Robot driving module real-time optimization method and system based on multi-modal perception

Abstract

The invention is suitable for the technical field of robot control, and provides a robot driving module real-time optimization method and system based on multi-modal perception, and the method comprises the following steps: sampling a driving module, and obtaining multi-modal original data representing electrical, thermal and mechanical vibration; calculating an electricity-heat correlation degree, an electricity-vibration correlation degree and a heat-vibration correlation degree as initial features; calculating a coupling stress index based on the original data and the initial characteristics, dynamically adjusting scaling factors of a current limit value, a rotating speed limit value and a controller bandwidth limit value according to a real-time numerical value of the coupling stress index in combination with a preset mapping function, determining a real-time feed-forward compensation coefficient, and reconstructing a safety boundary; constructing an optimization problem taking tracking errors, energy loss and control smoothness as optimization targets by taking the security boundary as a hard constraint; and solving the optimization problem to obtain the optimal current loop proportional gain and integral gain. According to the method, the system performance can be maintained to the greatest extent on the premise of ensuring the safety.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, specifically to a real-time optimization method and system for robot drive modules based on multimodal perception. Background Technology

[0002] Traditional drive control systems typically rely on single or limited electrical parameters such as current, voltage, and speed for control, lacking simultaneous monitoring and comprehensive analysis of multiple physical fields, including temperature field distribution and mechanical vibration. This results in the system's inability to fully perceive its own operating status, especially under high load and high dynamic conditions, where the coupling effect between electricity, heat, and mechanical vibration is ignored, making it difficult for the controller to predict potential risks of overheating, resonance, or stability degradation.

[0003] While some advanced control strategies incorporate online optimization, they typically focus on optimizing only a single performance metric (such as minimum tracking error), neglecting the synergistic effects of multiple objectives like energy efficiency, temperature rise, and vibration. Furthermore, the optimization process is often disconnected from the system's real-time safety boundaries, easily generating mathematically optimal but practically high-risk control commands (such as approaching overheating limits), lacking a mechanism to integrate real-time risk perception into optimization objectives and constraints. Additionally, when dealing with disturbances such as sudden load increases or uneven terrain, traditional feedforward compensation coefficients are often fixed values ​​or calculated based on simple models, failing to dynamically adjust according to the real-time perceived system coupling state, resulting in slow dynamic response and limited disturbance rejection capabilities. Therefore, a real-time optimization method and system for robot drive modules based on multimodal perception is needed to address these issues. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a real-time optimization method and system for robot drive modules based on multimodal perception, so as to solve the problems existing in the above-mentioned background technology.

[0005] This invention is implemented as follows: a real-time optimization method for a robot drive module based on multimodal perception, the method comprising the following steps: The drive module is sampled to obtain multimodal raw data characterizing electrical, thermal, and mechanical vibrations; based on a sliding time window, time-frequency analysis and correlation analysis are performed on the multimodal raw data to dynamically calculate the electro-thermal correlation, electro-vibration correlation, and thermal-vibration correlation as initial features; The coupling stress index is calculated in real time based on the original data and initial features. Based on the real-time value of the coupling stress index and combined with the preset mapping function, the scaling factors of the current limit, speed limit and controller bandwidth limit are dynamically adjusted to determine the real-time feedforward compensation coefficient and reconstruct the safety boundary. Using safety boundaries as hard constraints, a real-time optimization problem is constructed with tracking error, energy loss, and control smoothness as optimization objectives. Based on the real-time state, the optimization problem is solved to obtain the optimal current loop proportional gain and integral gain. The optimal control parameters obtained from the solution are applied to the current regulator to achieve real-time optimized control of the robot's drive motor.

[0006] As a further aspect of the present invention, the steps of dynamically calculating the electro-thermal correlation, electro-vibrational correlation, and thermal-vibrational correlation specifically include: By analyzing the covariance relationship between the effective value of the quadrature shaft current and the winding temperature rise rate over time, and by combining the weighted correction with the deviation between the current speed and the rated speed, the electro-thermal correlation degree is obtained. Spectral analysis was performed on the direct-axis current signal and the axial vibration acceleration signal respectively. The coherence intensity of the current spectrum component and the vibration spectrum component at the main current harmonic frequency was calculated. The electro-vibration correlation degree was obtained by weighting the harmonics with their importance as weights. The thermal-vibration correlation degree is obtained by calculating the Pearson correlation coefficient between the winding temperature time series signal and the root mean square value vibration acceleration time series signal within the window.

[0007] As a further aspect of the present invention, the step of calculating the coupling stress index in real time based on the original data and initial features specifically includes: A coupled stress index basic model is constructed. The model is based on the product of normalized real-time torque current command, winding temperature rise rate, electro-vibration correlation degree and vibration amplitude, and a power function with the ratio of rated voltage to real-time voltage as the base. A multiphysics cross-coupling compensation model is constructed. The model linearly combines the product of the electro-thermal correlation degree and the relative temperature rise, as well as the product of the thermal-vibration correlation degree and the relative vibration intensity, as compensation terms for the basic model. The output of the base model and the output of the compensation term are weighted and fused to obtain the final coupling stress index, which is given by the formula: CSI(t) = γ1 × CSIbase(t) + γ2 × CSIcoup(t), where CSI(t) is the coupling stress index at time t, CSIbase(t) is the output value of the base model, CSIcoup(t) is the output value of the cross-coupling compensation term, and γ1 and γ2 are weight coefficients.

[0008] As a further aspect of the present invention, the step of dynamically adjusting the scaling factors of the current limit, speed limit, and controller bandwidth limit specifically includes: By inputting the real-time coupling stress index, current adjustment threshold and real-time bus voltage drop into an S-shaped decay function, a current limit scaling factor between 0 and 1 is calculated. By inputting the real-time coupling stress index, the speed adjustment threshold, and the real-time winding temperature into an exponential decay function, a speed limit scaling factor between 0 and 1 is calculated. By inputting the real-time coupling stress index, bandwidth adjustment threshold and real-time electro-vibration correlation into a hyperbolic tangent adjustment function, a bandwidth scaling factor between 0 and 1 is calculated. The real-time current limit, real-time speed limit, and controller parameter range are obtained based on the scaling factor.

[0009] As a further aspect of the present invention, determining the real-time feedforward compensation coefficient specifically includes: inputting the real-time coupling stress index, the rotational speed change rate, and the real-time electro-thermal correlation degree into a dynamic response function, calculating the feedforward gain scaling factor, and multiplying the feedforward gain scaling factor by the rated feedforward gain to obtain the real-time feedforward compensation coefficient.

[0010] As a further aspect of the present invention, the steps of constructing a real-time optimization problem to obtain the optimal current loop proportional gain and integral gain specifically include: An optimization objective function is constructed that includes four sub-objectives: the accuracy of speed tracking, the energy efficiency of motor operation, the stability of control output, and the smoothness of changes in control parameters. The real-time current limit, speed limit, and controller parameter range are set as hard constraints in the optimization solution process. The real-time feedforward compensation coefficient is taken as a known quantity, and the proportional gain and integral gain of the current loop are taken as variables to be optimized. Using the gradient descent algorithm, the objective function is solved online under the hard constraints, and the optimal values ​​of proportional gain and integral gain are updated in real time.

[0011] Another objective of this invention is to provide a real-time optimization system for a robot drive module based on multimodal perception, the system comprising: The correlation determination module is used to sample the drive module and obtain multimodal raw data characterizing electrical, thermal and mechanical vibrations; based on the sliding time window, time-frequency analysis and correlation analysis are performed on the multimodal raw data to dynamically calculate the electro-thermal correlation, electro-vibration correlation and thermal-vibration correlation as initial features; The safety boundary reconstruction module is used to calculate the coupling stress index in real time based on the original data and initial characteristics. Based on the real-time value of the coupling stress index and combined with the preset mapping function, the scaling factors of the current limit, speed limit and controller bandwidth limit are dynamically adjusted to determine the real-time feedforward compensation coefficient and reconstruct the safety boundary. The optimization problem-solving module is used to construct a real-time optimization problem with tracking error, energy loss, and control smoothness as optimization objectives, using a safety boundary as a hard constraint; based on the real-time state, the optimization problem is solved to obtain the optimal current loop proportional gain and integral gain; The control parameter application module is used to apply the solved optimal control parameters to the current regulator to achieve real-time optimized control of the robot drive motor.

[0012] Compared with the prior art, the beneficial effects of the present invention are: By simultaneously acquiring and analyzing multimodal raw data of electrical, thermal, and mechanical vibrations, and dynamically calculating the electro-thermal, electro-vibrational, and thermal-vibrational correlations, the interaction and influence mechanisms between different physical fields can be accurately characterized. Through the coupling stress index calculated in real time based on multi-source data and correlation characteristics, the risk level of approaching the overall operational limit can be continuously and quantitatively assessed. Based on the coupling stress index, dynamically adjusted current limits, speed limits, controller bandwidth scaling factors, and feedforward compensation coefficients can maximize system performance while ensuring safety, avoiding drastic performance interruptions caused by traditional protection mechanisms, and finding the optimal solution that meets current safety constraints. Attached Figure Description

[0013] Figure 1 This is a flowchart of a real-time optimization method for robot drive modules based on multimodal perception.

[0014] Figure 2 This is a flowchart illustrating the calculation of correlation degree in a real-time optimization method for robot drive modules based on multimodal perception.

[0015] Figure 3 This is a flowchart illustrating the calculation of the coupling stress index in a real-time optimization method for robot drive modules based on multimodal perception.

[0016] Figure 4 This is a flowchart illustrating the determination of scaling factors in a real-time optimization method for robot drive modules based on multimodal perception.

[0017] Figure 5 This is a flowchart illustrating the process of obtaining optimal control parameters in a real-time optimization method for robot drive modules based on multimodal perception.

[0018] Figure 6 This is a schematic diagram of the structure of a real-time optimization system for a robot drive module based on multimodal perception. Detailed Implementation

[0019] 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 specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0021] like Figure 1 As shown, this embodiment of the invention provides a real-time optimization method for a robot drive module based on multimodal perception. The method includes the following steps: S100 samples the drive module to obtain multimodal raw data characterizing electrical, thermal, and mechanical vibrations; based on a sliding time window, it performs time-frequency analysis and correlation analysis on the multimodal raw data, dynamically calculating the electro-thermal correlation, electro-vibration correlation, and thermal-vibration correlation as initial features; S200 calculates the coupling stress index in real time based on the original data and initial characteristics. Based on the real-time value of the coupling stress index and combined with the preset mapping function, it dynamically adjusts the scaling factors of the current limit, speed limit and controller bandwidth limit to determine the real-time feedforward compensation coefficient and reconstruct the safety boundary. S300 uses a safety boundary as a hard constraint to construct a real-time optimization problem with tracking error, energy loss, and control smoothness as optimization objectives; based on the real-time state, it solves the optimization problem to obtain the optimal current loop proportional gain and integral gain. The S400 applies the solved optimal control parameters to the current regulator to achieve real-time optimized control of the robot drive motor.

[0022] In this embodiment of the invention, firstly, a synchronous sampling circuit is used to simultaneously acquire multi-modal raw data of electrical, thermal, and mechanical vibration of the drive module at a sampling frequency of not less than 10kHz, and the data of each channel is denoised and normalized. The multi-modal raw data specifically includes: 1. Electrical data: three-phase stator currents Ia, Ib, and Ic, DC bus voltage Vdc, and motor winding back electromotive force; 2. Thermal data: stator winding temperature Tw, rotor permanent magnet temperature Tm, and power device junction temperature Tj, acquired through a distributed temperature sensor array; 3. Mechanical vibration data: motor axial vibration acceleration ax, radial vibration accelerations ay and az, and vibration spectrum characteristics, acquired through a triaxial MEMS accelerometer; 4. Operating condition data: motor speed ωm, load torque τ, and ambient temperature Tamb. Then, a sliding time window Wt is defined, and the length Lw of the sliding time window is adaptively adjusted according to the current motor speed: Lw = max(0.1, 10 / ωm) seconds. Based on a sliding time window, time-frequency analysis and correlation analysis are performed on the multimodal raw data to dynamically calculate the electro-thermal correlation, electro-vibration correlation, and thermal-vibration correlation, and these correlations are used as initial features.

[0023] Next, based on the original data and initial characteristics, a coupled stress index is calculated to comprehensively characterize the risk level of the system approaching its operational limits. Then, based on the real-time value of the coupled stress index and a preset mapping function, the scaling factors of the current limit, speed limit, and controller bandwidth limit are dynamically adjusted to determine the real-time feedforward compensation coefficient, thereby reconstructing the real-time safety boundary of the control system. Then, using the safety boundary as a hard constraint, a real-time optimization problem is constructed with tracking error, energy loss, and control smoothness as optimization objectives. Based on the real-time system state, this problem is solved online using a gradient descent algorithm to obtain the optimal current loop proportional-integral gain. Finally, the solved optimal control parameters are applied to the current regulator to achieve real-time optimized control of the robot's drive motor.

[0024] like Figure 2 As shown, in a preferred embodiment of the present invention, the steps of dynamically calculating the electro-thermal correlation, electro-vibration correlation, and thermal-vibration correlation specifically include: S101. By analyzing the covariance relationship between the effective value of the quadrature shaft current and the winding temperature rise rate over time, and combining it with the weighted correction of the deviation between the current speed and the rated speed, the electro-thermal correlation degree is obtained. S102, perform spectral analysis on the direct-axis current signal and the axial vibration acceleration signal respectively, calculate the coherence intensity of the current spectrum component and the vibration spectrum component at the main current harmonic frequency, and perform weighted summation with the importance of each harmonic as the weight to obtain the electro-vibration correlation degree. S103, calculate the Pearson correlation coefficient between the winding temperature time series signal and the root mean square value of vibration acceleration time series signal within the window to obtain the thermal-vibration correlation degree.

[0025] In this embodiment of the invention, the electro-thermal correlation degree RET= exp(−α ∣ωm−ωnom∣), where Iq,rms is the effective value of the quadrature-axis current. The winding temperature rise rate is represented by Cov, where Cov represents the covariance, σ1 represents the standard deviation of the effective value of the quadrature-axis current, σ2 represents the standard deviation of the winding temperature rise rate, α is the speed deviation attenuation coefficient, and ωnom is the rated speed. Electro-oscillation correlation REV = wk. Where FFT()fk represents the Fourier transform value at the characteristic frequency fk. Let f be the direct-axis current, fk be the k-th harmonic frequency, wk be the weighting coefficient for that harmonic frequency, ϵ be a small constant to prevent division by zero, and Nh be the harmonic order considered. The thermal-vibration correlation coefficient RTV can be directly calculated using the Pearson correlation coefficient algorithm. Finally, the calculated RET, REV, and RTV need to be normalized to the [0,1] interval as input features for subsequent steps.

[0026] like Figure 3 As shown, in a preferred embodiment of the present invention, the step of calculating the coupling stress index in real time based on the original data and initial features specifically includes: S201, Construct a basic model of coupled stress index. The model is based on the product of normalized real-time torque current command, winding temperature rise rate, electro-vibration correlation degree and vibration amplitude, and a power function with the ratio of rated voltage to real-time voltage as the base. S202, construct a multi-physics cross-coupling compensation model. The model linearly combines the product of the electro-thermal correlation degree and the relative temperature rise, as well as the product of the thermal-vibration correlation degree and the relative vibration intensity, as compensation terms for the basic model. S203 weights and fuses the output of the basic model with the output of the compensation term to obtain the final coupled stress index.

[0027] In this embodiment of the invention, the basic model of the coupling stress index is: CSIbase(t)=[Iq*(t) / Imax]×[1+λ×dTw / dt]×[1+η×(REV×Avib / Anom)]×[Vnom / Vdc(t)]^k, where t is the current time, Iq*(t) is the current quadrature axis current command value, Imax is the maximum allowable continuous current of the motor, λ is the temperature rise rate influence coefficient, Avib is the real-time vibration main frequency amplitude, Anom is the vibration amplitude under rated operating conditions, η is the vibration risk influence coefficient, Vnom is the rated DC bus voltage, Vdc(t) is the real-time DC bus voltage, and k is the voltage stress index (>1), which reflects the nonlinearity of the voltage drop's impact on the system margin. The multiphysics cross-coupling compensation model is: CSIcoup(t) = β1 × RET(t) × [(Tw(t) - Tamb) / (Tmax - Tamb)] + β2 × RTV(t) × [arms(t) / arms_max], where Tw(t) is the real-time winding temperature, Tmax is the maximum allowable operating temperature of the winding, arms(t) is the real-time root mean square value of vibration acceleration, arms_max is the maximum allowable root mean square value of vibration acceleration, and β1 and β2 are the weighting coefficients of the cross-coupling term. Finally, the output of the basic model and the output of the compensation term are weighted and fused to obtain the final coupling stress index, which is: CSI(t) = γ1 × CSIbase(t) + γ2 × CSIcoup(t), where CSI(t) is the coupling stress index at time t, CSIbase(t) is the output value of the basic model, CSIcoup(t) is the output value of the cross-coupling compensation term, and γ1 and γ2 are the weighting coefficients.

[0028] like Figure 4 As shown, in a preferred embodiment of the present invention, the step of dynamically adjusting the scaling factors of the current limit, speed limit, and controller bandwidth limit specifically includes: S204 inputs the real-time coupling stress index, current adjustment threshold and real-time bus voltage drop into an S-shaped decay function to calculate a current limit scaling factor between 0 and 1. S205 inputs the real-time coupling stress index, speed adjustment threshold and real-time winding temperature into an exponential decay function to calculate a speed limit scaling factor between 0 and 1. S206, by inputting the real-time coupling stress index, bandwidth adjustment threshold and real-time electro-vibration correlation into a hyperbolic tangent adjustment function, a bandwidth scaling factor between 0 and 1 is calculated; S207, based on the scaling factor, obtains the real-time current limit, real-time speed limit, and controller parameter value range.

[0029] In this embodiment of the invention, the formula for calculating the current limit scaling factor αI is: αI(t)=1 / (1+exp(βI×(CSI(t)-CSIthI)))×[1-δ×((Vnom-Vdc(t)) / Vnom)], Ilimit(t)=Imax×αI(t), where CSIthI is the current scaling trigger threshold, βI is the adjustment slope coefficient, δ is the voltage drop compensation coefficient, and Ilimit(t) is the real-time current limit. The formula for calculating the speed limit scaling factor αω is: αω(t)=exp(-βω×max(0,CSI(t)-CSIthω))×[1-κ×((Tw(t)-Tref) / (Tmax-Tref))], ωlimit(t)=ωmax×αω(t), where CSIthω is the speed scaling trigger threshold, βω is the attenuation coefficient, κ is the temperature influence coefficient, Tref is the temperature reference value, ωmax is the maximum allowable speed of the motor, and ωlimit(t) is the real-time speed limit. The formula for calculating the controller bandwidth scaling factor αBW is: αBW=1-γBW×tanh((CSI(t)-CSIthBW) / σ)+ξ×(REV(t) / REVmax), where CSIthBW is the bandwidth adjustment trigger threshold, γBW is the maximum bandwidth adjustment amplitude, σ is the adjustment smoothing coefficient, ξ is the vibration correlation compensation coefficient, and REVmax is the reference maximum value of the electro-vibration correlation degree. Applying αBW to the preset nominal ranges [Kpmin, Kpmax] and [Kimin, Kimax] of the proportional gain Kp and integral gain Ki creates a dynamic constraint on the range of parameter values. That is, the values ​​of Kp and Ki must fall within the intervals [αBW(t)×Kpmin, αBW(t)×Kpmax] and [αBW(t)×Kimin, αBW(t)×Kimax].

[0030] In this embodiment of the invention, the real-time coupling stress index, the rate of change of rotational speed, and the real-time electro-thermal correlation are also input into a dynamic response function to calculate the feedforward gain scaling factor. The feedforward gain scaling factor is multiplied by the rated feedforward gain to obtain the real-time feedforward compensation coefficient. Specifically, the formula for calculating the feedforward gain scaling factor αff is: αff(t)=1 / (1+ζ×CSI(t)×|dω / dt|)×(1+ν×RET(t)), Kff(t)=Kffnom×αff(t), where ζ is the dynamic response suppression coefficient, |dω / dt| is the absolute value of the rate of change of rotational speed, ν is the electro-thermal correlation enhancement coefficient, Kffnom is the rated feedforward gain, and Kff(t) is the real-time feedforward compensation coefficient.

[0031] like Figure 5As shown, in a preferred embodiment of the present invention, the steps of constructing a real-time optimization problem and obtaining the optimal current loop proportional gain and integral gain specifically include: S301, construct an optimization objective function containing four sub-objectives: the accuracy of speed tracking, the energy efficiency of motor operation, the stability of control output, and the smoothness of the changes in control parameters themselves; S302 sets the real-time current limit, speed limit, and controller parameter range as hard constraints in the optimization solution process. S303, take the real-time feedforward compensation coefficient as a known quantity, and take the proportional gain and integral gain of the current loop as variables to be optimized; S304 employs the gradient descent algorithm to iteratively solve the objective function online under the hard constraints, updating the optimal values ​​of the proportional gain and integral gain in real time.

[0032] In this embodiment of the invention, the tracking accuracy objective is reflected by minimizing the sum of squared rotational speed errors within a time window; the energy efficiency objective is reflected by minimizing copper losses caused by current within the same time period; the control stability objective is reflected by minimizing the rate of change of the control current command; and the parameter smoothness objective is reflected by minimizing the change of the control parameter itself in adjacent time moments. The four sub-objectives are weighted and summed to obtain the optimization objective function. The real-time measured rotational speed error and current value are used as inputs, and the proportional gain and integral gain of the current loop are used as the two degrees of freedom to be optimized. Under hard constraints, an online optimization algorithm based on gradient direction estimation is used to perform real-time, continuous minimization search of the optimization objective function to determine the parameter values ​​that allow the system to reach the current optimal equilibrium point.

[0033] like Figure 6 As shown, this embodiment of the invention also provides a real-time optimization system for a robot drive module based on multimodal perception, the system comprising: The correlation determination module 100 is used to sample the drive module and obtain multimodal raw data characterizing electrical, thermal and mechanical vibrations; based on the sliding time window, time-frequency analysis and correlation analysis are performed on the multimodal raw data to dynamically calculate the electro-thermal correlation, electro-vibration correlation and thermal-vibration correlation as initial features; The safety boundary reconstruction module 200 is used to calculate the coupling stress index in real time based on the original data and initial characteristics. According to the real-time value of the coupling stress index, combined with the preset mapping function, the scaling factor of the current limit, speed limit and controller bandwidth limit is dynamically adjusted to determine the real-time feedforward compensation coefficient and reconstruct the safety boundary. The optimization problem-solving module 300 is used to construct a real-time optimization problem with tracking error, energy loss and control smoothness as optimization objectives, using the safety boundary as a hard constraint; based on the real-time state, the optimization problem is solved to obtain the optimal current loop proportional gain and integral gain; The control parameter application module 400 is used to apply the solved optimal control parameters to the current regulator to achieve real-time optimized control of the robot drive motor.

[0034] In a preferred embodiment of the present invention, the correlation degree determination module 100 includes: The electrothermal correlation unit is used to obtain the electrothermal correlation by analyzing the covariance relationship between the effective value of the quadrature axis current and the winding temperature rise rate over time, and by weighting and correcting the deviation between the current speed and the rated speed. The electro-vibration correlation unit is used to perform spectral analysis on the direct-axis current signal and the axial vibration acceleration signal respectively, calculate the coherence intensity of the current spectrum component and the vibration spectrum component at the main current harmonic frequency, and perform weighted summation with the importance of each harmonic as the weight to obtain the electro-vibration correlation degree. The thermal-vibration correlation unit is used to calculate the Pearson correlation coefficient between the winding temperature time series signal and the root mean square value vibration acceleration time series signal within the window, thus obtaining the thermal-vibration correlation.

[0035] In a preferred embodiment of the present invention, the security boundary reconstruction module 200 includes: The basic model unit is used to construct the basic model of the coupled stress index. The model is based on the product of normalized real-time torque current command, winding temperature rise rate, electro-vibration correlation degree and vibration amplitude, and a power function with the ratio of rated voltage to real-time voltage as the base. The compensation model unit is used to construct a multi-physics cross-coupling compensation model. The model linearly combines the product of the electro-thermal correlation degree and the relative temperature rise, as well as the product of the thermal-vibration correlation degree and the relative vibration intensity, as compensation terms for the basic model. The stress index unit is used to weight and fuse the output of the base model and the output of the compensation term to obtain the final coupled stress index. The formula is: CSI(t)=γ1×CSIbase(t)+γ2×CSIcoup(t), where CSI(t) is the coupled stress index at time t, CSIbase(t) is the output value of the base model, CSIcoup(t) is the output value of the cross-coupling compensation term, and γ1 and γ2 are weight coefficients.

[0036] In a preferred embodiment of the present invention, the security boundary reconstruction module 200 further includes: The current scaling factor unit is used to input the real-time coupling stress index, current adjustment threshold and real-time bus voltage drop into an S-shaped decay function to calculate a current limit scaling factor between 0 and 1. The speed scaling factor unit is used to input the real-time coupling stress index, speed adjustment threshold and real-time winding temperature into an exponential decay function to calculate a speed limit scaling factor between 0 and 1. The bandwidth scaling factor unit is used to input the real-time coupling stress index, bandwidth adjustment threshold and real-time electro-vibration correlation degree into a hyperbolic tangent adjustment function to calculate a bandwidth scaling factor between 0 and 1. The limit determination unit is used to obtain the real-time current limit, real-time speed limit, and controller parameter value range based on the scaling factor.

[0037] The above description only details the preferred embodiments of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0038] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0039] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the disclosure in the specification and embodiments. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

Claims

1. A real-time optimization method for robot drive modules based on multimodal perception, characterized in that, The method includes the following steps: The drive module is sampled to obtain multimodal raw data characterizing electrical, thermal, and mechanical vibrations; based on a sliding time window, time-frequency analysis and correlation analysis are performed on the multimodal raw data to dynamically calculate the electro-thermal correlation, electro-vibration correlation, and thermal-vibration correlation as initial features; The coupling stress index is calculated in real time based on the original data and initial features. Based on the real-time value of the coupling stress index and combined with the preset mapping function, the scaling factors of the current limit, speed limit and controller bandwidth limit are dynamically adjusted to determine the real-time feedforward compensation coefficient and reconstruct the safety boundary. Using safety boundaries as hard constraints, a real-time optimization problem is constructed with tracking error, energy loss, and control smoothness as optimization objectives. Based on the real-time state, the optimization problem is solved to obtain the optimal current loop proportional gain and integral gain. The optimal control parameters obtained from the solution are applied to the current regulator to achieve real-time optimized control of the robot's drive motor.

2. The real-time optimization method for robot drive modules based on multimodal perception according to claim 1, characterized in that, The steps for dynamically calculating the electro-thermal correlation, electro-vibrational correlation, and thermal-vibrational correlation specifically include: By analyzing the covariance relationship between the effective value of the quadrature shaft current and the winding temperature rise rate over time, and by combining the weighted correction with the deviation between the current speed and the rated speed, the electro-thermal correlation degree is obtained. Spectral analysis was performed on the direct-axis current signal and the axial vibration acceleration signal respectively. The coherence intensity of the current spectrum component and the vibration spectrum component at the main current harmonic frequency was calculated. The electro-vibration correlation degree was obtained by weighting the harmonics with their importance as weights. The thermal-vibration correlation degree is obtained by calculating the Pearson correlation coefficient between the winding temperature time series signal and the root mean square value vibration acceleration time series signal within the window.

3. The real-time optimization method for robot drive modules based on multimodal perception according to claim 1, characterized in that, The steps for real-time calculation of the coupled stress index based on raw data and initial features specifically include: A coupled stress index basic model is constructed. The model is based on the product of normalized real-time torque current command, winding temperature rise rate, electro-vibration correlation degree and vibration amplitude, and a power function with the ratio of rated voltage to real-time voltage as the base. A multiphysics cross-coupling compensation model is constructed. The model linearly combines the product of the electro-thermal correlation degree and the relative temperature rise, as well as the product of the thermal-vibration correlation degree and the relative vibration intensity, as compensation terms for the basic model. The output of the base model and the output of the compensation term are weighted and fused to obtain the final coupling stress index, which is given by the formula: CSI(t) = γ1 × CSIbase(t) + γ2 × CSIcoup(t), where CSI(t) is the coupling stress index at time t, CSIbase(t) is the output value of the base model, CSIcoup(t) is the output value of the cross-coupling compensation term, and γ1 and γ2 are weight coefficients.

4. The real-time optimization method for robot drive modules based on multimodal perception according to claim 1, characterized in that, The steps for dynamically adjusting the scaling factors of current limits, speed limits, and controller bandwidth limits specifically include: By inputting the real-time coupling stress index, current adjustment threshold and real-time bus voltage drop into an S-shaped decay function, a current limit scaling factor between 0 and 1 is calculated. By inputting the real-time coupling stress index, the speed adjustment threshold, and the real-time winding temperature into an exponential decay function, a speed limit scaling factor between 0 and 1 is calculated. By inputting the real-time coupling stress index, bandwidth adjustment threshold and real-time electro-vibration correlation into a hyperbolic tangent adjustment function, a bandwidth scaling factor between 0 and 1 is calculated. The real-time current limit, real-time speed limit, and controller parameter range are obtained based on the scaling factor.

5. The real-time optimization method for robot drive modules based on multimodal perception according to claim 4, characterized in that, The determination of the real-time feedforward compensation coefficient specifically includes: inputting the real-time coupled stress index, the rotational speed change rate, and the real-time electro-thermal correlation degree into a dynamic response function, calculating the feedforward gain scaling factor, and multiplying the feedforward gain scaling factor by the rated feedforward gain to obtain the real-time feedforward compensation coefficient.

6. The real-time optimization method for robot drive modules based on multimodal perception according to claim 4, characterized in that, The steps to construct a real-time optimization problem and obtain the optimal current loop proportional gain and integral gain include: An optimization objective function is constructed that includes four sub-objectives: the accuracy of speed tracking, the energy efficiency of motor operation, the stability of control output, and the smoothness of changes in control parameters. The real-time current limit, speed limit, and controller parameter range are set as hard constraints in the optimization solution process. The real-time feedforward compensation coefficient is taken as a known quantity, and the proportional gain and integral gain of the current loop are taken as variables to be optimized. Using the gradient descent algorithm, the objective function is solved online under the hard constraints, and the optimal values ​​of proportional gain and integral gain are updated in real time.

7. A real-time optimization system for robot drive modules based on multimodal perception, characterized in that, The system includes: The correlation determination module is used to sample the drive module and obtain multimodal raw data characterizing electrical, thermal and mechanical vibrations; based on the sliding time window, time-frequency analysis and correlation analysis are performed on the multimodal raw data to dynamically calculate the electro-thermal correlation, electro-vibration correlation and thermal-vibration correlation as initial features; The safety boundary reconstruction module is used to calculate the coupling stress index in real time based on the original data and initial characteristics. Based on the real-time value of the coupling stress index and combined with the preset mapping function, the scaling factors of the current limit, speed limit and controller bandwidth limit are dynamically adjusted to determine the real-time feedforward compensation coefficient and reconstruct the safety boundary. The optimization problem-solving module is used to construct a real-time optimization problem with tracking error, energy loss, and control smoothness as optimization objectives, using a safety boundary as a hard constraint; based on the real-time state, the optimization problem is solved to obtain the optimal current loop proportional gain and integral gain; The control parameter application module is used to apply the solved optimal control parameters to the current regulator to achieve real-time optimized control of the robot drive motor.

8. The real-time optimization system for robot drive modules based on multimodal perception according to claim 7, characterized in that, The correlation determination module includes: The electrothermal correlation unit is used to obtain the electrothermal correlation by analyzing the covariance relationship between the effective value of the quadrature axis current and the winding temperature rise rate over time, and by weighting and correcting the deviation between the current speed and the rated speed. The electro-vibration correlation unit is used to perform spectral analysis on the direct-axis current signal and the axial vibration acceleration signal respectively, calculate the coherence intensity of the current spectrum component and the vibration spectrum component at the main current harmonic frequency, and perform weighted summation with the importance of each harmonic as the weight to obtain the electro-vibration correlation degree. The thermal-vibration correlation unit is used to calculate the Pearson correlation coefficient between the winding temperature time series signal and the root mean square value vibration acceleration time series signal within the window, thus obtaining the thermal-vibration correlation.

9. The real-time optimization system for robot drive modules based on multimodal perception according to claim 7, characterized in that, The security boundary reconstruction module includes: The basic model unit is used to construct the basic model of the coupled stress index. The model is based on the product of normalized real-time torque current command, winding temperature rise rate, electro-vibration correlation degree and vibration amplitude, and a power function with the ratio of rated voltage to real-time voltage as the base. The compensation model unit is used to construct a multi-physics cross-coupling compensation model. The model linearly combines the product of the electro-thermal correlation degree and the relative temperature rise, as well as the product of the thermal-vibration correlation degree and the relative vibration intensity, as compensation terms for the basic model. The stress index unit is used to weight and fuse the output of the base model and the output of the compensation term to obtain the final coupled stress index. The formula is: CSI(t)=γ1×CSIbase(t)+γ2×CSIcoup(t), where CSI(t) is the coupled stress index at time t, CSIbase(t) is the output value of the base model, CSIcoup(t) is the output value of the cross-coupling compensation term, and γ1 and γ2 are weight coefficients.

10. The real-time optimization system for robot drive modules based on multimodal perception according to claim 7, characterized in that, The security boundary reconstruction module also includes: The current scaling factor unit is used to input the real-time coupling stress index, current adjustment threshold and real-time bus voltage drop into an S-shaped decay function to calculate a current limit scaling factor between 0 and 1. The speed scaling factor unit is used to input the real-time coupling stress index, speed adjustment threshold and real-time winding temperature into an exponential decay function to calculate a speed limit scaling factor between 0 and 1. The bandwidth scaling factor unit is used to input the real-time coupling stress index, bandwidth adjustment threshold and real-time electro-vibration correlation degree into a hyperbolic tangent adjustment function to calculate a bandwidth scaling factor between 0 and 1. The limit determination unit is used to obtain the real-time current limit, real-time speed limit, and controller parameter value range based on the scaling factor.

Images