Dynamic weight compensation method and device based on low-torque servo

Abstract

The application relates to a dynamic weight compensation method and device based on low-torque servo, which comprises the following steps: obtaining mechanical response data and user motion characteristics of a target device, performing pre-defined action type matching, and obtaining a servo compensation parameter set; driving a main resistance unit of the target device to generate a reference resistance according to the mechanical response data, performing double-path control analysis in combination with the servo compensation parameter set, and obtaining double-path compensation data; obtaining mechanism vibration data of the target device, suppressing torque fluctuation of the double-path compensation data, and obtaining a low-torque compensation signal; performing mechanical state prediction according to the user motion characteristics and the mechanism vibration data, and outputting a low-torque balance threshold; and performing closed-loop fusion on the low-torque compensation signal and the low-torque balance threshold, and generating global compensation information. The application can realize dynamic optimization of resistance compensation, and can avoid over-compensation or under-compensation.

Description

Technical Field

[0001] This invention relates to the field of instrument torque and weight compensation technology, and in particular to a dynamic weight compensation method and device based on low-torque servo. Background Technology

[0002] Low-torque servo systems play a crucial role in precision motion control and human-machine interface devices. The core challenge lies in achieving high-precision, low-energy dynamic resistance compensation. However, most existing dynamic compensation methods rely on static mechanical models or single sensor data, making it difficult to adapt to nonlinear load variations in complex motion scenarios, resulting in insufficient compensation accuracy and increased energy consumption. Furthermore, traditional methods perform poorly in suppressing mechanism vibration and torque fluctuations, and lack dynamic prediction and closed-loop constraints of the mechanical state, making the compensation process prone to over-adjustment or under-adjustment, affecting the system's stability and reliability. Summary of the Invention

[0003] To address the existing problems, this invention aims to solve at least one of the technical problems existing in the prior art. To this end, this invention provides a dynamic weight compensation method and apparatus based on low-torque servo technology. This method can generate a low-torque balance threshold through mechanical state prediction and fuse it with the compensation signal in a closed loop, achieving dynamic optimization of resistance compensation and avoiding over-compensation or under-compensation.

[0004] To achieve the above objectives, the present invention provides a dynamic weight compensation method based on low-torque servo, comprising: The mechanical response data of the target device and the user's motion characteristics are obtained, and the predefined action types are matched to obtain the servo compensation parameter set. Based on the mechanical response data, the main resistance unit of the target device is driven to generate a reference resistance, and dual-path control analysis is performed in combination with the servo compensation parameter set to obtain dual-path compensation data; The mechanical vibration data of the target device is acquired, and torque fluctuation suppression is performed on the dual-channel compensation data to obtain a low torque compensation signal. Based on the user's motion characteristics and the mechanism's vibration data, the mechanical state is predicted, and a low torque balance threshold is output. Closed-loop fusion is performed on the low torque compensation signal and the low torque balance threshold to generate global compensation information.

[0005] Preferably, the step of acquiring the mechanical response data of the target device and the user's motion characteristics, performing predefined action type matching, and obtaining a servo compensation parameter set includes: The mechanical response data is obtained by acquiring the original tensile data sequence through the strain sensor array of the target device and performing time-domain noise reduction. The triaxial acceleration and angular velocity data streams of the inertial measurement unit of the target device are acquired, and motion features are extracted to obtain the user motion features; Based on a preset compensation parameter benchmark library, the mechanical response data and the user motion characteristics are indexed by a predefined type to obtain a predefined action type; Based on the predefined action type, the mechanical response data and the user motion characteristics are converted into equivalent compensation for reference inertia and transmission chain to obtain the servo compensation parameter set.

[0006] Preferably, the step of driving the main resistance unit of the target device to generate a reference resistance based on the mechanical response data, and performing dual-path control analysis in conjunction with the servo compensation parameter set to obtain dual-path compensation data includes: The mechanical response data is subjected to frequency domain energy spectrum separation to construct a dynamic mechanical feature vector; Based on the dynamic mechanical feature vector, the magnetic circuit output nonlinear approximation of the main resistance unit is performed to obtain the reference resistance distribution set; The servo compensation parameter set is decomposed into translational and rotational control according to the preset kinematic orthogonal decoupling mechanism to obtain a two-degree-of-freedom control input group. The reference resistance distribution set and the two-degree-of-freedom control input set are dynamically and interactively corrected to obtain a hybrid control protocol. The main resistance unit is reconstructed using the power drive modulation signal according to the hybrid control protocol to obtain the dual-channel compensation data.

[0007] Preferably, the step of acquiring the mechanical vibration data of the target device and suppressing torque fluctuations in the dual-channel compensation data to obtain a low-torque compensation signal includes: The vibration data of the mechanism are subjected to spectral feature separation to obtain the main resonant mode component and the secondary resonant mode component; Based on the dual-path compensation data, phase lag compensation is performed on the main resonant mode component to obtain the set of critical oscillation frequency points; Based on the set of critical oscillation frequency points, the driving waveform envelope boundary constraint is applied to the dual-channel compensation data to obtain the anti-saturation compensation signal; The active damping injection coefficient is obtained by performing coupled resonance energy inversion calculation on the sub-resonant mode components and the anti-saturation compensation signal. Based on the dual-path compensation data and the active damping injection coefficient, the pulse width modulation depth signal is reconstructed to obtain the low torque compensation signal.

[0008] Preferably, the step of performing coupled resonance energy inversion calculations on the sub-resonant mode components and the anti-saturation compensation signal to obtain the active damping injection coefficient includes: Energy spectrum focusing extraction is performed on the sub-resonant mode components to obtain a set of resonant peak frequencies; Based on the anti-saturation compensation signal, drive phase difference mapping is performed to generate an anti-phase compensation sequence; Based on the set of resonant peak frequencies and the anti-phase compensation sequence, a dual-domain damping correlation calculation is performed to generate an initial anti-phase damping function. Perform orthogonal decoupling and gain factor correction on the initial antiphase damping function to obtain a set of gain damping functions; The active damping injection coefficient is obtained by performing energy inversion calculation on the gain damping function set.

[0009] Preferably, the step of predicting the mechanical state based on the user's motion characteristics and the mechanism's vibration data, and outputting a low torque balance threshold, includes: Periodic action phase segmentation is performed on the user's motion features to obtain a motion state label sequence; The dominant frequency band energy features of the vibration data of the mechanism are extracted to obtain the mechanical focusing frequency band information; A dynamic interactive state mapping relationship is constructed based on the motion state marker sequence and the mechanical aggregation frequency band information; The dynamic interactive state mapping relationship is quantitatively stable to obtain the critical working region boundary; Based on the boundary of the critical working region, a low torque safety margin conversion is performed to generate dynamic safety torque constraint parameters. The dynamic safety torque constraint parameters are adaptively matched to obtain the low torque balance threshold.

[0010] Preferably, the adaptive matching of the dynamic safety torque constraint parameters to obtain the low torque balance threshold includes: The dynamic safety torque constraint parameters are decomposed into axial torque constraint components and radial torque constraint components. The coupling relationship between the axial torque constraint component and the radial torque constraint component is analyzed to obtain the torque component correlation coefficient. Based on the torque component correlation coefficient, the axial torque constraint component and the radial torque constraint component are weighted and fused to obtain a comprehensive dynamic torque constraint vector. The boundary matching degree of the comprehensive dynamic torque constraint vector is calculated based on the preset torque feedback standard data to obtain the torque constraint deviation. The comprehensive dynamic torque constraint vector is dynamically thresholded based on the torque constraint deviation to obtain a low torque balance threshold.

[0011] Preferably, the step of performing closed-loop fusion of the low-torque compensation signal and the low-torque balance threshold to generate global compensation information includes: Obtain the reference resistance of the main resistance unit, and perform waveform reference alignment on the low torque compensation signal to obtain a reference consistent compensation signal; Based on the low torque balance threshold, the benchmark consistent compensation signal is subjected to boundary limitation calculation and heterogeneous channel fusion to obtain the composite resistance intermediate signal; Based on the vibration data of the mechanism, the intermediate signal of the composite resistance is offset-corrected to obtain a phase-coordinated signal; The phase-coordinated signal is subjected to energy distribution equalization and global resistance interpolation fitting based on the servo compensation parameter set to obtain the global compensation information.

[0012] The present invention also provides a dynamic weight compensation device based on low-torque servo, applied to any one of the above-described dynamic weight compensation methods based on low-torque servo, comprising: The acquisition module is used to acquire the mechanical response data of the target device and the user's motion characteristics, perform predefined action type matching, and obtain a set of servo compensation parameters. The analysis module is used to drive the main resistance unit of the target device to generate a reference resistance based on the mechanical response data, and to perform dual-path control analysis in combination with the servo compensation parameter set to obtain dual-path compensation data. The processing module is used to acquire the mechanical vibration data of the target device, suppress torque fluctuations in the dual-channel compensation data, and obtain a low torque compensation signal. The association module is used to predict the mechanical state based on the user's motion characteristics and the mechanism's vibration data, and output a low torque balance threshold. A construction module is used to perform closed-loop fusion of the low torque compensation signal and the low torque balance threshold to generate global compensation information.

[0013] The present invention provides a dynamic weight compensation method and device based on low-torque servo, which has the following beneficial effects: By matching user motion characteristics with equipment mechanical response data, precise parameter adaptation for predefined motion types is achieved, effectively solving the problem of decreased compensation accuracy caused by insufficient parameter matching in traditional methods, and significantly improving the accuracy of resistance control. Combining mechanism vibration data with dual-path compensation data for torque fluctuation suppression optimizes the low-torque compensation signal, effectively overcoming the shortcomings of traditional methods in vibration suppression, reducing the risk of mechanical resonance, and improving system response stability. Generating a low-torque balance threshold through mechanical state prediction and fusing it with the compensation signal in a closed loop achieves dynamic optimization of resistance compensation, avoiding overcompensation or undercompensation, and reducing energy consumption while ensuring system stability. Based on dual-path control analysis and global compensation information generation, it can adapt to nonlinear load changes, solving the problem that traditional static models cannot cope with complex working conditions, and significantly improving the system's adaptability to diverse motion scenarios. The combination of dynamic prediction and closed-loop constraint mechanisms ensures real-time coordination between the compensation process and mechanical state, effectively reducing system instability caused by lack of constraints in traditional methods and enhancing the long-term operational reliability of the equipment. Attached Figure Description

[0014] Figure 1 This is a flowchart of a dynamic weight compensation method based on low-torque servo provided by the present invention; Figure 2 This is a structural diagram of a dynamic weight compensation device based on low-torque servo provided by the present invention.

[0015] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments.

[0018] Reference Figure 1 As shown, the present invention provides a dynamic weight compensation method based on low-torque servo, comprising: Step S1: Obtain the mechanical response data of the target device and the user's motion characteristics, perform predefined action type matching, and obtain the servo compensation parameter set; Step S2: Based on the mechanical response data, drive the main resistance unit of the target device to generate the reference resistance, and combine it with the servo compensation parameter set to perform dual-path control analysis to obtain dual-path compensation data; Step S3: Obtain the mechanical vibration data of the target equipment, suppress torque fluctuations in the dual-channel compensation data, and obtain a low torque compensation signal; Step S4: Based on the user's motion characteristics and the mechanism's vibration data, predict the mechanical state and output the low torque balance threshold; Step S5: Perform closed-loop fusion on the low torque compensation signal and the low torque balance threshold to generate global compensation information.

[0019] Based on the steps described above, the detailed process is as follows: Step S1: A triaxial force sensor and joint angle encoder deployed in the device's transmission mechanism acquire raw mechanical response data streams at a sampling rate of 2000Hz. Simultaneously, a surface electromyography (EMG) sensor and an inertial measurement unit capture the user's bioelectrical signals. Kalman filtering is applied to the raw force signals to eliminate 10-50Hz mechanical noise interference caused by transmission gaps. The EMG signals are then RMS rectified and integrated using a sliding window to convert them into temporal characteristics of muscle activation.

[0020] The processed mechanical response data and user motion characteristics are input into the motion classification engine, which uses a dynamic time warping algorithm to perform multi-dimensional similarity matching between the real-time motion trajectory and a preset motion template library. When the shoulder flexion angle of the bench press reaches its peak within 50ms and the triceps activation rate exceeds the threshold of 85%, a concentric phase compensation flag is triggered. The matching result drives the compensation parameter generator, which outputs a set of servo compensation parameters including dynamic damping coefficient, impedance gradient curve, and peak torque limit. The dynamic damping coefficient is generated proportionally to the muscle activation rate, with the parameter vector ranging from 0.15 to 0.35 N·s / m.

[0021] Step S2: The permanent magnet brushless motor of the main resistance unit receives real-time load torque commands from the mechanical response data and generates a reference resistance output through field-oriented control. This reference resistance completes closed-loop regulation of the current loop within a 150μs control cycle, and uses space vector pulse width modulation to control the torque error within ±0.8Nm. Simultaneously, the servo compensation parameter set is decoupled into a feedforward compensation path and an impedance adjustment path: the feedforward path constructs a second-order resistance model based on the dynamic damping coefficient to generate the lead compensation amount; the impedance path generates a phase compensation matrix based on the impedance gradient curve through Lyapunov stability analysis.

[0022] The dual signals are fused in a state observer, with the reference resistance serving as the main control variable. Feedforward compensation and impedance compensation are injected into the high-frequency and low-frequency components, respectively. Real-time band splitting is performed: low-frequency compensation below 20Hz is directly superimposed on the reference resistance, while high-frequency components (20-200Hz) are selectively injected through a bandpass filter. The final output includes dual-channel compensation data containing spectrum-optimized resistance. This scheme enables the equipment to dynamically switch from 5Nm centripetal resistance to 2Nm centrifugal resistance within 0.5s.

[0023] Step S3: A piezoelectric accelerometer was used to acquire the vibration time-domain waveform of the equipment's drive shaft at a sampling rate of 10kHz. A Fast Fourier Transform (FFT) was used to separate the two main resonant frequencies of 150Hz and 380Hz. A vibration feature extractor was constructed to calculate the energy proportion of each frequency and detect the resonant phase shift. Frequency domain analysis was performed on the dual-channel compensation data to identify the PWM drive spectrum components coinciding with the main resonant frequencies. An anti-resonance filter bank was established, and the zero-point position of the transfer function was designed to precisely match the 150Hz resonant point, generating a reverse suppression signal. A notch filter was used for attenuation at the frequency points. A timing alignment unit synchronized the suppression signal with the dual-channel compensation data. Vibration energy inversion calculations were performed, converting the kinetic energy of the resonant components into a potential energy compensation term, which was injected into the current loop reference value of the motor driver. In the final low-torque compensation signal, the torque fluctuation caused by the 150Hz resonance was suppressed.

[0024] Step S4: Based on the joint angular velocity time-series curves and electromyographic activation patterns in user motion characteristics, a phase-labeling model of the movement cycle is established to identify the peak point of concentric contraction and the critical point of eccentric transition. Simultaneously, the frequency domain energy distribution characteristics of the mechanism's vibration data are extracted, and the vibration modes of different stiffness segments of the transmission system are separated using wavelet packet decomposition. A motion-vibration coupling state matrix is ​​constructed to establish a nonlinear mapping relationship between joint motion trajectories and vibration energy in specific frequency bands.

[0025] A dynamic stability assessment algorithm is used to analyze the eigenvalue distribution of the matrix. When the proportion of high-frequency vibration energy exceeds a preset threshold under a specific motion phase, it is identified as a mechanical instability risk area. An initial torque safety constraint is generated based on the risk area boundary. Temperature drift compensation is performed using the servo system thermodynamic model, and friction coefficient decay is modeled and corrected based on the equipment's cumulative operating time. Finally, a dynamically adjusted low-torque balance threshold is output. This constraint value is updated in real time with the motion state and strictly limited within the safe torque range of the servo unit.

[0026] Step S5: A dual-channel signal synchronization mechanism is established to timestamp and calibrate the low-torque compensation signal with the reference resistance of the main resistance unit, eliminating control timing deviations. A boundary constraint engine performs dynamic amplitude limiting on the optimized compensation signal, ensuring its instantaneous amplitude remains within the range defined by the low-torque balance threshold. A heterogeneous signal fusion algorithm is executed, decomposing the compensation signal into translational and rotational degrees of freedom components, which are then selectively superimposed with the reference resistance at different frequency bands: arithmetic superposition is used for low-frequency resistance components, while coherent filtering is applied to high-frequency compensation components.

[0027] Phase correction is performed on the fused signal based on the vibration feedback data of the mechanism, and an adaptive phase-locked loop is used to compensate for mechanical transmission delay. Energy equalization is performed on the phase coordination signal, and local abrupt changes are eliminated through a resistance gradient smoothing algorithm. Finally, after processing by the transmission chain impedance matching unit, the output provides global compensation information that highly conforms to human biomechanical characteristics, achieving a continuous and smooth transition of the equipment's output resistance throughout the entire operation cycle.

[0028] The present invention provides a dynamic weight compensation method and device based on low-torque servo, which has the following beneficial effects: By matching user motion characteristics with equipment mechanical response data, precise parameter adaptation for predefined motion types is achieved, effectively solving the problem of decreased compensation accuracy caused by insufficient parameter matching in traditional methods, and significantly improving the accuracy of resistance control. Combining mechanism vibration data with dual-path compensation data for torque fluctuation suppression optimizes the low-torque compensation signal, effectively overcoming the shortcomings of traditional methods in vibration suppression, reducing the risk of mechanical resonance, and improving system response stability. Generating a low-torque balance threshold through mechanical state prediction and fusing it with the compensation signal in a closed loop achieves dynamic optimization of resistance compensation, avoiding overcompensation or undercompensation, and reducing energy consumption while ensuring system stability. Based on dual-path control analysis and global compensation information generation, it can adapt to nonlinear load changes, solving the problem that traditional static models cannot cope with complex working conditions, and significantly improving the system's adaptability to diverse motion scenarios. The combination of dynamic prediction and closed-loop constraint mechanisms ensures real-time coordination between the compensation process and mechanical state, effectively reducing system instability caused by lack of constraints in traditional methods and enhancing the long-term operational reliability of the equipment.

[0029] Furthermore, in one embodiment, the mechanical response data of the target device and the user's motion characteristics are acquired, and predefined action types are matched to obtain a servo compensation parameter set, including: The acquisition of mechanical response data for the target equipment relies on a strain sensor array on the surface of its transmission mechanism. This array consists of multiple high-sensitivity metal strain gauges fixed to the surfaces of key force-bearing components such as connecting rods and crankshafts in a specific mechanical layout, ensuring coverage of the main areas of the equipment subjected to tensile forces during training movements. Each strain gauge converts mechanical deformation into a weak electrical signal, which is amplified by an instrumentation amplifier to output a sequence of raw tensile force data. During acquisition, the sensors synchronously record the tensile force changes of the equipment under different operating modes at a fixed sampling rate (e.g., 1000Hz), and the data includes tensile force amplitude, frequency distribution, and phase information.

[0030] Due to noise sources such as mechanical vibration (e.g., knocking noise caused by transmission chain gaps) and electromagnetic interference (e.g., electromagnetic noise from motors) during equipment operation, the raw data undergoes time-domain denoising to extract effective mechanical features. The denoising employs an adaptive filtering algorithm, dynamically adjusting filter weights by real-time calculation of the signal autocorrelation function and noise variance. This focuses on suppressing mechanical vibration noise in the 50-200Hz frequency band while retaining the low-frequency effective load signal in the 10-50Hz range (this frequency band corresponds to the coupling frequency between human muscle contraction and equipment transmission). The processed mechanical response data removes high-frequency noise interference and retains key features reflecting the actual stress state of the equipment, providing fundamental physical quantities for subsequent action type identification and compensation parameter generation.

[0031] The acquisition of user motion characteristics of the target device relies on an inertial measurement unit (IMU) integrated into the device body or the user's contact area. This unit includes a three-axis accelerometer and a three-axis gyroscope, which output the three-axis linear acceleration and three-axis angular velocity of the device and user motion, respectively. Both are acquired synchronously with the same timestamp to form a data stream. The accelerometer data reflects the translational characteristics of the overall device motion, while the gyroscope data reflects the rotational characteristics. The combination of the two comprehensively describes the spatial motion trajectory of the user's actions.

[0032] The motion feature extraction process is based on multi-dimensional kinematic analysis. First, the acceleration signal is segmented using a sliding window, and the root mean square value within each window is calculated to characterize the acceleration intensity. Peak factors are extracted to identify impact movements. The angular velocity signal from the gyroscope is integrated to obtain the change in joint angle, and the rate of change of angle and angular acceleration are further calculated to reflect the acceleration and deceleration characteristics of the movement. A pattern recognition algorithm is used to classify the feature sequence, identifying the type of movement the user is currently performing (e.g., squat, bench press, deadlift), and recording key time points within the movement cycle (e.g., the concentric contraction initiation point, the eccentric transition critical point). The final user motion features include motion type labels, joint motion trajectories, and muscle exertion timing, providing a basis for the personalized adaptation of subsequent compensation parameters.

[0033] The pre-defined compensation parameter benchmark library is a standardized set of parameters accumulated before the equipment leaves the factory or during long-term operation. It is stored in categories according to motion type (such as pushing, pulling, squatting) and equipment model (such as training equipment of different tonnages), and includes core parameters such as benchmark resistance curves, dynamic damping coefficients, and equivalent stiffness of the transmission chain. The construction of this benchmark library is based on training with a large amount of historical motion data, covering the mechanical characteristics and compensation requirements of common training movements.

[0034] The indexing process takes the acquired mechanical response data (force amplitude, frequency distribution) and user motion features (action type labels, joint motion trajectories) as input and implements it through a feature matching algorithm. Specifically, the system compares the force frequency distribution in the mechanical response data with the typical force spectrum of each action type in the benchmark library and calculates the cosine similarity; simultaneously, it aligns the joint motion trajectory in the user motion features with the trajectory template of the corresponding action type in the benchmark library in time and calculates the trajectory overlap. When the force similarity exceeds a set threshold and the trajectory alignment error is less than 50ms, the current action is determined to belong to the predefined type. The indexing result generates a predefined action type (such as "squat concentric" and "bench press eccentric"), which is directly associated with the corresponding compensation parameter group in the benchmark library, providing a type basis for subsequent conversion.

[0035] A dual operation of physical quantity conversion and dynamic compensation is employed to establish an energy conversion model between actual motion and equipment transmission characteristics. Based on predefined motion types, standardized parameter modules from the compensation parameter benchmark library are invoked: for "squat concentric" type motions, benchmark equivalent mass and viscous damping coefficient are extracted; for "bench press eccentric" type motions, reference values ​​for rotational inertia and elastic recovery coefficients are obtained. Simultaneously, the inherent characteristic parameters of the transmission chain calibrated at the equipment's factory are loaded, including the transmission system stiffness and friction loss feature set of the moving pairs.

[0036] The reference inertia compensation conversion operation for mechanical response data involves inputting the raw tension data collected by the strain sensor array into the mass acceleration conversion model, dividing it by the pre-stored equivalent mass parameter to obtain the acceleration sequence per unit mass, and then performing a scalar multiplication operation between this acceleration sequence and the reference damping coefficient for the corresponding motion type to generate the basic resistance compensation amount. This process employs iterative calculation using the equation of motion to ensure the energy conservation relationship between the compensation amount and the actual motion state.

[0037] The equivalent compensation conversion operation of the transmission chain for user motion characteristics involves: analyzing the joint angle change rate collected by the inertial measurement unit, calculating the elastic potential energy generated by joint motion using the Hooke's Law elastic deformation model (elastic force = equivalent stiffness × joint angle change), and combining it with the reference elastic coefficients (linear stiffness parameters related to the motion type) to generate an additional resistance compensation amount to correct the resistance deviation caused by joint deformation. The potential energy value is input into a preset transmission chain stiffness matching function and convolved with the elastic coefficients in the reference library to output the additional resistance compensation amount. Simultaneously, abrupt muscle force exertion points in the motion characteristics are detected, triggering a friction loss compensation mechanism to generate an additional friction compensation component.

[0038] In the time domain, vector superposition is performed on the basic resistance compensation, additional resistance compensation, and friction compensation components to output a servo compensation parameter set containing the dynamic damping coefficient, elastic compensation gradient, and time-varying friction coefficient. This parameter set is calibrated and certified to strictly match the physical characteristics of the equipment's drivetrain, ensuring that the compensated resistance curve is completely synchronized with the user's intended action.

[0039] This embodiment effectively extracts the true tension characteristics of the equipment's transmission mechanism by combining a strain sensor array with time-domain noise reduction processing, reducing the impact of mechanical vibration and electromagnetic interference on the data and providing a more accurate mechanical basis for subsequent action type recognition. The triaxial acceleration and angular velocity data synchronously collected by the inertial measurement unit, after action feature extraction, can accurately capture the changes in joint angles and the timing of force application during user movement, solving the problem of large errors in user motion feature recognition in traditional methods. Based on the parameter conversion of the reference inertia and equivalent compensation of the transmission chain, a dynamic compensation parameter set is generated in combination with the actual transmission characteristics of the equipment. This effectively offsets the influence of transmission chain inertia and joint deformation on resistance, ensuring that the compensated resistance in low-torque scenarios is highly consistent with the user's action intention, thus improving the accuracy and reliability of dynamic weight compensation.

[0040] Furthermore, in one embodiment, the main resistance unit of the target device is driven to generate a reference resistance based on the mechanical response data, and dual-path control analysis is performed in conjunction with the servo compensation parameter set to obtain dual-path compensation data, including: The core mechanical features related to the mechanical response data and movement patterns are extracted. Frequency domain energy spectrum separation is performed on the mechanical response data, and the time domain signal is converted into a frequency domain distribution using Fast Fourier Transform. The signal is divided into multiple sub-frequency bands by setting a frequency resolution. The energy density (the ratio of energy to frequency bandwidth) of each sub-frequency band is calculated, and effective frequency bands related to human movement (usually 10-500Hz, covering the coupling frequency of muscle contraction and equipment transmission) are selected.

[0041] After removing high-frequency noise (>500Hz) and low-frequency drift (<10Hz), the energy density and center frequency of each effective sub-band are used as characteristic parameters and arranged in a time series to form a dynamic mechanical characteristic vector. This vector fully preserves the frequency domain characteristics of the device's mechanical response under different operating modes, providing a precise frequency dimension basis for subsequent magnetic circuit output adjustment.

[0042] The electromagnetic output characteristics of the main resistance unit are affected by the magnetic circuit material, current excitation, and mechanical structure, exhibiting significant nonlinear characteristics (such as magnetic saturation effect and eddy current loss). The frequency and energy information in the dynamic mechanical characteristic vector directly reflect the target action's resistance requirements (e.g., high rigidity resistance is required in the centripetal phase, and elastic buffer resistance is required in the centrifugal phase). To achieve a high degree of matching between the magnetic circuit output and the requirements, a nonlinear mathematical model of the magnetic circuit needs to be established: using current as the input variable and magnetic circuit output force as the output variable, a nonlinear function relationship is obtained by fitting historical calibration data. Based on typical frequency points in the dynamic mechanical characteristic vector (e.g., centripetal peak frequency and centrifugal valley frequency) as approximation nodes, the parameters of the nonlinear model are optimized using the least squares method or polynomial interpolation algorithm, so that the error of the magnetic circuit output force at the target frequency point is less than ±0.2Nm. The optimized magnetic circuit parameter set is the reference resistance distribution set, containing the reference resistance values ​​corresponding to each frequency under different action modes (e.g., 8Nm for 50Hz centripetal squat and 3Nm for 150Hz centrifugal bench press), providing a standard reference for subsequent dual-path control analysis.

[0043] The servo compensation parameter set contains multi-dimensional parameters related to the device's motion degrees of freedom, which need to be decomposed into independent translational and rotational control inputs through a kinematic orthogonal decoupling mechanism. This decoupling mechanism, based on the kinematic model of the device's transmission mechanism, decomposes the output of the main resistance unit into translational components along the motion trajectory (such as linear push / pull forces) and rotational components around the joint axis (such as torques). In practice, a kinematic coordinate system is established for the device, defining the translational direction (such as the x-axis) and the rotational axis (such as the z-axis). The generalized forces / torques in the compensation parameter set are then converted into translational and rotational control inputs using the Jacobian matrix.

[0044] For example, the damping coefficients in the compensation parameter set are decomposed into translational damping and rotational damping according to the direction of motion, and the stiffness coefficients are decomposed into translational stiffness and rotational stiffness. The decomposed translational control input group contains a force command sequence along the x-axis, and the rotational control input group contains a torque command sequence around the z-axis. The two are orthogonal and uncoupled, laying the foundation for subsequent two-degree-of-freedom independent control.

[0045] The reference resistance distribution set reflects the standard resistance requirements of the target motion, and the two-degree-of-freedom control input set provides independent adjustment dimensions for the control commands. The dynamic interactive correction process is achieved through the synergy of feedforward control and feedback control: feedforward control predicts the output of the main resistance unit based on the reference resistance distribution set and generates initial control commands; feedback control adjusts the parameters of the control input set (such as the gain coefficient of translational damping and the compensation amount of rotational stiffness) according to the deviation between the actual output of the main resistance unit and the predicted value.

[0046] The specific correction logic is as follows: when the error between the actual resistance and the reference resistance exceeds ±0.1 Nm, feedback adjustment is triggered. The parameters of the control input group are updated through the proportional-integral (PI) controller, so that after the feedforward command and the feedback correction are superimposed, the matching degree between the total control output and the reference resistance is improved to over 95%. The final hybrid control protocol includes a force command sequence for translation control and a torque command sequence for rotation control. The two are synchronized through timestamps to ensure the consistency of control timing.

[0047] The power drive modulation signal of the main resistance unit needs to convert the hybrid control protocol into specific electrical signal parameters. The reconfiguration process is based on the characteristics of the main resistance unit's drive circuit, using pulse width modulation (PWM) technology to convert the control command into a duty cycle signal, while ensuring the stability of the output torque through current loop regulation.

[0048] The force command sequence for translation control is modulated by PWM to generate a duty cycle signal, with the duty cycle linearly related to the force magnitude (e.g., a 50% duty cycle corresponds to 5 Nm). The torque command sequence for rotation control is converted by an angle-to-current conversion module to generate a current reference value, with the current value proportional to the torque magnitude (e.g., 1A current corresponds to 2 Nm torque). The final output dual-channel compensation data includes the PWM duty cycle sequence for translation and the current reference sequence for rotation. Both are output synchronously through the drive circuit to achieve low-torque dynamic compensation for the main resistance unit, ensuring that the resistance output is highly consistent with the user's intended action.

[0049] This embodiment effectively filters out mechanical vibration and electromagnetic interference noise through temporal denoising of mechanical response data, while retaining the tensile characteristics that reflect the actual stress state of the equipment, providing a more accurate mechanical basis for subsequent action type recognition. The triaxial acceleration and angular velocity data synchronously collected by the inertial measurement unit, after action feature extraction, can accurately capture the joint angle changes and force sequence of the user's movement, solving the problem of large errors in user motion feature recognition in traditional methods. The combination of a preset compensation parameter benchmark library and a predefined type index enables rapid matching of action types and compensation parameters, avoiding compensation deviations caused by misjudgment of action types and improving the adaptability of compensation parameters to actual actions. Based on the parameter conversion of benchmark inertia and equivalent compensation of the transmission chain, a dynamic compensation parameter set is generated in conjunction with the actual transmission characteristics of the equipment, effectively offsetting the influence of transmission chain inertia and joint deformation on resistance, ensuring that the compensated resistance in low-torque scenarios is highly consistent with the user's action intention. The synergistic effect of steps such as frequency domain energy spectrum separation and nonlinear approximation, kinematic decoupling and dynamic interactive correction improves the fit between magnetic circuit output and action requirements, enhances control flexibility and precision, and ultimately achieves efficient and stable execution of low torque dynamic weight compensation.

[0050] In one embodiment, acquiring the mechanical vibration data of the target device and performing torque fluctuation suppression on the dual-channel compensation data to obtain a low-torque compensation signal includes: The acquisition of vibration data relies on accelerometers on the surface of the equipment's transmission mechanism. These sensors synchronously record the triaxial vibration time-domain signals during equipment operation at a fixed sampling rate. The raw vibration signals contain broadband mechanical noise (such as knocking from transmission chain gaps and bearing rolling sounds) and vibration components related to the target motion. To extract vibration features directly related to low torque compensation, spectral analysis of the time-domain signal is required: a Fast Fourier Transform is used to convert the time-domain signal into a frequency domain distribution, and the signal is divided into multiple sub-frequency bands by setting a frequency resolution.

[0051] The energy density (the ratio of energy to frequency bandwidth) of each sub-band is calculated, and the sub-bands are sorted by energy percentage. The primary resonant mode components (typically corresponding to the natural frequencies of the equipment's transmission chain, such as 150Hz and 380Hz) with an energy percentage exceeding 80% are selected, along with the secondary resonant mode components (such as non-natural frequency components like 50Hz and 200Hz) with a lower energy percentage. During the separation process, high-frequency noise (>500Hz) and low-frequency drift (<50Hz) are simultaneously removed, retaining the effective vibration frequency band of 10-500Hz. The resulting primary and secondary resonant mode components are stored as frequency domain signals, providing fundamental data for phase compensation of the primary resonant mode and energy suppression of the secondary resonant mode.

[0052] The dual-channel compensation data includes a reference resistance signal generated based on the servo compensation parameter set. This signal may experience phase lag due to control delay or mechanical inertia when driving the main resistance unit, resulting in a phase difference with the vibration signal of the main resonant mode. The goal of phase lag compensation is to adjust the phase of the compensation signal so that it is opposite to the vibration phase of the main resonant mode, thereby canceling the resonance effect. In practice, the phase information of the main resonant mode component is first extracted (the instantaneous phase is calculated using Hilbert transform). Then, the phase of the reference resistance signal from the dual-channel compensation data is adjusted so that the phase difference between the adjusted compensation signal and the main resonant mode is close to 180° (out of phase).

[0053] During the adjustment process, the phase compensation amount needs to be calculated based on the mechanical characteristics of the equipment's transmission chain (such as transmission ratio and stiffness) to ensure phase matching between the compensation signal and the vibration signal. When the phase difference between the compensation signal and the main resonant mode is adjusted to a critical value (e.g., ±5°), the vibration energy of the main resonant mode is significantly suppressed. At this point, the peak frequency point of the main resonant mode (i.e., the critical oscillation frequency point) is recorded, forming a set containing multiple critical frequencies (e.g., {150Hz, 380Hz}). This set is directly used for the envelope constraint of subsequent driving waveforms to clearly identify the resonant frequency bands that need to be suppressed.

[0054] Each frequency in the set of critical oscillation frequencies corresponds to the peak position of the main resonant mode. The vibrational energy is most concentrated in these frequency bands, making them prone to servo system saturation (output torque exceeding the rated value) due to overshoot of the drive waveform. The core of the drive waveform envelope boundary constraint is to prevent waveform distortion due to resonance by limiting the amplitude of the compensation signal at these frequency points. A bandpass filter bank is generated based on the set of critical oscillation frequencies, with each filter's center frequency corresponding to a critical frequency, and its bandwidth covering a ±10Hz range around that frequency.

[0055] The reference resistance signal of the dual-channel compensation data is input into the filter bank to detect the amplitude of the sub-signals at each critical frequency point, and an upper limit for the amplitude is set (e.g., ±4.8 Nm, corresponding to the safety threshold of low-torque servo). When the amplitude of a sub-signal at a certain frequency point exceeds the upper limit, it is limited to a safe range by a limiter; if the amplitude does not exceed the upper limit, the original signal is maintained. The processed sub-signals are then combined by inverse transformation of the bandpass filter bank to obtain the anti-saturation compensation signal. The amplitude of this signal is strictly constrained near the critical oscillation frequency point, avoiding performance degradation of the servo system due to resonance overshoot, while retaining effective compensation components in other frequency bands, providing a stable input basis for energy suppression and active damping injection of subsequent sub-resonances.

[0056] Secondary resonant mode components are frequency bands in the vibration of a mechanism whose energy proportion is lower than that of the primary resonant mode (such as non-transmission chain natural frequencies like 50Hz and 200Hz). Although their vibration energy is weak, they are prone to high-frequency oscillations due to local resonance of the mechanical structure or control delay, affecting the compensation accuracy in low-torque scenarios. The anti-saturation compensation signal has completed the phase lag compensation and amplitude constraint of the primary resonant mode, but the energy of the secondary resonant mode still needs to be suppressed by active damping.

[0057] The core of coupled resonance energy inversion calculation is to correlate the energy characteristics of the sub-resonant mode with the dynamic characteristics of the anti-saturation compensation signal, and calculate the damping coefficient that needs to be injected to counteract the influence of the sub-resonance. In practice, the energy time-series characteristics of the sub-resonant mode components (such as the energy density change curve over time) are extracted, and their phase correlation with the anti-saturation compensation signal is analyzed (calculated through cross-correlation function).

[0058] An inversion model is established based on the principle of energy conservation: Sub-resonance energy = Anti-saturation compensation energy + Active damping energy. By adjusting the amplitude and phase of the active damping energy in real time, the sub-resonance energy is effectively absorbed. This process requires modifying the inversion parameters by considering the damping characteristics of the equipment's transmission chain (such as material damping ratio and structural damping coefficient), ultimately obtaining an active damping injection coefficient that matches the frequency and energy of the sub-resonance mode. This coefficient is directly used for damping adjustment of the subsequent pulse width modulation signal.

[0059] The dual-channel compensation data includes a reference resistance signal generated based on the servo compensation parameter set. This signal has been analyzed and dynamically corrected through dual-channel control, possessing basic resistance characteristics that match the user's intended actions. The active damping injection coefficient provides a basis for damping adjustment for high-frequency vibrations in the sub-resonant mode. The goal of pulse width modulation depth signal reconstruction is to combine the two to generate a final compensation signal that balances low torque characteristics and vibration suppression.

[0060] The reference drag signal from the dual-channel compensation data is converted into a pulse-width modulated duty cycle reference value. The dynamic range of the duty cycle is adjusted according to the active damping injection coefficient: for the sub-resonant frequency band, high-frequency oscillations are suppressed by reducing the fluctuation amplitude of the duty cycle; for the non-sub-resonant frequency band, a smooth transition of the duty cycle is maintained to ensure drag continuity. The adjusted duty cycle signal is smoothed by a low-pass filter (cutoff frequency 500Hz) to eliminate high-frequency switching noise, and finally outputs a low-torque compensation signal. The amplitude of this signal is strictly limited within the safety threshold of low-torque servo (e.g., ≤5Nm), and vibration energy is effectively suppressed, achieving smoothness and stability of drag output in low-torque scenarios.

[0061] This embodiment effectively extracts the main and secondary resonant mode components through spectral feature separation of the vibration data, providing accurate vibration characteristic basis for subsequent targeted compensation and improving the resolution and reliability of vibration energy analysis. Phase lag compensation of the main resonant mode significantly suppresses torque fluctuations caused by the main resonance by adjusting the phase of the compensation signal to be superimposed in opposite directions with the vibration phase, reducing the risk of performance degradation of the servo system due to resonance overshoot. Boundary constraints on the drive waveform envelope limit the amplitude of the critical oscillation frequency point, avoiding servo saturation caused by resonance distortion of the compensation signal and ensuring the stability of the output torque in low-torque scenarios. The coupled resonant energy inversion calculation of the secondary resonant mode and the anti-saturation compensation signal accurately matches the amplitude and phase of the secondary resonant energy. Active damping injection effectively absorbs high-frequency vibration energy, reducing the impact of local resonance of the mechanical structure on compensation accuracy. Finally, pulse width modulation depth signal reconstruction combines the reference resistance with active damping, generating an optimized compensation signal that maintains good smoothness and continuity in the low-torque range, achieving efficient and stable execution of dynamic weight compensation.

[0062] Furthermore, in one embodiment, the sub-resonant mode components are coupled with the anti-saturation compensation signal to perform resonant energy inversion calculation to obtain the active damping injection coefficient, including: Secondary resonant modal components are non-natural frequency components in the vibration data of a mechanism whose energy proportion is lower than that of the main resonant mode (such as 50Hz, 200Hz, and other non-transmission chain natural frequencies). Although their vibration energy is weak, they are prone to high-frequency oscillations due to local resonance of the mechanical structure or control delay, affecting the compensation accuracy in low-torque scenarios. The core of energy spectrum focusing extraction is to screen out characteristic frequency bands with locally concentrated energy through frequency domain analysis. The secondary resonant modal components are subjected to frequency domain transformation (such as fast Fourier transform) to convert the time-domain vibration signal into a frequency domain distribution; then the energy density (the ratio of energy to frequency bandwidth) of each sub-frequency band is calculated and sorted according to energy proportion; finally, peak points with energy density significantly higher than the surrounding frequency bands are screened out, and the frequencies corresponding to these peak points constitute the set of resonant peak frequencies. This set clarifies the frequency bands that need to be focused on in the secondary resonance, providing a target frequency basis for subsequent anti-phase compensation and ensuring that the compensation energy is accurately applied to the high-frequency vibration region.

[0063] The anti-saturation compensation signal has completed phase lag compensation and amplitude constraint for the main resonant mode, but its phase relationship with the secondary resonant mode still needs to be adjusted to achieve effective suppression. The goal of driving phase difference mapping is to generate a reference sequence for anti-phase compensation by calculating the phase difference between the secondary resonant mode component and the anti-saturation compensation signal. The instantaneous phase of the secondary resonant mode component is extracted to obtain the phase information of the anti-saturation compensation signal; then the phase difference between the two is calculated to determine the phase offset that needs to be compensated in anti-phase; finally, the phase of the anti-saturation compensation signal is adjusted according to this phase offset to generate an anti-phase compensation sequence with the phase opposite to that of the secondary resonant mode. This sequence provides a phase reference for subsequent dual-domain damping calculations, ensuring that the damping energy is in phase opposite to the secondary resonant vibration, thus achieving effective cancellation.

[0064] The set of resonant peak frequencies clarifies the sub-resonant frequency bands that need to be suppressed, and the anti-phase compensation sequence provides a phase reference. Dual-domain damping correlation calculations require combining both to determine the damping parameters for each frequency band. At the characteristic resonant peak frequency points, the energy amplitude of the sub-resonant mode and the phase matching degree with the anti-phase compensation sequence are analyzed. The initial value of the damping coefficient is adjusted according to the energy amplitude (e.g., the higher the energy, the larger the initial damping coefficient), and the phase shift of the damping is corrected according to the phase matching degree (e.g., adjusting the damping sign when the phase difference is >5°). An initial anti-phase damping function set containing the damping coefficients at each characteristic frequency point is generated. This function set provides preliminary damping parameters for subsequent energy inversion calculations, ensuring a high degree of matching between the active damping energy and the sub-resonant vibration in the frequency domain and phase, thus improving suppression efficiency.

[0065] The initial anti-phase damping function set contains damping parameters for each characteristic resonant peak frequency. However, coupling effects exist between multi-band damping (such as high-frequency damping interfering with low-frequency vibrations). Orthogonal decoupling eliminates the mutual influence between multi-band damping. Centered on the characteristic resonant peak frequency, the initial damping function is projected onto the orthogonal basis function space, separating the independent damping components for each frequency band. Combining the actual damping characteristics of the equipment's transmission chain (such as material damping ratio and structural damping coefficient), the gain of the orthogonally decoupled damping components is corrected: for frequency bands with high energy proportions, the damping gain is appropriately increased to enhance the suppression effect; for frequency bands with low energy proportions, the gain is reduced to avoid mechanical stiffness caused by excessive damping. The corrected damping components are recombined into a gain-damping function set, which realizes independent control and coordinated optimization of multi-band damping, ensuring that the damping energy of each frequency band accurately acts on the target vibration region and avoiding energy crosstalk between frequency bands.

[0066] The gain-damping function set clearly defines the required damping energy distribution at each characteristic resonant peak frequency. However, in actual compensation, energy absorption requires active damping injection. The core of energy inversion calculation is to determine the amplitude and phase of the damping injection based on the matching relationship between the energy characteristics of the sub-resonant mode and the damping function. The total energy of the sub-resonant mode and the proportion of energy in each characteristic frequency band are calculated. Combined with the damping gain of the gain-damping function set, the required active damping energy for each frequency band is derived. According to the principle of energy conservation, the active damping energy is converted into a damping injection coefficient (the coefficient is directly proportional to the energy and inversely proportional to the vibration frequency). By adjusting the dynamic range of the damping injection coefficient in real time (e.g., limiting the maximum injected energy to not exceed the rated value of the servo system), it is ensured that the active damping effectively suppresses sub-resonant vibration without exceeding the safe operating boundary of the equipment. The final active damping injection coefficient is directly used for damping adjustment of the subsequent pulse width modulation signal, achieving precise absorption of sub-resonant energy and ensuring stability of low-torque compensation.

[0067] Although the initial anti-phase damping function set already contains the damping parameters for each characteristic resonant peak frequency, coupling effects may exist between the damping of multiple frequency bands (such as the interference of high-frequency damping on low-frequency vibrations). Orthogonal decoupling is necessary to eliminate the mutual influence between frequency bands. Using the Gram-Schmidt orthogonalization method, the initial damping function is projected onto the orthogonal basis function space, centered on the characteristic resonant peak frequency, to separate the independent damping components for each frequency band. Subsequently, considering the actual damping characteristics of the equipment's transmission chain (such as material damping ratio and structural damping coefficient), the gain of the orthogonally decoupled damping components is corrected: for frequency bands with high energy proportions, the damping gain is appropriately increased to enhance the suppression effect; for frequency bands with low energy proportions, the gain is reduced to avoid mechanical stiffness caused by excessive damping.

[0068] The corrected damping components are recombined into a set of gain damping functions. This set of functions enables independent control and coordinated optimization of multi-frequency band damping, ensuring that the damping energy of each frequency band is accurately applied to the target vibration region and avoiding energy crosstalk between frequency bands.

[0069] The gain-damping function set clearly defines the required damping energy distribution at each characteristic resonant peak frequency, but in actual compensation, energy absorption needs to be achieved through active damping injection. The core of energy inversion calculation is to determine the amplitude and phase of the damping injection based on the matching relationship between the energy characteristics of the sub-resonant mode and the damping function.

[0070] The total energy of the subresonant mode and the proportion of energy in each characteristic frequency band are calculated. Combined with the damping gain of the gain-damping function set, the required active damping energy for each frequency band is derived. Based on the principle of energy conservation, the active damping energy is converted into a damping injection coefficient (the coefficient is directly proportional to the energy and inversely proportional to the vibration frequency). By adjusting the dynamic range of the damping injection coefficient in real time (e.g., limiting the maximum injected energy to not exceed the servo system's rated value), it is ensured that active damping effectively suppresses subresonant vibration without exceeding the equipment's safe operating boundaries. The final active damping injection coefficient is directly used for damping adjustment of the subsequent pulse width modulation signal, achieving precise absorption of subresonant energy and stable low-torque compensation.

[0071] This embodiment accurately locates the key frequency band of high-frequency vibration by focusing and extracting the energy spectrum of the sub-resonant mode components, providing a clear energy suppression target for subsequent compensation and avoiding energy waste caused by indiscriminate suppression. Driving phase difference mapping, by adjusting the phase of the compensation signal to be superimposed in opposite directions with the phase of the sub-resonant vibration, significantly reduces the energy transfer efficiency of high-frequency vibration and reduces abnormal vibrations in the mechanical structure caused by local resonance. Orthogonal decoupling eliminates the coupling effect between multi-band dampers, ensuring that the damping energy of each band acts independently on the target region, thus improving the accuracy of compensation.

[0072] Furthermore, in one embodiment, 6, the dynamic weight compensation method based on low-torque servo according to claim 1, is characterized in that, based on user motion characteristics and mechanism vibration data, mechanical state prediction is performed, and a low-torque balance threshold is output, including: The acquisition of user motion characteristics relies on the inertial measurement unit (IMU) of the device body or moving contact parts. This unit synchronously outputs triaxial acceleration and angular velocity data streams, reflecting the translational and rotational characteristics of the user's movements. The core of periodic motion phase segmentation is to identify the repetition cycle of the motion and the characteristic points of each stage through time series analysis. In specific implementation, the acceleration and angular velocity data are first segmented into sliding windows, with the window length set to 1 / 2-2 / 3 of the motion cycle (e.g., when the squat motion cycle is 2 seconds, the window length is 1-1.3 seconds). The root mean square (RMS) value and peak factor (peak value / RMS) within each window are calculated to identify the starting point of the motion (RMS rises sharply from low), the peak point of concentric contraction (RMS reaches its maximum value), and the critical point of eccentric transition (RMS drops sharply from high). Subsequently, the motion cycle duration is determined based on the time interval between adjacent characteristic points, and each cycle is divided into concentric, eccentric, and other sub-stages. A unique motion state label is assigned to each sub-stage (e.g., "squat concentric" "bench press eccentric"). The final generated motion state label sequence is arranged in chronological order and includes motion type, stage label and corresponding timestamp, providing a time dimension benchmark for subsequent correlation analysis with mechanical vibration energy.

[0073] Vibration data is collected by accelerometers on the surface of the equipment's transmission mechanism, outputting a time-domain signal containing a wideband vibration range of 10-500Hz. The goal of dominant frequency band energy feature extraction is to identify key frequency bands where vibration energy is concentrated, providing frequency-domain feature data for mechanical state analysis. Specifically, the time-domain vibration signal is first converted to a frequency-domain distribution using a Fast Fourier Transform (FFT). Then, the energy density of each sub-band is calculated and sorted by energy percentage. Dominant frequency bands with an energy percentage exceeding 80% of the total energy are selected, and their center frequency and energy percentage are recorded. Simultaneously, energy in low-frequency and high-frequency bands is suppressed, retaining only the energy distribution characteristics of the dominant frequency bands. The resulting mechanical concentration frequency band information is presented as a frequency-energy curve, clearly identifying the main distribution areas of vibration energy under different operating modes, providing frequency-domain feature input for the subsequent construction of dynamic interactive state mapping.

[0074] The construction of a dynamic interactive state mapping relationship aims to establish a correlation between the user's motion state and the distribution of mechanical vibration energy, reflecting the energy characteristics of mechanical vibration at different action stages. In specific implementation, firstly, each timestamp in the motion state tag sequence is time-aligned with the frequency-energy curve of the mechanical aggregation frequency band information, ensuring that the action stage and vibration energy data are synchronized in the time dimension. Subsequently, for each motion state (e.g., "squat concentric"), its corresponding vibration energy data (e.g., the energy proportion of the 150Hz frequency band) is extracted, and the amplitude, frequency distribution, and phase characteristics of the vibration energy in that state are calculated. A mapping table between motion states and vibration energy is established through statistical analysis (e.g., mean, variance), where the key is the motion state label (e.g., "squat concentric"), and the value is the corresponding vibration energy feature vector (containing parameters such as dominant frequency band, energy proportion, and phase shift). This mapping relationship reflects the energy distribution law of mechanical vibration at different action stages, providing a state-energy correlation basis for subsequent quantitative stability assessment, ensuring that the assessment results highly match the actual motion scenario.

[0075] The dynamic interactive state mapping establishes a correlation between the user's motion state and mechanical vibration energy, but its stability needs to be quantitatively assessed to determine the boundaries of safe system operation. Key characteristic parameters are extracted from the dynamic interactive state mapping, including the vibration energy amplitude corresponding to the motion state, the energy proportion of the dominant frequency band, and phase deviation. Based on these parameters, stability assessment indicators are constructed, such as energy volatility (variance of energy amplitude), phase consistency (mean of phase difference between adjacent periods), and frequency band energy concentration (standard deviation of the energy proportion of the dominant frequency band).

[0076] By statistically analyzing the distribution patterns of these indicators, the critical conditions for the system to transition from stability to instability are determined: when the energy fluctuation rate exceeds a certain threshold (such as twice the standard deviation of the mean), the phase deviation exceeds ±10°, or the energy concentration in the frequency band is below 85%, the system enters the unstable region. Finally, the motion states and vibration energy characteristics corresponding to these critical conditions are used as boundaries to form the critical working region boundary, which clearly defines the safe fluctuation range of mechanical vibration energy under different action stages.

[0077] The critical operating region boundary defines the safe fluctuation range of mechanical vibration energy, but it needs to be converted into torque constraint parameters controllable by the servo system. A mapping relationship between vibration energy and servo torque is established: through training with historical data or theoretical calculations, the linear or nonlinear relationship between vibration energy amplitude and servo output torque is determined (e.g., for every 10% increase in energy amplitude, the torque needs to increase by 0.5 Nm to maintain stability). Based on the energy fluctuation range of the critical operating region boundary, the corresponding torque safety margin is calculated: for example, if the upper limit of the energy amplitude in the critical region is Emax, then the corresponding upper limit of torque is Tmax = Emax × conversion coefficient. Based on the safety threshold of the low-torque servo (e.g., 80% of the rated torque), the upper limit of torque is corrected to ensure it does not exceed the safe operating range of the equipment. The finally generated dynamic safe torque constraint parameters include the upper and lower limits of torque and the allowable fluctuation range. These parameters are directly used to generate subsequent balance threshold constraints, ensuring that the compensated resistance is within the safe range and matches the mechanical state.

[0078] While the dynamic safety torque constraint parameters have a clearly defined safety range, they need to be adapted and adjusted based on the actual motion characteristics of the equipment and user needs to form the final balance threshold constraint. The equivalent stiffness and damping coefficient of the equipment's transmission chain, as well as the force application patterns in the user's motion characteristics (such as the frequency and amplitude of explosive force output), are obtained. Based on these parameters, the dynamic safety torque constraint parameters are modified: for high-stiffness transmission chains, the upper limit of torque is increased to accommodate rapid resistance changes; for low-damping systems, the torque fluctuation range is reduced to suppress high-frequency oscillations. The time dependence of the constraint parameters is adjusted by considering key time points in the user's motion characteristics (such as the initiation point of centripetal contraction) to ensure that the torque constraint remains stable during critical phases of the action (such as peak force). The resulting low-torque balance threshold includes upper and lower limits of torque dynamically adjusted with the motion state, as well as fluctuation restrictions. This constraint is directly used for resistance output control in subsequent closed-loop fusion, achieving smoothness and safety of resistance output in low-torque scenarios.

[0079] This embodiment accurately identifies the periodic phases of user actions through periodic action phase division, providing a time-dimensional benchmark for mechanical state analysis and avoiding compensation deviations caused by misjudgment of action phases. Dominant frequency band energy feature extraction clarifies the concentrated frequency bands of the mechanism's vibration energy, providing key frequency-domain features for dynamic interactive state mapping and ensuring that correlation analysis focuses on the region with the most concentrated energy. Quantitative stability assessment, through statistical analysis of parameters such as energy fluctuation and phase consistency, determines the critical boundary for safe system operation, providing a clear instability warning standard for subsequent torque constraint generation. Low torque safety margin conversion correlates vibration energy with servo torque, converting the energy fluctuation range into a controllable torque range, ensuring that the compensated resistance is within the safety threshold and matches the mechanical state. Adaptability matching of dynamic safety torque constraint parameters combines equipment transmission characteristics and user motion characteristics, correcting the actual applicability of the constraint parameters and avoiding constraint failures caused by differences in mechanical stiffness or user force application patterns.

[0080] Furthermore, in one embodiment, adaptive matching of the dynamic safety torque constraint parameters is performed to obtain a low torque balance threshold, including: Dynamic safety torque constraint parameters describe the maximum allowable torque range of a device under different operating conditions. Their physical meaning lies in limiting the amplitude of the servo system's output torque to prevent mechanical damage or control failure due to overload. However, the torque experienced by the device in actual operation is not unidirectional but includes an axial component along the motion trajectory (such as a push / pull force) and a radial component perpendicular to the motion trajectory (such as a lateral offset force). To accurately reflect the torque constraint requirements in different directions, the dynamic safety torque constraint parameters need to be decomposed into a multi-dimensional space.

[0081] An orthogonal coordinate system is established with the equipment's motion axis as the reference, defining the axial direction as the direction of motion and the radial direction as the plane perpendicular to the motion axis. Then, the dynamic safety torque constraint parameters (usually expressed as the upper limit of the torque amplitude) are projected onto this coordinate system, and the axial and radial components are separated through mathematical transformations (such as coordinate rotation or principal component analysis). The axial component reflects the torque constraint along the direction of motion, and its value is determined by the axial stiffness and safety factor of the equipment's drivetrain; the radial component reflects the torque constraint in the vertical direction, and is determined by the radial stiffness and lateral force resistance of the drivetrain. The two decomposed components are stored as independent constraint parameters, providing directional dimension feature inputs for subsequent coupling relationship analysis.

[0082] Although the axial and radial torque constraint components have different directions, they may exhibit coupling effects in actual motion. For example, when a large torque is applied to the equipment in the axial direction, an additional torque may be generated in the radial direction due to the elastic deformation of the transmission chain; conversely, radial offset force may cause fluctuations in the axial torque. To quantify this coupling relationship, it is necessary to analyze the correlation between the two components in the time domain or frequency domain.

[0083] Time-series data of axial and radial torque constraint components are extracted, their cross-correlation functions are calculated, and their temporal synchronicity is assessed. Frequency domain analysis (such as FFT) is used to calculate their coherence functions, determining the strength of coupling at different frequencies. Based on these analyses, a dimensionless parameter (e.g., a value between 0 and 1, where 1 represents complete coupling and 0 represents no coupling) is defined as the torque component correlation coefficient to describe the degree of coupling. This correlation coefficient reflects the interaction strength between axial and radial torque constraints, providing a quantitative basis for subsequent weighted fusion and ensuring that the comprehensive constraint vector accurately reflects the collaborative constraint requirements in both directions.

[0084] Axial and radial torque constraint components describe torque limitations in different directions, but in practical applications, they need to be integrated into a single comprehensive constraint vector to fully guide the torque control of the servo system. The core of weighted fusion is to adjust the weights based on the correlation coefficient between the two components, so that components with a high degree of coupling occupy a larger proportion in the comprehensive vector.

[0085] The weighting coefficients for the axial and radial torque constraint components are determined, with the weights proportional to the correlation coefficients. The magnitude of the comprehensive dynamic torque constraint vector is calculated using linear weighting: Comprehensive magnitude = Axial component × Axial weight + Radial component × Radial weight. The phase information of both components is preserved, and the direction of the comprehensive constraint vector is obtained through vector synthesis. This vector not only contains the maximum allowable magnitudes of the axial and radial torques but also reflects their phase coordination, providing complete constraint characteristics for subsequent boundary matching calculations and ensuring that the torque output by the servo system does not exceed the safety threshold in any direction.

[0086] The preset torque feedback standard data represents the safe torque range verified by the equipment during long-term operation. It includes upper and lower limits of torque under different operating modes and load conditions, and is typically stored in time series or frequency domain distribution format. The time series data of the integrated dynamic torque constraint vector is aligned with the corresponding time periods of the standard data to ensure that the operating phase (e.g., centripetal, centrifugal) and torque constraints are synchronized in the time dimension. Subsequently, the amplitude of the integrated constraint vector is compared point-by-point with the allowable range of the standard data: if the integrated amplitude is within the standard range (i.e., between the preset upper and lower limits), it is marked as "matched"; if it exceeds the upper limit or falls below the lower limit, it is marked as "mismatched".

[0087] The boundary matching degree is calculated by statistically analyzing the proportion of matching points to the total number of points (e.g., matching degree = number of matching points / total number of points × 100%). Simultaneously, the specific deviation values ​​(comprehensive amplitude - standard upper or lower limit) of all mismatched points are recorded. These deviation values ​​are the torque constraint deviations, reflecting the degree of deviation between the comprehensive constraint vector and the safety standard. This deviation directly indicates the gap between the current constraint and actual safety requirements, providing a quantitative basis for subsequent threshold adjustments.

[0088] Torque constraint deviations include positive deviations (overall amplitude exceeding the upper limit) and negative deviations (overall amplitude below the lower limit). These need to be corrected to ensure the constraints meet both safety requirements and adapt to actual motion demands. Analyzing the distribution of deviations: if most deviations are positive, the constraint is too loose, potentially causing servo system overload; if most are negative, the constraint is too tight, limiting the device's dynamic performance. Adjusting the amplitude of the overall constraint vector based on the mean and extreme values ​​of the deviations: for positive deviations, reduce the amplitude at the corresponding time point to the standard upper limit; for negative deviations, increase the amplitude to the standard lower limit.

[0089] During the adjustment process, phase information must be preserved to ensure that the corrected constraint vector is consistent with the original constraint in timing. The adjusted constraint vector needs to be re-verified to ensure its matching degree with the standard data, ensuring that the corrected threshold is within a safe range (e.g., not exceeding 90% of the rated torque) and matches the actual motion state (e.g., higher torque is allowed in the centripetal phase, and lower restrictions are imposed in the centrifugal phase). The obtained low-torque balance threshold includes the corrected amplitude range and phase information. This constraint is directly used for resistance output control in subsequent closed-loop fusion, achieving smoothness and safety of resistance output in low-torque scenarios.

[0090] This embodiment decomposes the dynamic safety torque constraint parameters into axial and radial components through multidimensional spatial decomposition, comprehensively covering the torque constraint requirements in different directions during equipment movement. This avoids the limitations of single-direction constraints and provides more complete constraint characteristics for subsequent analysis. Coupling relationship analysis quantifies the interaction strength between axial and radial torques, enabling the comprehensive constraint vector to reflect the synergistic effect of the two during movement and improving the matching degree between constraints and actual motion states. Weighted fusion adjusts the component weights according to the degree of coupling, optimizing the amplitude and phase of the comprehensive constraint vector to ensure that its constraints in any direction meet safety requirements. Boundary matching degree calculation, by comparing with preset standard data, clarifies the degree of deviation between the comprehensive constraints and the safety threshold, providing a quantitative basis for dynamic correction.

[0091] Furthermore, in one embodiment, closed-loop fusion is performed on the low torque compensation signal and the low torque balance threshold to generate global compensation information, including: The reference resistance of the main resistance unit is the basic resistance output of the equipment in the uncompensated state. It is usually determined by the mechanical characteristics of the transmission chain (such as gear ratio and bearing friction coefficient) and the rated parameters of the motor (such as rated torque and current-torque conversion factor). This reference resistance can be obtained through static calibration before the equipment leaves the factory or established through training with historical data during long-term operation. The drive circuit of the main resistance unit is called to read its output torque value in the current uncompensated state. After filtering to remove high-frequency noise, the value is determined as the reference resistance value.

[0092] The low-torque compensation signal is a time-domain signal containing dynamic compensation information generated in the preceding steps. Its waveform must be strictly aligned with the reference resistance in the time dimension to ensure the effectiveness of the compensation. Waveform reference alignment is achieved through timestamp synchronization: both the reference resistance signal and the optimized compensation signal of the main resistance unit have precise timestamps (accuracy up to 1ms). By comparing the timestamp deviations between the two, the compensation signal is time-shifted to ensure that their start points, peak points, and end points completely coincide on the time axis. The aligned compensation signal and the reference resistance signal maintain consistency in waveform characteristics (such as amplitude change rate and phase relationship), forming a consistent reference compensation signal, providing a unified reference for subsequent boundary definition and fusion.

[0093] The low torque balance threshold is a dynamic safe torque range, encompassing an upper and lower limit, and an allowable fluctuation range. It limits the compensated resistance output to within the equipment's safe operating boundary. The core of the boundary constraint calculation is to limit the amplitude of the reference-consistent compensation signal within this threshold range: for amplitudes exceeding the upper limit, a limiter truncates them to the upper limit; for amplitudes below the lower limit, a DC bias adjustment raises them to the lower limit. During the calculation, the phase information of the compensation signal is preserved, and only the amplitude range is adjusted to ensure that the constrained signal still matches the intended action.

[0094] Heterogeneous channel fusion is performed on the reference consistent compensation signal and the original reference resistance signal of the main resistance unit. Since the compensation signal and the reference resistance signal come from different control channels (compensation channel and main resistance channel), a fusion algorithm is needed to eliminate the differences between the channels. Using the reference resistance signal as a reference, and based on the effectiveness of the compensation signal (e.g., a weight of 0.7 when the compensation error is less than ±0.1 Nm, otherwise 0.3), a weighted sum of the two is calculated to generate a composite resistance intermediate signal. This signal retains the basic characteristics of the reference resistance while incorporating the dynamic adjustment information of the compensation signal, providing a fused input basis for subsequent phase correction.

[0095] The vibration data of the mechanism reflects the actual vibration state of the equipment's transmission chain during operation, and its phase information is directly related to the force exertion stage of the mechanical motion (such as centripetal and centrifugal). Although the intermediate signal of the composite resistance is bounded and fused, its phase deviates from the phase of the actual mechanical vibration (such as due to control delay or signal transmission error), and offset correction needs to be performed using vibration data.

[0096] The instantaneous phase of the mechanism vibration data is extracted (calculated using Hilbert transform) and compared with the phase of the intermediate signal of the composite resistance to calculate the phase difference. If the phase difference exceeds ±5° (a safety threshold determined by the mechanical characteristics of the equipment), the phase of the intermediate signal of the composite resistance is adjusted: by delaying or advancing the timestamp of the signal, the phase of the adjusted signal is made consistent with the phase of the vibration data. During the correction process, the mechanical delay of the equipment's transmission chain (such as transmission ratio and shaft stiffness) needs to be considered to correct the offset of the timestamp, ensuring that the phase of the corrected signal is synchronized with the actual force application stage of the mechanical motion. The final phase-coordinated signal has a phase highly consistent with the mechanism vibration data, providing an input signal synchronized with the mechanical state for subsequent energy distribution equalization and interpolation fitting.

[0097] The phase-synchronized signal has achieved phase synchronization with the vibration data of the mechanism, but its resistance distribution still has local abrupt changes or discontinuities (such as due to the fusion error between the compensation signal and the reference resistance). It needs to be optimized through energy distribution equalization and global interpolation fitting to generate a smooth resistance output suitable for the entire motion cycle.

[0098] The core of energy distribution equalization is adjusting the uniformity of the resistance amplitude of the phase-coordinated signal within the motion cycle. The resistance amplitude sequence of the phase-coordinated signal is extracted and, combined with motion trajectory data from the servo compensation parameter set, the peak and trough regions of the resistance distribution are identified. Based on the "resistance smoothness" index in the compensation parameter set, the amplitude in the peak regions is attenuated, and the amplitude in the trough regions is increased, ensuring that the rate of change of the resistance amplitude within the motion cycle is controlled within a safe range (e.g., ≤5% / ms). This process eliminates local resistance abrupt changes by dynamically adjusting the signal amplitude distribution, ensuring that the resistance output is consistent with the smoothness requirements of human motion perception.

[0099] The goal of global resistance interpolation fitting is to convert discrete phase-coordinated signals into continuous resistance output, covering the entire trajectory of the device's motion. Using the motion trajectory in the servo compensation parameter set as a reference, the discrete time points of the phase-coordinated signals are mapped to continuous positions on the trajectory. Piecewise linear interpolation or cubic spline interpolation algorithms are employed to generate resistance values ​​at arbitrary positions based on the resistance amplitude and trajectory position at adjacent time points. During interpolation, the "resistance continuity" constraint in the compensation parameter set must be considered to ensure that the interpolated resistance curve has no abrupt changes. The interpolation result is superimposed on the signal after energy distribution equalization to form global compensation information covering the entire motion cycle. This information is directly used for the drive control of the servo system, achieving continuity and comfort in resistance output under low-torque scenarios.

[0100] This embodiment ensures strict synchronization between the low-torque compensation signal and the reference resistance of the main resistance unit in the time dimension through waveform reference alignment, avoiding compensation failure caused by time deviation and improving the matching accuracy between the compensation signal and the motion intention. Boundary limitation calculation combined with the low-torque balance threshold limits the amplitude of the compensation signal within a safe range, preventing performance degradation of the servo system due to overload and avoiding resistance fluctuations caused by insufficient compensation. Heterogeneous channel fusion integrates the reference resistance and compensation signal through weighted integration, preserving the stability of the basic resistance and the dynamic adjustment characteristics of the compensation signal, enhancing the overall adaptability of the output resistance. Based on the offset correction of the mechanism vibration data, the phase deviation caused by control delay or signal transmission error is eliminated by synchronizing phase information, making the compensation signal highly consistent with the actual force application stage of the mechanical motion, improving the real-time performance of the resistance output.

[0101] Reference Figure 2 As shown, the present invention also provides a dynamic weight compensation device based on low-torque servo, applicable to any of the above-mentioned dynamic weight compensation methods based on low-torque servo, comprising: The acquisition module is used to acquire the mechanical response data of the target device and the user's motion characteristics, perform predefined action type matching, and obtain a set of servo compensation parameters. The analysis module is used to drive the main resistance unit of the target device to generate a reference resistance based on the mechanical response data, and to perform dual-path control analysis in combination with the servo compensation parameter set to obtain dual-path compensation data. The processing module is used to acquire the mechanical vibration data of the target equipment, suppress torque fluctuations in the dual-channel compensation data, and obtain a low torque compensation signal. The correlation module is used to predict the mechanical state based on user motion characteristics and mechanism vibration data, and output a low torque balance threshold. The module is used to perform closed-loop fusion of the low torque compensation signal and the low torque balance threshold to generate global compensation information.

[0102] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the system and each module described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0103] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A dynamic weight compensation method based on low-torque servo, characterized in that, include: The raw tensile data sequence is acquired by the strain sensor array of the target device, and time-domain noise reduction is performed to obtain mechanical response data; The system acquires the triaxial acceleration and angular velocity data streams from the inertial measurement unit of the target device and extracts motion features to obtain user motion features. Based on a preset compensation parameter reference library, the system performs predefined type indexing on the mechanical response data and the user motion features to obtain predefined motion types. Based on the predefined motion types, the system performs reference inertia and transmission chain equivalent compensation conversion on the mechanical response data and the user motion features to obtain a servo compensation parameter set. The mechanical response data is separated by frequency domain energy spectrum to obtain dynamic mechanical feature vector; the main resistance unit is nonlinearly approximated by magnetic circuit output based on the dynamic mechanical feature vector to obtain a reference resistance distribution set; the servo compensation parameter set is decomposed by translation control and rotation control according to a preset kinematic orthogonal decoupling mechanism to obtain a two-degree-of-freedom control input group. The reference resistance distribution set and the two-degree-of-freedom control input set are dynamically and interactively corrected to obtain a hybrid control protocol. The main resistance unit is reconstructed using the power drive modulation signal according to the hybrid control protocol to obtain dual-channel compensation data; The vibration data of the mechanism are subjected to spectral feature separation to obtain the main resonant mode components and the secondary resonant mode components; Phase lag compensation is performed on the main resonant mode component based on the dual-channel compensation data to obtain a set of critical oscillation frequency points; based on the set of critical oscillation frequency points, driving waveform envelope boundary constraints are applied to the dual-channel compensation data to obtain an anti-saturation compensation signal; The active damping injection coefficient is obtained by performing coupled resonance energy inversion calculation on the sub-resonant mode components and the anti-saturation compensation signal. Based on the dual-path compensation data and the active damping injection coefficient, pulse width modulation depth signal reconstruction is performed to obtain a low torque compensation signal; Periodic motion phase segmentation is performed on the user motion characteristics to obtain a motion state label sequence; dominant frequency band energy features are extracted from the mechanism vibration data to obtain mechanical clustering frequency band information; a dynamic interactive state mapping relationship is constructed based on the motion state label sequence and the mechanical clustering frequency band information. The dynamic interactive state mapping relationship is quantitatively stable to obtain the critical working region boundary; based on the critical working region boundary, a low torque safety margin conversion is performed to generate dynamic safety torque constraint parameters. The dynamic safety torque constraint parameters are adaptively matched to obtain the low torque balance threshold. Obtain the reference resistance of the main resistance unit, and perform waveform reference alignment on the low torque compensation signal to obtain a reference consistent compensation signal; Based on the low torque balance threshold, the reference consistent compensation signal is subjected to boundary limitation calculation and heterogeneous channel fusion to obtain a composite resistance intermediate signal; based on the mechanism vibration data, the composite resistance intermediate signal is offset-corrected to obtain a phase coordination signal; based on the servo compensation parameter set, the phase coordination signal is subjected to energy distribution equalization and global resistance interpolation fitting to obtain global compensation information.

2. The dynamic weight compensation method based on low-torque servo according to claim 1, characterized in that, The step of performing coupled resonance energy inversion calculations on the sub-resonant mode components and the anti-saturation compensation signal to obtain the active damping injection coefficients includes: Energy spectrum focusing extraction is performed on the sub-resonant mode components to obtain a set of resonant peak frequencies; Based on the anti-saturation compensation signal, drive phase difference mapping is performed to generate an anti-phase compensation sequence; Based on the set of resonant peak frequencies and the anti-phase compensation sequence, a dual-domain damping correlation calculation is performed to generate an initial anti-phase damping function. Perform orthogonal decoupling and gain factor correction on the initial antiphase damping function to obtain a set of gain damping functions; The active damping injection coefficient is obtained by performing energy inversion calculation on the gain damping function set.

3. The dynamic weight compensation method based on low-torque servo according to claim 1, characterized in that, The process of adapting the dynamic safety torque constraint parameters to obtain a low torque balance threshold includes: The dynamic safety torque constraint parameters are decomposed into axial torque constraint components and radial torque constraint components. The coupling relationship between the axial torque constraint component and the radial torque constraint component is analyzed to obtain the torque component correlation coefficient. Based on the torque component correlation coefficient, the axial torque constraint component and the radial torque constraint component are weighted and fused to obtain a comprehensive dynamic torque constraint vector. The boundary matching degree of the comprehensive dynamic torque constraint vector is calculated based on the preset torque feedback standard data to obtain the torque constraint deviation. The comprehensive dynamic torque constraint vector is dynamically thresholded based on the torque constraint deviation to obtain a low torque balance threshold.

4. A dynamic weight compensation device based on low-torque servo, characterized in that, The dynamic weight compensation method based on low-torque servo, applied to any one of claims 1-3, includes: The acquisition module is used to acquire raw tensile data sequences through the strain sensor array of the target device, perform time-domain noise reduction to obtain the mechanical response data; acquire the triaxial acceleration and angular velocity data stream of the inertial measurement unit of the target device, and extract motion features to obtain the user motion features; index the mechanical response data and the user motion features into predefined types based on a preset compensation parameter benchmark library to obtain predefined motion types; and perform benchmark inertia and transmission chain equivalent compensation conversion on the mechanical response data and the user motion features based on the predefined motion types to obtain the servo compensation parameter set. The analysis module is used to perform frequency domain energy spectrum separation on the mechanical response data to construct a dynamic mechanical feature vector; based on the dynamic mechanical feature vector, perform nonlinear approximation of the magnetic circuit output of the main resistance unit to obtain a reference resistance distribution set; decompose the servo compensation parameter set into translational and rotational control according to a preset kinematic orthogonal decoupling mechanism to obtain a two-degree-of-freedom control input group; perform dynamic interactive correction on the reference resistance distribution set and the two-degree-of-freedom control input group to obtain a hybrid control protocol; and reconstruct the power drive modulation signal of the main resistance unit according to the hybrid control protocol to obtain the dual-path compensation data. The processing module is used to perform spectral feature separation on the vibration data of the mechanism to obtain the main resonant mode component and the secondary resonant mode component; perform phase lag compensation on the main resonant mode component based on the dual-channel compensation data to obtain a set of critical oscillation frequency points; perform driving waveform envelope boundary constraints on the dual-channel compensation data based on the set of critical oscillation frequency points to obtain an anti-saturation compensation signal; perform coupled resonant energy inversion calculation on the secondary resonant mode component and the anti-saturation compensation signal to obtain the active damping injection coefficient; and perform pulse width modulation depth signal reconstruction based on the dual-channel compensation data and the active damping injection coefficient to obtain a low torque compensation signal. The association module is used to perform periodic action phase segmentation on the user's motion characteristics to obtain a motion state label sequence; extract dominant frequency band energy features from the mechanism vibration data to obtain mechanical clustering frequency band information; construct a dynamic interactive state mapping relationship based on the motion state label sequence and the mechanical clustering frequency band information; perform quantitative stability evaluation on the dynamic interactive state mapping relationship to obtain the critical working area boundary; perform low torque safety margin conversion based on the critical working area boundary to generate dynamic safety torque constraint parameters; and perform adaptive matching on the dynamic safety torque constraint parameters to obtain a low torque balance threshold. The module is configured to acquire the reference resistance of the main resistance unit, perform waveform reference alignment on the low torque compensation signal to obtain a reference consistent compensation signal, perform boundary limitation calculation and heterogeneous channel fusion on the reference consistent compensation signal according to the low torque balance threshold to obtain a composite resistance intermediate signal, perform offset correction on the composite resistance intermediate signal based on the mechanism vibration data to obtain a phase coordination signal, and perform energy distribution equalization and global resistance interpolation fitting on the phase coordination signal according to the servo compensation parameter set to obtain the global compensation information.

Images