denoising processing on the torque feedback data and the absolute angle position data, and generates the real-time joint torque value and the real-time angle value.

[0013] As a further scheme of the present application, the specific function of the physiological feature analysis module is implemented as:

[0014] A spectrum conversion sub-module applies windowing and truncation processing to the original electromyographic voltage signals in the time domain, maps the time domain signals to a frequency domain space by using a fast Fourier transform algorithm, calculates energy amplitudes of a plurality of frequency components, and generates the power spectral density distribution representing frequency structure characteristics of electromyographic signals;

[0015] A frequency extraction sub-module traverses the power spectral density distribution, calculates a cumulative power spectral energy, identifies a frequency boundary point that equally divides the cumulative power spectral energy into two equal parts, continuously extracts the frequency boundary point according to a time sliding window, and generates a sequence of frequency points in the frequency domain;

[0016] The fatigue evaluation submodule takes the sequence of frequency points in the frequency domain as the dependent variable and takes the time sequence as the independent variable to establish a least square linear regression equation, analyzes the slope term of the least square linear regression equation to quantify the degree of the trend of the frequency center drifting to the low frequency, and generates the electromyographic fatigue gradient.

[0017] As a further scheme of the present application, the specific function of the motion compensation calculation module is implemented as:

[0018] The dynamics solver module obtains the link mass parameters and the inertia tensor matrix of the robot arm, combines the real-time joint torque value and the real-time angle value, inversely solves the force state of the end effector of the robot arm in the Cartesian space coordinate system by using the Lagrange dynamics equation, and generates the end driving torque;

[0019] The torque decomposition submodule obtains a preset standard rehabilitation training trajectory tangent vector, orthogonally projects the end driving torque to the standard rehabilitation training trajectory tangent vector direction to separate out the target torque component, and defines the remaining vector after excluding the target torque component from the end driving torque as the compensation torque component.

[0020] The coefficient generation submodule calculates the ratio between the module length of the compensation torque component and the module length of the end driving torque, weights and corrects the ratio in combination with the current motion smoothness index, and generates the motion compensation coefficient.

[0021] As a further scheme of the present application, the specific function of the adaptive control execution module is implemented as:

[0022] The resistance adjustment submodule establishes a nonlinear mapping relationship between the electromyographic fatigue gradient and the motor output impedance, dynamically reduces the training resistance set value as the electromyographic fatigue gradient increases, calculates the target current value of the plurality of joint motors according to the adjusted training resistance set value, and generates the motor current control instruction;

[0023] The constraint construction submodule monitors the motion compensation coefficient in real time, generates a virtual force field boundary with stiffness and damping characteristics outside the standard trajectory based on the current deviation direction when it is monitored that the motion compensation coefficient exceeds the preset compensation range, calculates the geometric position data of the virtual force field boundary, and generates the spatial coordinate constraint parameter;

[0024] The strategy fusion submodule takes the motor current control instruction as the bottom torque following basis and takes the spatial coordinate constraint parameter as the position restriction condition, logically superimposes torque control and position restriction by using an impedance controller, and generates the rehabilitation control strategy.

[0025] As a further scheme of the present application, the specific process that the frequency extraction submodule calculates the sequence of median frequency points in the frequency domain comprises:

[0026] The power spectrum density distribution in the current analysis window is acquired, and the total power energy value of the full frequency band from the direct current component to the highest effective frequency in the current analysis window is calculated;

[0027] The power spectrum density distribution is subjected to integral accumulation operation starting from the zero frequency, and the proportional relationship between the accumulated energy value and the total power energy value of the full frequency band is monitored in real time;

[0028] When the accumulated energy value first reaches 50% of the total power energy value of the full frequency band, the current corresponding frequency value is locked, and the frequency value is marked as the median frequency point of the current analysis window;

[0029] With continuous sliding of the time window, the above calculation process is repeated, and the sequentially obtained median frequency points are arranged in time sequence to generate the sequence of median frequency points in the frequency domain.

[0030] As a further scheme of the present application, the process that the fatigue evaluation submodule constructs the myoelectric fatigue gradient comprises:

[0031] The sequence of median frequency points in the frequency domain with a preset length is acquired, a data set including time variable and frequency variable is constructed, and a variation trend straight line of the data set is fitted by using a linear regression algorithm;

[0032] The slope parameter of the variation trend straight line is extracted, the sign and the numerical value of the slope parameter are judged, and if the slope parameter is a negative value, the absolute value of the slope parameter is standardized and defined as a muscle fatigue quantitative index after processing;

[0033] The muscle fatigue quantitative index is normalized and corrected in combination with physiological tolerance criteria of multiple muscle groups, and the myoelectric fatigue gradient is generated.

[0034] As a further scheme of the present application, the process that the dynamics solver submodule generates the end driving torque comprises:

[0035] The length, mass center position and mass of the links of the robot arm stored in the system database are called;

[0036] The angular velocity and angular acceleration of multiple joints are calculated based on the real-time angle value, and the theoretical joint torque required to offset the gravity term, the Coriolis force term and the centrifugal force term is calculated according to the inverse solution algorithm of rigid body dynamics;

[0037] A difference vector between the real-time joint torque value and the theoretical joint torque is calculated, and the difference vector is mapped to an end operation space by using a Jacobian matrix transpose method to generate the end driving torque.

[0038] As a further scheme of the present application, the process in which the constraint construction sub-module generates the spatial coordinate constraint parameter comprises:

[0039] A three-dimensional spatial point set of a standard rehabilitation trajectory is acquired, and a normal deviation distance between a current actual position of an end effector and the standard rehabilitation trajectory is calculated when the motion compensation coefficient exceeds a preset threshold value;

[0040] A virtual elastic potential energy field is constructed according to the normal deviation distance, an repulsion force direction of the virtual elastic potential energy field is set to be perpendicular to an inside of the standard rehabilitation trajectory, and an repulsion force gain is exponentially increased according to a magnitude of the normal deviation distance;

[0041] Isopotal surface coordinate data of the virtual elastic potential energy field and a corresponding stiffness coefficient matrix are extracted to generate the spatial coordinate constraint parameter.

[0042] As a further scheme of the present application, a specific formula in which the coefficient generation sub-module calculates the motion compensation coefficient is:

[0043] ;

[0044] wherein, represents the motion compensation coefficient, represents a Euclidean norm of the compensation torque component, represents a Euclidean norm of the end driving torque, represents a preset compensation cumulative penalty factor, represents a real-time position error value in which the end effector deviates from a tangent vector of the standard rehabilitation training trajectory, represents a starting moment of a training period, represents a current moment.

[0045] Compared with the prior art, the present application has the following advantages and positive effects:

[0046] In the application, by activating the multi-channel sensor and combining joint torque and position feedback to accurately capture real-time motion data, frequency domain analysis and linear regression operation are performed on the electromyographic signal to construct the electromyographic fatigue gradient, the end driving torque is calculated by using the rigid body dynamics model, and the compensation torque ratio is calculated, the resistance increment is dynamically adjusted according to the compensation coefficient and the fatigue state, and the reverse virtual wall space constraint is constructed, so as to realize the deep perception and quantitative analysis of the physiological characteristics and motion state of the patient, ensure the adaptive matching of the best rehabilitation strategy under the premise of ensuring the safety of training, and effectively improve the intelligent level and accuracy of rehabilitation training. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 The principle block diagram of the intelligent rehabilitation nursing management system based on data analysis of the application is shown in the figure;

[0048] Figure 2 The detailed execution flow chart of the motion state acquisition module of the application is shown in the figure;

[0049] Figure 3 The detailed execution flow chart of the physiological characteristic analysis module of the application is shown in the figure;

[0050] Figure 4 The detailed execution flow chart of the motion compensation calculation module of the application is shown in the figure;

[0051] Figure 5 The detailed execution flow chart of the adaptive control execution module of the application is shown in the figure. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical scheme and advantages of the application clearer, the technical scheme realized by software is described in detail below combined with system architecture diagram and embodiment. It should be understood that the specific embodiments described herein are only used to explain the technical scheme of the application, and do not constitute a limitation on the protection scope.

[0053] All user-related information (including but not limited to biometric information, identity information, behavior data, device information and other data that can be used for identity verification and personalized services) involved in the application are collected and processed on the premise that the user is fully informed and voluntarily authorizes and agrees. The collection, storage and use of all information strictly comply with the applicable national and regional laws and regulations, and meet the relevant data protection standards and policy requirements. The scope of use of data is limited to the purpose necessary for providing the technical services of the application, and reasonable technical and management measures will be taken in information protection and privacy security to ensure the security and confidentiality of user personal information.

[0054] In the description of the present application, the system architecture relationship or data processing flow indicated by the terms "hierarchy", "module", "interface", "data flow", "client", "server" and the like are defined based on the corresponding architecture diagram or flowchart of the embodiment, and such expression manner is only used to clearly explain the logical relationship of each element in the technical solution, but not to limit the physical deployment form. The "multiple" contains two or more technical units, including but not limited to multiple data nodes, processing threads, service instances or functional components, and other scalable elements, and the specific number is determined according to the actual business scenario.

[0055] Please refer to Figure 1 and Figure 2 The present application provides a technical solution: an intelligent rehabilitation nursing management system based on data analysis includes:

[0056] A motion state acquisition module activates a multi-channel sensor to collect raw electromyographic voltage signals of target muscle groups and compensatory muscle groups, reads real-time joint torque values output by a rehabilitation robot joint torque sensor, and obtains real-time angle values fed back by a position encoder;

[0057] The specific function implementation of the motion state acquisition module is:

[0058] A signal sensing sub-module activates a multi-channel electromyographic electrode attached to the skin surface of a patient in response to a system start instruction to synchronously capture weak bioelectric signals of target muscle groups and compensatory muscle groups at a preset high-frequency sampling rate, pre-amplifies and filters the weak bioelectric signals, and generates raw electromyographic voltage signals;

[0059] A kinematics reading sub-module accesses a joint driver interface of a rehabilitation robot in real time through a communication bus, reads torque feedback data of multiple joint torque sensors under high dynamic motion, simultaneously obtains absolute angle position data output by a joint position encoder, time-stamps and denoises the torque feedback data and the absolute angle position data, and generates real-time joint torque values and real-time angle values.

[0060] The motion state acquisition module is constructed as a perception front end of the entire system, responsible for accurately obtaining bioelectric signals and mechanical data in the human-machine interaction process. The specific function implementation of this module strictly follows the dual parallel architecture of signal sensing and kinematics reading.

[0061] The signal sensing sub-module first responds to the start acquisition instruction issued by the system main control unit. In this embodiment, the instruction code is set to . In response to the instruction, the signal sensing sub-module activates the multi-channel electromyography electrode patch attached to the skin surface of the patient's upper limb biceps brachii and trapezius muscle. The biceps brachii is the target muscle group, and the trapezius muscle is the compensatory muscle group. In order to ensure the quality of signal acquisition, the electrode patch selects Ag / AgCl wet electrode with a diameter of , and the skin is cleaned before attachment to ensure that the skin contact impedance is less than . The signal sensing sub-module integrates a precision instrument amplification circuit, which is configured to capture the weak bioelectric signals of the above muscle groups at a preset high-frequency sampling rate of . Considering that the amplitude of the original bioelectric signal is usually between and , and is easily disturbed by environmental noise, the signal sensing sub-module first amplifies the signal by a gain of times through a preamplifier, and sets the common-mode rejection ratio to be greater than . Subsequently, the amplified analog signal enters the hardware filter circuit. The circuit is specifically constructed to include a high-pass filter with a cutoff frequency of to remove baseline drift, and a low-pass filter with a cutoff frequency of to prevent signal aliasing, and a notch filter with a center frequency of to eliminate power frequency interference. After the above analog conditioning, the signal is quantized to a digital sequence of bits by an analog-to-digital converter, thereby generating an original electromyography voltage signal.

[0062] The original electromyography voltage signal refers to the voltage time sequence that can digitally represent the potential change generated by muscle fiber excitation after amplification, filtering and analog-to-digital conversion.

[0063] In order to verify the effectiveness of signal acquisition, an experiment was conducted on a hemiplegic rehabilitation patient. Table 1 shows the signal-to-noise ratio comparison data of the signal sensing sub-module at different filtering stages.

[0064] Table 1 Signal-to-noise ratio data table of signal sensing sub-module at different processing stages

[0065] Processing stage Signal peak-peak value Noise level Signal-to-noise ratio Signal integrity Original input 1.2 mV 0.45 mV 8.52 dB 65.4% After pre-amplification 1.2V 0.15V 18.06 dB 82.1% After filtering processing 1.15V 0.02V 35.19 dB 99.8%

[0066] As shown in Table 1, after the step-by-step processing of the signal sensing sub-module, the signal-to-noise ratio is significantly improved to , effectively eliminating environmental noise and providing a pure data source for subsequent analysis.

[0067] The kinematics reading sub-module is built as the perception nerve center of the rehabilitation robot arm. The sub-module establishes physical connection with the joint driver interface of the robot arm through the industrial EtherCAT real-time communication bus, and the communication period is strictly limited to The sub-module is configured to access torque sensors integrated at each joint of the robot arm in real time. In the present embodiment, the torque sensors are strain gauge torque sensors with a range of ± 10 Nm and an accuracy of 0.1% FS. In the high dynamic motion mode, i.e., when the angular velocity is greater than 1000° / s, the sub-module reads the raw voltage value output by the joint torque sensor at a frequency of 10 kHz and converts it into a physical quantity, i.e., torque feedback data, according to the sensor calibration curve. At the same time, the sub-module reads the absolute position encoder data installed at the motor shaft end in parallel, and the encoder has a resolution of 16 bits, thereby obtaining the current absolute angle position data of the robot arm. In view of the slight time deviation in sensor data acquisition, the sub-module internally runs a time stamp alignment algorithm. The algorithm reads the distributed clock on the bus to unify the sampling time of the torque data and the angle data and map them to the same time reference axis, and controls the maximum alignment error to be within 1 ms. Subsequently, the sub-module applies a sliding average filter with a length of 100 to denoise the data and eliminate the sharp noise generated by mechanical vibration, and finally generates synchronous and smooth real-time joint torque values and real-time angle values, which are packaged and stored in the shared memory area for calling by downstream modules.

[0068] The above-mentioned EtherCAT refers to an Ethernet control automation technology, which is an open-architecture Ethernet fieldbus system with high-speed refresh and low jitter characteristics, and is suitable for motion control systems with extremely high real-time requirements.

[0069] Please refer to Figure 1 and Figure 3 , the physiological feature analysis module, performs a fast Fourier transform on the original electromyographic voltage signal to calculate the power spectral density distribution, extracts the frequency domain median frequency point sequence from the power spectral density distribution, calls a linear regression model to calculate the time evolution slope value of the frequency domain median frequency point sequence, and constructs an electromyographic fatigue gradient.

[0070] The specific function implementation of the physiological feature analysis module is as follows:

[0071] The spectrum conversion sub-module applies windowing and truncation processing to the original electromyographic voltage signal in the time domain, maps the time domain signal to the frequency domain space using a fast Fourier transform algorithm, calculates the energy amplitude of multiple frequency components, and generates a power spectral density distribution representing the frequency structure characteristics of the electromyographic signal.

[0072] ​​​​​​The frequency extraction submodule traverses the power spectral density distribution and calculates cumulative power spectrum energy, identifies a frequency boundary point that equally divides the cumulative power spectrum energy into two equal parts, continuously extracts the frequency boundary point according to a time sliding window, and generates a sequence of median frequency points in the frequency domain;

[0073] The specific process of the frequency extraction submodule to calculate the sequence of median frequency points in the frequency domain includes:

[0074] The power spectral density distribution in the current analysis window is obtained, and the total power energy value of the full frequency band from the direct current component to the highest effective frequency in the current analysis window is calculated.

[0075] The power spectral density distribution is integrated and accumulated from zero frequency, and the proportional relationship between the accumulated energy value and the total power energy value of the full frequency band is monitored in real time.

[0076] When the accumulated energy value first reaches fifty percent of the total power energy value of the full frequency band, the current corresponding frequency value is locked, and the frequency value is marked as the median frequency point of the current analysis window.

[0077] With the continuous sliding of the time window, the above calculation process is repeated, and the sequentially obtained median frequency points are arranged in time sequence to generate a sequence of median frequency points in the frequency domain.

[0078] The fatigue assessment submodule takes the sequence of median frequency points in the frequency domain as the dependent variable and the time sequence as the independent variable, establishes a least squares linear regression equation, analyzes the slope term of the least squares linear regression equation to quantify the degree of low-frequency drift of the frequency center, and generates an electromyographic fatigue gradient.

[0079] The process of the fatigue assessment submodule to construct the electromyographic fatigue gradient includes:

[0080] A frequency domain median frequency point sequence of a preset length is obtained, a data set including time and frequency variables is constructed, and a one-dimensional linear regression algorithm is used to fit a trend line of the data set.

[0081] The slope parameter of the trend line is extracted, and the sign and value of the slope parameter are determined. If the slope parameter is negative, the absolute value of the slope parameter is standardized and defined as a muscle fatigue quantitative indicator.

[0082] The muscle fatigue quantitative indicator is normalized and corrected in combination with physiological tolerance benchmarks of multiple muscle groups to generate an electromyographic fatigue gradient.

[0083] The core task of the physiological feature analysis module is to convert the time domain electromyographic signal into a frequency domain feature and quantify muscle fatigue.

[0084] The spectrum conversion submodule first applies windowing truncation processing to the input raw electromyographic voltage signal. In order to reduce spectrum leakage, the submodule selects a Hamming window, the window length is set to sampling points, corresponding to a time length of about , and the window overlap rate is set to . For each windowed time domain data segment, the submodule calls a fast Fourier transform logic unit to perform operation. The specific operation logic is to convert point time domain sequence into point complex frequency domain sequence by using a butterfly operation structure. Subsequently, the submodule calculates the energy amplitude of each frequency component, that is, calculates the modulus square of the complex number and divides by the window normalization coefficient, to generate the power spectral density distribution. The distribution is stored in an array form, covering the frequency band from to , and the frequency resolution is .

[0085] The above Hamming window refers to a commonly used window function in signal processing, which can effectively reduce the sidelobe leakage effect in spectrum analysis and improve the spectrum resolution by smoothly attenuating both ends of the time domain signal.

[0086] The frequency extraction submodule is configured to calculate the median frequency point in the frequency domain based on the power spectral density distribution. The submodule first obtains the power spectral density distribution array within the current analysis window, and calculates the total power energy value of the full frequency band from the direct current component, that is, to the highest effective frequency by accumulation operation. In the embodiment, the highest effective frequency is set to . Subsequently, the submodule starts from the frequency index zero, and performs point-by-point integration and accumulation operation on the power spectral density, and monitors the current accumulation energy value in real time. The submodule is internally provided with a comparator logic for real-time judgment of the size relationship between the ratio of the accumulation energy value to the total power energy value of the full frequency band and . When it is monitored that the ratio first reaches or exceeds , the submodule immediately locks the current frequency index, and marks the corresponding physical frequency value as the median frequency point of the current analysis window. With each sliding of the time window, the submodule repeats the above calculation process, and sequentially stores the obtained median frequency points in a first-in-first-out queue in time order to generate a frequency domain median frequency point sequence.

[0087] The fatigue evaluation submodule is configured to quantify the muscle fatigue degree. The submodule reads the frequency domain median frequency point sequence of a preset length. In the embodiment, the preset length is set to the data of the last seconds, that is, about The sub-module constructs a data set containing time variable as independent variable matrix and frequency variable as dependent variable matrix. Then, the sub-module calls the least square linear regression algorithm unit. The algorithm unit performs matrix operation to calculate the regression coefficient vector. The sub-module extracts the slope item of the regression equation. Physiologically, muscle fatigue is manifested as a shift of median frequency to low frequency, i.e. the slope should be negative. The sub-module judges the sign and value of the slope item. If the slope item is negative, it indicates that there is a fatigue trend. The sub-module normalizes the absolute value of the slope parameter, i.e. calculates the ratio of the absolute value of the slope to the preset reference slope of severe fatigue. In the embodiment, the reference slope of severe fatigue is set to . Finally, the sub-module combines the physiological tolerance benchmark of the specific muscle group stored in the database to normalize and correct the index, generating the final muscle fatigue gradient. In the embodiment, the physiological tolerance benchmark of the biceps brachii is set to , and that of the trapezius is set to .

[0088] Table 2 shows the measured data and calculation results of the module in an isometric contraction training lasting seconds.

[0089] Table 2 Muscle fatigue gradient calculation process data table

[0090] Time period MDF sequence mean value Calculation slope Determination result Fatigue index Electromyographic fatigue gradient 0−10s 95.4 Hz −0.05 No obvious fatigue 0.06 0.06 10−20s 88.2 Hz −0.42 Mild fatigue 0.52 0.52 20−30s 76.5 Hz −0.85 Severe fatigue 1.06 1.06

[0091] As shown in Table 2, the absolute value of the median frequency slope increases significantly as the training progresses, and the muscle fatigue gradient calculated by the system accurately reflects the change in muscle state, indicating that the system can quantitatively capture the small physiological fatigue trend.

[0092] Please refer to Figure 1 and Figure 4 , the motion compensation algorithm module inputs the real-time joint torque value and real-time angle value into the rigid body dynamics model to solve the end driving torque, decomposes the end driving torque into target torque component and compensation torque component, calculates the proportion value of the compensation torque component in the end driving torque, and generates the motion compensation coefficient.

[0093] The specific function implementation of the motion compensation algorithm module is as follows:

[0094] The dynamics solver module obtains the link mass parameters and inertia tensor matrix of the robot arm, combines the real-time joint torque value and real-time angle value, and uses the Lagrange dynamics equation to inversely solve the force state of the robot arm end effector in the Cartesian space coordinate system, generating the end driving torque.

[0095] The process of the dynamics solver module generating the end driving torque includes:

[0096] Call the mechanical arm link length, link mass center position and link mass parameters stored in the system database;

[0097] Based on the real-time angle value, the angular velocity and angular acceleration of multiple joints are calculated, and according to the rigid body dynamics inverse algorithm, the theoretical joint torque required to offset the gravity term, the Coriolis force term and the centrifugal force term is calculated;

[0098] The difference vector between the real-time joint torque value and the theoretical joint torque is calculated, and the difference vector is mapped to the end operation space by using the Jacobian matrix transpose method to generate the end driving torque;

[0099] The torque decomposition sub-module obtains the preset standard rehabilitation training trajectory tangent vector, orthogonally projects the end driving torque to the standard rehabilitation training trajectory tangent vector direction to separate out the target torque component, and defines the remaining vector after excluding the target torque component in the end driving torque as the compensation torque component.

[0100] The coefficient generation sub-module calculates the ratio between the modulus of the compensation torque component and the modulus of the end driving torque, and combines the current motion smoothness index to weight and correct the ratio to generate the motion compensation coefficient.

[0101] The specific formula for calculating the motion compensation coefficient by the coefficient generation sub-module is:

[0102] ;

[0103] Wherein, represents the motion compensation coefficient, represents the Euclidean norm of the compensation torque component, represents the Euclidean norm of the end driving torque, represents the preset compensation cumulative penalty factor, represents the real-time position error value of the end effector deviating from the standard rehabilitation training trajectory tangent vector, represents the starting time of the training period, represents the current time.

[0104] The motion compensation calculation module accurately quantifies the compensation behavior of the patient during the training process through dynamics calculation and torque decomposition technology.

[0105] The dynamics calculation sub-module is configured to back-propagate the end force according to the motion state of the robot. The sub-module first calls the pre-stored mechanical arm link parameters from the system database, including the link length , , , and the link mass parameter , , ​ Joint angle and angular velocity and acceleration are acquired in real time. The inverse dynamics module calculates the joint torque required to counteract the gravity and inertia of the robot arm. The difference between the real-time joint torque and the theoretical joint torque is calculated. The difference represents the interaction torque exerted by the patient. The interaction torque is mapped to the Cartesian space using the Jacobian matrix to obtain the end-effector force.

[0106] The Jacobian matrix refers to the transpose of the robot Jacobian matrix, which is used to linearly map the joint torque vector to the end-effector force or torque vector in the Cartesian space.

[0107] The torque decomposition module and the coefficient generation module work together to quantify the compensation level. The torque decomposition module obtains the tangent vector of the standard rehabilitation training trajectory at the current time. The module performs vector projection operations to project the end-effector driving torque onto the tangent vector of the standard rehabilitation training trajectory, calculating the target torque component. At the same time, the compensation torque component is calculated through vector subtraction. This step decomposes the force exerted by the patient into effective work force and ineffective compensation force perpendicular to the movement direction.

[0108] The coefficient generation module calculates the motion compensation coefficient based on the following formula:

[0109] ;

[0110] Wherein, represents the motion compensation coefficient, which is used to comprehensively evaluate the current compensation level; represents the Euclidean norm of the compensation torque component, i.e. the size of the ineffective torque; represents the Euclidean norm of the end-effector driving torque, i.e. the total driving torque exerted by the patient; represents the preset compensation cumulative penalty factor, which is used to adjust the influence weight of historical position error on the current compensation coefficient. In this embodiment, the value is ; represents the position error integral term, i.e. the real-time position error value of the end-effector deviating from the tangent vector of the standard rehabilitation training trajectory in time; represents the starting time of the training period; represents the current time.

[0111] In this embodiment, each parameter in the formula is configured and valued as follows to verify its technical effect. Assuming that at a certain time, the end-effector driving torque vector is calculated in real time by the dynamics solving submodule , and its Euclidean norm is Assuming the standard trajectory tangent vector is along the positive X-axis, the compensated component is calculated as follows: Its norm is Assume that the training has continued from the start to the current time. The system is based on Frequency sampling, the average distance of the end effector deviating from the standard trajectory is The approximate value of the integral is Substitute the above values ​​into the formula for calculation: First, calculate the torque ratio. Then calculate the penalty term as follows: The final calculated compensation coefficient is: .

[0112] Table 3. Calculation Examples of Motion Compensation Coefficient

[0113]

[0114] This result indicates that, although instantaneous torque compensation accounts for only a small percentage... However, due to the patient's persistent cumulative positional deviation, the final compensation coefficient is amplified by the penalty term. The advantage of this formula is that it not only considers the current error in the direction of force, but also introduces historical evaluation over time through the integral term. This effectively identifies implicit compensatory behaviors where the body posture is skewed for a long time, even if the direction of force is barely correct, making the compensation assessment more comprehensive and rigorous.

[0115] Please see Figure 1 and Figure 5 The adaptive control execution module calculates the resistance increment value based on the electromyographic fatigue gradient and generates motor current control commands. When the motion compensation coefficient exceeds the preset threshold, it calculates the spatial coordinate constraint parameters of the reverse virtual wall and constructs a rehabilitation control strategy based on the motor current control commands and spatial coordinate constraint parameters.

[0116] The specific functions of the adaptive control execution module are as follows:

[0117] The resistance adjustment submodule establishes a nonlinear mapping relationship between the electromyographic fatigue gradient and the motor output impedance. As the electromyographic fatigue gradient increases, the training resistance setting value is dynamically reduced. Based on the adjusted training resistance setting value, the target current value of multiple joint motors is calculated, and motor current control commands are generated.

[0118] The constraint construction submodule monitors the motion compensation coefficient in real time. When the motion compensation coefficient exceeds the preset compensation range, it generates a virtual force field boundary with stiffness and damping characteristics outside the standard trajectory based on the current deviation direction, calculates the geometric position data of the virtual force field boundary, and generates spatial coordinate constraint parameters.

[0119] The process of generating spatial coordinate constraint parameters by the constraint construction submodule includes:

[0120] Obtain the three-dimensional spatial point set of the standard rehabilitation trajectory, and when the motion compensation coefficient exceeds the preset threshold, calculate the normal deviation distance between the current actual position of the end effector and the standard rehabilitation trajectory.

[0121] A virtual elastic potential energy field is constructed based on the normal deviation distance. The direction of the repulsive force of the virtual elastic potential energy field is set to be perpendicular to the inside of the standard rehabilitation trajectory, and the repulsive force gain is increased exponentially according to the magnitude of the normal deviation distance.

[0122] Extract the equipotential surface coordinate data and the corresponding stiffness coefficient matrix of the virtual elastic potential energy field to generate spatial coordinate constraint parameters;

[0123] The strategy fusion submodule uses motor current control commands as the underlying torque following basis and spatial coordinate constraint parameters as position constraints. Through an impedance controller, torque control and position constraints are logically superimposed to generate a rehabilitation control strategy.

[0124] The adaptive control execution module dynamically adjusts the control strategy of the robotic arm based on the front-end analysis results to achieve intelligent assistance.

[0125] The resistance adjustment submodule is configured to dynamically adjust the physical parameters of rehabilitation training based on the aforementioned generated electromyographic fatigue gradient. The submodule internally uses a preset nonlinear mapping function, where the damping coefficient equals the initial damping setpoint multiplied by a negative power of the natural constant, where the power is the product of the decay rate constant and the electromyographic fatigue gradient. In this embodiment, the initial damping setpoint is set to... The decay rate constant is set to If the current electromyographic fatigue gradient is calculated as follows: Then the system calculates the new damping coefficient. The calculation process is as follows: the exponential part is... Multiply equal natural constant The power is approximately The final new damping coefficient is Multiply equal Subsequently, the submodule calculates the target current value that each joint motor needs to output based on the adjusted resistance setting and the current movement speed, and generates motor current control commands, thereby automatically reducing the current when the patient is fatigued. The training resistance helps prevent muscle damage.

[0126] The constraint construction submodule is configured to intervene when severe compensation is detected. This submodule monitors the motion compensation coefficient in real time. The system sets a preset threshold. In the above example, the calculated compensation coefficient is: The threshold has been exceeded. At this point, the constraint construction submodule immediately triggers the virtual wall generation logic. This submodule acquires the three-dimensional spatial point set of the standard rehabilitation trajectory and calculates the normal deviation distance from the current position of the end effector to the nearest point on the trajectory. A virtual elastic potential energy field is constructed based on the normal deviation distance. This submodule sets the virtual stiffness not to be constant, but to increase exponentially with distance. In this embodiment, the basic stiffness is set to be... The gain coefficient is If the current deviation distance is The stiffness provided by the virtual wall is calculated as follows: Multiplied by the natural constant The result of subtracting one from the power is:

[0127] This submodule calculates the corresponding repulsive force vector and defines the repulsive force and the corresponding stiffness matrix as spatial coordinate constraint parameters.

[0128] The strategy fusion submodule ultimately fuses the above parameters through an impedance controller. It receives motor current control commands as feedforward torque and spatial coordinate constraint parameters as feedback correction force, generating the final force through logical superposition. The rehabilitation control strategy signals drive the robotic arm to provide appropriate training resistance and form a flexible tunnel in space, forcibly constraining the patient's movements within a safe range.

[0129] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the protection scope defined by the technical solution of the present invention.

Claims

1. An intelligent rehabilitation and nursing management system based on data analysis, characterized in that, The system includes: The motion state acquisition module activates multi-channel sensors to acquire raw electromyographic voltage signals of the target muscle group and compensatory muscle group, reads the real-time joint torque value output by the joint torque sensor of the rehabilitation robotic arm, and obtains the real-time angle value fed back by the position encoder. The physiological feature analysis module performs a fast Fourier transform on the original electromyographic voltage signal to calculate the power spectral density distribution, extracts the median frequency point sequence in the frequency domain from the power spectral density distribution, calls a linear regression model to calculate the time evolution slope value of the median frequency point sequence in the frequency domain, and constructs the electromyographic fatigue gradient. The motion compensation calculation module inputs the real-time joint torque value and the real-time angle value into the rigid body dynamics model to solve the end-drive torque, decomposes the end-drive torque into a target torque component and a compensation torque component, calculates the proportion of the compensation torque component in the end-drive torque, and generates a motion compensation coefficient. The adaptive control execution module calculates the resistance increment value based on the electromyographic fatigue gradient and generates a motor current control command. When the motion compensation coefficient exceeds a preset threshold, it calculates the spatial coordinate constraint parameters of the reverse virtual wall and constructs a rehabilitation control strategy based on the motor current control command and the spatial coordinate constraint parameters.

2. The intelligent rehabilitation and nursing management system based on data analysis according to claim 1, characterized in that, The specific functions of the motion state acquisition module are as follows: The signal sensing submodule responds to the system start command to activate the multi-channel electromyographic electrode pads attached to the patient's skin surface, and synchronously captures the weak bioelectric signals of the target muscle group and the compensating muscle group at a preset high-frequency sampling rate. The weak bioelectric signals are pre-amplified and filtered to generate the original electromyographic voltage signal. The kinematic reading submodule accesses the joint actuator interface of the rehabilitation robotic arm in real time via the communication bus, reads torque feedback data from multiple joint torque sensors under high dynamic motion, and simultaneously acquires absolute angle position data output by the joint position encoder. The torque feedback data and the absolute angle position data are then time-stamped and denoised to generate the real-time joint torque value and the real-time angle value.

3. The intelligent rehabilitation and nursing management system based on data analysis according to claim 1, characterized in that, The specific functions of the physiological characteristic analysis module are as follows: The spectrum conversion submodule applies windowing truncation processing to the original electromyographic voltage signal in the time domain, uses the fast Fourier transform algorithm to map the time domain signal to the frequency domain space, calculates the energy amplitude of multiple frequency components, and generates the power spectral density distribution that characterizes the frequency structure of the electromyographic signal. The frequency extraction submodule traverses the power spectral density distribution and calculates the cumulative power spectral energy, identifies the frequency boundary points that divide the cumulative power spectral energy into two equal parts, and continuously extracts the frequency boundary points according to the time sliding window to generate the frequency domain median frequency point sequence. The fatigue assessment submodule uses the mid-frequency point sequence in the frequency domain as the dependent variable and the time series as the independent variable to establish a least squares linear regression equation. It analyzes the slope term of the least squares linear regression equation to quantify the degree of frequency center drift towards lower frequencies and generates the electromyographic fatigue gradient.

4. The intelligent rehabilitation and nursing management system based on data analysis according to claim 1, characterized in that, The specific functions of the motion compensation calculation module are as follows: The dynamics solution submodule obtains the link mass parameters and inertia tensor matrix of the robotic arm, and combines the real-time joint torque value and the real-time angle value to use the Lagrange dynamics equation to solve the force state of the robotic arm end effector in the Cartesian coordinate system, thereby generating the end-effector driving torque. The torque decomposition submodule obtains the preset standard rehabilitation training trajectory tangent vector, orthogonally projects the end driving torque onto the direction of the standard rehabilitation training trajectory tangent vector to separate the target torque component, and defines the remaining vector after removing the target torque component from the end driving torque as the compensating torque component. The coefficient generation submodule calculates the ratio between the modulus of the compensating torque component and the modulus of the end drive torque, and performs a weighted correction on the ratio in conjunction with the current motion smoothness index to generate the motion compensation coefficient.

5. The intelligent rehabilitation and nursing management system based on data analysis according to claim 1, characterized in that, The specific functions of the adaptive control execution module are as follows: The resistance adjustment submodule establishes a nonlinear mapping relationship between the electromyographic fatigue gradient and the motor output impedance. As the electromyographic fatigue gradient increases, the training resistance setting value is dynamically reduced. Based on the adjusted training resistance setting value, the target current values ​​of multiple joint motors are calculated, and the motor current control command is generated. The constraint construction submodule monitors the motion compensation coefficient in real time. When the motion compensation coefficient exceeds the preset compensation range, it generates a virtual force field boundary with stiffness and damping characteristics outside the standard trajectory based on the current deviation direction, calculates the geometric position data of the virtual force field boundary, and generates the spatial coordinate constraint parameters. The strategy fusion submodule uses the motor current control command as the underlying torque following basis and the spatial coordinate constraint parameters as position constraints. Through the impedance controller, the torque control and position constraints are logically superimposed to generate the rehabilitation control strategy.

6. The intelligent rehabilitation and nursing management system based on data analysis according to claim 3, characterized in that, The specific process by which the frequency extraction submodule calculates the mid-frequency point sequence in the frequency domain includes: Obtain the power spectral density distribution within the current analysis window, and calculate the total power energy value across the entire frequency band from the DC component to the highest effective frequency within the current analysis window; The power spectral density distribution is integrated and accumulated starting from zero frequency, and the ratio between the accumulated energy value and the total power energy value of the entire frequency band is monitored in real time. When the accumulated energy value first reaches 50% of the total power energy value of the entire frequency band, the corresponding frequency value is locked and the frequency value is marked as the median frequency point of the current analysis window. As the time window slides continuously, the above calculation process is repeated, and the median frequency points obtained in sequence are arranged in chronological order to generate the frequency domain median frequency point sequence.

7. The intelligent rehabilitation and nursing management system based on data analysis according to claim 3, characterized in that, The process of constructing the electromyographic fatigue gradient by the fatigue assessment submodule includes: Obtain the frequency point sequence of the frequency domain of the preset length, construct a dataset including time variables and frequency variables, and fit the trend line of the dataset using a univariate linear regression algorithm; Extract the slope parameter of the trend line, determine the sign and magnitude of the slope parameter, and if the slope parameter is negative, define the absolute value of the slope parameter as a quantitative index of muscle fatigue after standardization. The muscle fatigue quantification index is normalized and corrected by combining the physiological tolerance benchmarks of multiple muscle groups to generate the electromyographic fatigue gradient.

8. The intelligent rehabilitation and nursing management system based on data analysis according to claim 4, characterized in that, The process by which the dynamics calculation submodule generates the end-drive torque includes: Call the robot arm link length, link center of mass position and link mass parameters stored in the system database; Based on the real-time angle values, the angular velocities and angular accelerations of multiple joints are calculated. According to the inverse kinematics algorithm, the theoretical joint torques required to counteract the gravity, Coriolis force, and centrifugal force terms are calculated. The difference vector between the real-time joint torque value and the theoretical joint torque is calculated, and the difference vector is mapped to the end effector space using the Jacobian matrix transpose method to generate the end effector drive torque.

9. The intelligent rehabilitation and nursing management system based on data analysis according to claim 5, characterized in that, The process by which the constraint construction submodule generates the spatial coordinate constraint parameters includes: Obtain a three-dimensional spatial point set of the standard rehabilitation trajectory; when the motion compensation coefficient exceeds a preset threshold, calculate the normal deviation distance between the current actual position of the end effector and the standard rehabilitation trajectory. A virtual elastic potential energy field is constructed based on the normal deviation distance. The repulsive force direction of the virtual elastic potential energy field is set to be perpendicular to the inside of the standard rehabilitation trajectory, and the repulsive force gain is increased exponentially according to the magnitude of the normal deviation distance. Extract the equipotential surface coordinate data and the corresponding stiffness coefficient matrix of the virtual elastic potential energy field to generate the spatial coordinate constraint parameters.

10. The intelligent rehabilitation and nursing management system based on data analysis according to claim 4, characterized in that, The specific formula for calculating the motion compensation coefficient by the coefficient generation submodule is as follows: ； in, This represents the motion compensation coefficient. The Euclidean norm representing the compensating torque component, The Euclidean norm representing the end-drive torque. This represents the pre-defined cumulative penalty factor for compensation. This represents the real-time position error value of the end effector deviating from the tangent vector of the standard rehabilitation training trajectory. This represents the start time of the training cycle. It represents the current moment.

Images