A load adaptive control method and system for a robot joint module

Abstract

The application belongs to the field of robot joint driving control, and is a load adaptive control method and system of a robot joint module, comprising the following steps: S1: synchronously collecting operation data, adopting a weighted recursive adaptive sliding window filtering algorithm, combining speed noise reduction, and performing temperature deviation calibration; S2: on-line solving of load parameters, introduction of a nonlinear temperature coupling correction factor to correct the result and determination of the load working condition grade; S3: matching control gain according to the working condition, generating inertia feedforward, disturbance observation and thermal error compensation term; S4: superposition to double-loop instruction driving operation and feedback; the application can improve the identification accuracy of external load torque, equivalent rotational inertia and impact disturbance under different temperature conditions, and widen the effective temperature application interval of the control scheme.

Description

Technical Field

[0001] This invention belongs to the field of robot joint drive control technology, and in particular relates to a load adaptive control method and system for robot joint modules. Background Technology

[0002] The core component for motion execution in industrial robots and collaborative robots is the joint module. The module integrates drive motors, reduction mechanisms, and various types of sensors, and achieves precise motion output through torque closed-loop control. In the working environment, the load weight, external impact force, and ambient temperature will continuously change. The measured values ​​of the sensors will shift with the temperature rise, and the inherent parameters inside the model will drift synchronously.

[0003] In traditional control architectures, the signal filtering, load identification, and gain adjustment modules operate independently, with no real-time data exchange channels between modules. The filter window length and control gain are all preset fixed constants. When the module is subjected to variable loads or instantaneous impacts, the fixed parameters cannot be synchronously matched with the real-time operating conditions. Measurement deviations and model errors caused by temperature cannot be corrected, and there are significant limitations in the smoothness of motion and the accuracy of load identification.

[0004] Most existing robot joint module torque control schemes use fixed control parameters to complete motion regulation, relying solely on the basic model to solve for load parameters. Temperature changes during the operation of the joint module can cause sensor measurement deviations and model parameter drift, directly reducing the accuracy of load parameter identification. Conventional filtering methods use a fixed window length for signal processing, which cannot adapt to the dynamic signal characteristics changes caused by large-range rotational speed variations of joints. Traditional control strategies can only generate basic feedforward compensation quantities and do not establish corresponding compensation mechanisms for deformation errors caused by temperature. The control gain switching method often adopts a step switching form, which can easily induce joint motion fluctuations during the load state change process. The various parameter calculation units are independent of each other, and there is a lack of effective information exchange channels between the units, making it impossible to dynamically adjust the signal acquisition and processing strategy based on the real-time load status. Summary of the Invention

[0005] This invention proposes a load adaptive control method and system for robot joint modules to overcome the defect of decreased accuracy of load identification parameters caused by temperature drift. It corrects the identification results by means of nonlinear correction to adapt to the wide temperature range working environment of the joint module. At the same time, it solves the problem that fixed processing parameters cannot match dynamic operating conditions. It dynamically adjusts the signal acquisition and filtering strategy based on the operating condition information output by load identification, and achieves smooth transition of control gain to suppress motion fluctuations caused by load changes.

[0006] A load adaptive control method for a robot joint module includes: S1: Synchronously collect phase current, rotation speed, position error and temperature operation data of robot joint module, use weighted recursive adaptive sliding window filtering algorithm combined with rotation speed for noise reduction processing, perform temperature deviation calibration, and output standardized operation data; S2: Based on the standardized operating data, the external load torque, equivalent moment of inertia and impact disturbance are solved online. A nonlinear temperature coupling correction factor is introduced to correct the identification results. The load condition level is determined based on the corrected identification parameters. S3: Based on the load condition level and the identified parameters, match the control gain and generate inertia feedforward compensation term, disturbance observation compensation term and thermal error compensation term; S4: The control gain and compensation terms are superimposed on the current loop and torque loop control commands to drive the robot joint module to run, and the real-time feedback data is sent back to step S1 to form a closed loop.

[0007] Preferably, step S1 uses a weighted recursive adaptive sliding window filtering algorithm combined with rotational speed for noise reduction. Specifically, it includes: matching the dynamic window length by linear mapping according to the real-time rotational speed value of the joint module, configuring the time-series weighting coefficients monotonically increasing according to the sampling time sequence, and performing weighted average calculation on the original collected data.

[0008] Preferably, the temperature deviation calibration in step S1 includes: comparing the real-time temperature with a preset temperature threshold; when the real-time temperature exceeds the preset threshold range, calling the data correction model to perform deviation correction on the phase current data and rotation speed data.

[0009] Preferably, in step S2, a nonlinear temperature coupling correction factor is introduced to correct the identification results. Specifically, this includes obtaining the temperature-corrected identification parameters based on the original model identification parameters, the nonlinear temperature correction coefficient, the difference between the real-time temperature and the reference temperature, and the module's rated reference operating temperature using the principle of exponential calculation.

[0010] Preferably, in step S2, the load condition level is determined based on the corrected identification parameters, specifically including: determining the load condition as light load, rated load, heavy load, or impact disturbance based on the corrected external load torque, equivalent moment of inertia, and impact disturbance values.

[0011] This invention also discloses a load adaptive control system for a robot joint module, comprising: The data preprocessing module is used to synchronously collect phase current, rotation speed, position error and temperature operation data of robot joint modules, perform noise reduction filtering and temperature deviation calibration on the raw operation data, and output standardized operation data. The load identification module, connected to the data preprocessing module, is used to solve the external load torque, equivalent moment of inertia and impact disturbance online based on the standardized operating data, introduce a nonlinear temperature coupling correction factor to correct the identification results, and determine the load condition level based on the corrected identification parameters. An adaptive compensation module, connected to the load identification module, is used to match the control gain according to the load condition level and identification parameters, and generate an inertia feedforward compensation term, a disturbance observation compensation term, and a thermal error compensation term. The instruction-driven module, connected to the adaptive compensation module, is used to superimpose the control gain, inertia feedforward compensation term, disturbance observation compensation term and thermal error compensation term onto the current loop and torque loop control commands to drive the robot joint module to run. The instruction-driven module also sends real-time operation feedback data back to the data preprocessing module, forming a closed loop.

[0012] Preferably, the data preprocessing module is further configured to: employ a weighted recursive adaptive sliding window filtering algorithm, linearly map and match the dynamic window length according to the real-time rotation speed value of the joint module, and perform noise reduction processing by configuring the time-series weighting coefficients monotonically increasing according to the sampling time sequence; The real-time temperature is compared with the preset temperature threshold. When the real-time temperature exceeds the preset threshold range, deviation correction is performed on the phase current data and speed data.

[0013] Preferably, the load identification module further includes: obtaining temperature-corrected identification parameters based on the original model identification parameters, nonlinear temperature correction coefficient, the difference between real-time temperature and reference temperature, and the module's rated reference operating temperature using the principle of exponential calculation. Based on the corrected values ​​of external load torque, equivalent moment of inertia, and impact disturbance, the load condition is determined as light load, rated load, heavy load, or impact disturbance.

[0014] Preferably, the adaptive compensation module further includes: Establish a continuous nonlinear mapping rule between load conditions and control gain, and complete the continuous and smooth transition of control gain through interpolation calculation when the load condition level changes. Calculate the temperature gradient parameters of real-time temperature data, dynamically adjust the thermal error compensation coefficient based on the temperature gradient value, and generate a real-time thermal error compensation term.

[0015] Preferably, the instruction-driven module further includes: employing a normalized weighted adaptive fusion algorithm to allocate superimposed weights for the inertia feedforward compensation term, the disturbance observation compensation term, and the thermal error compensation term based on the current disturbance level and load size.

[0016] The present invention has the following beneficial effects: 1. This invention can improve the accuracy of identifying external load torque, equivalent moment of inertia and impact disturbance under different temperature conditions, and broaden the effective temperature range of the control scheme.

[0017] 2. The present invention can dynamically adjust the signal processing strategy according to the real-time load conditions, and at the same time realize the continuous and smooth switching of control gain, reduce motion fluctuations during load state changes, and improve the stability of joint module operation.

[0018] 3. By adding a calculation step related to thermal error compensation, this invention forms a multi-dimensional compensation amount collaborative control mechanism, which enhances the adaptability of the joint module to variable load and instantaneous impact disturbance scenarios. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the steps of a load adaptive control method for a robot joint module according to the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions in the embodiments of this invention will be clearly described below in conjunction with the examples.

[0021] Example 1 like Figure 1 As shown, this invention proposes a load adaptive control method and system for a robot joint module, wherein the load adaptive control method for the robot joint module specifically includes the following steps: S1. Synchronously collect phase current, rotation speed, position error and temperature operation data of robot joint module, perform noise reduction filtering and outlier removal preprocessing on the collected raw operation data, and output standardized operation data.

[0022] S2. Import the standardized operating data, solve the external load torque, equivalent moment of inertia and impact disturbance online through the load identification model, and determine the load condition level of the joint module by combining the temperature parameters.

[0023] S3. Based on the load condition level and identification parameters, match the control gain corresponding to the preset gain set, and synchronously generate inertia feedforward compensation term, disturbance observation compensation term and thermal error compensation term.

[0024] S4. The control gain, inertia feedforward compensation term, disturbance observation compensation term, and thermal error compensation term are superimposed on the current loop and torque loop control commands to drive the robot joint module to run.

[0025] After the robot joint module is powered on, it first executes the processing flow corresponding to S1. The signal acquisition unit continuously and synchronously acquires the phase current, rotation speed, position error and temperature raw operating data output by the sensors inside the joint module. All raw operating data are directly sent to the data preprocessing unit.

[0026] The data preprocessing unit executes two types of processing flows in parallel. The first type of flow executes S1-1 dynamic data processing, which calls the weighted recursive adaptive sliding window filtering algorithm to perform noise reduction filtering on the original running data.

[0027] The system reads the joint module rotation speed value in real time and completes the rotation speed and... The linear mapping calculation determines the dynamic window length corresponding to the current sampling period and the time-series weighting coefficients. According to the monotonically increasing configuration of sampling time sequence, sampling points closer to the current sampling time are assigned higher weights, which are then substituted into the filtering model: in, The filtered output data represents the result of the nth sampling point after weighted recursive adaptive sliding window filtering. It is dimensionless or has the same unit as the input data (the sensor output is a dimensionless sampled value).

[0028] The dynamic window length associated with rotational speed represents the total number of sampling points contained in the filter window during the current sampling period, expressed in sampling points. This parameter is linearly mapped and matched based on the real-time rotational speed of the joint module; the higher the rotational speed, the more dynamic the window. The larger the value.

[0029] The time-series weighting coefficient represents the data of the i-th sampling point within the window. The assigned weights are dimensionless; these coefficients are configured to monotonically increase according to the sampling time sequence, with the weights increasing the closer the sampling point is to the current sampling time (index n). The larger the coefficient, the better.

[0030] The raw data represents the original output value of the sensor at the i-th sampling point within the window, such as phase current, rotational speed, etc. It is dimensionless or represents the original unit of sensor output.

[0031] Perform filtering operations to suppress random noise interference from the sensor.

[0032] The second type of process executed by the data preprocessing unit is S1-2 temperature deviation calibration. The unit loads the pre-calibrated temperature threshold and compares the real-time acquired temperature with the threshold. When the real-time temperature exceeds the threshold range, the data correction model is activated to perform deviation correction on the phase current data and speed data. After the filtered calculation output is processed by temperature deviation calibration, standardized operating data is obtained. Standardized operating data is sent as the first output data stream to the load parameter identification unit corresponding to S2; Meanwhile, the load parameter identification unit continuously transmits the calculated load condition level back to the data preprocessing unit. The data preprocessing unit changes the acquisition frame rate according to the received condition level. High frame rate sampling mode is enabled for impact disturbance conditions and heavy load conditions, while low frame rate sampling mode is switched for light load steady state conditions to reduce the unit's computing power consumption.

[0033] The load parameter identification unit receives standardized operating data from S1, loads a preset load identification model, and substitutes in phase current, speed fluctuation, and position error information to perform numerical solutions and obtain the initial state. That is, the original external load torque Original equivalent moment of inertia and the original impact disturbance .

[0034] Furthermore, the preset load identification model is a recursive least squares load identification model based on the joint dynamics equations, using phase current, rotational speed, and position error as identification inputs for online numerical solution. Specifically, the standardized phase current is converted into electromagnetic output torque based on the torque constant of the joint module motor, angular acceleration is calculated, and impact disturbance characterization quantities are constructed, namely: in, For the first Rotation speed during the sampling period For the first Position error of the sampling period, The sampling period is It is the torque constant. To standardize the phase current, , These are preset weighting coefficients.

[0035] Based on the discrete dynamics of the joint module, the identification equation is established: in, In order to identify the observation, Preset the viscous damping coefficient for the joint module. For external load torque, For equivalent rotational inertia, This refers to the impact disturbance.

[0036] Write the parameters to be identified as a parameter vector, and the regression vector as: The recursive least squares algorithm is used for online updates. The update process is as follows: in, For the first Parameter estimates for the sampling period, Let covariance matrix be the variance matrix. Forgetting factor, The identity matrix is ​​used. Through the above iterative calculations, the original external load torque, original equivalent moment of inertia, and original impact disturbance are obtained in real time; when the rate of change of position error and the speed fluctuation increase synchronously, the regression vector... As the magnitude increases, the original impact disturbance obtained from the solution also increases, thus characterizing the instantaneous impact condition.

[0037] The unit initiates the S2-1 model temperature correction process, retrieves the real-time temperature value synchronously transmitted by S1, and calculates the real-time temperature and... The difference between them Substituting the nonlinear temperature coupling correction model: in, These are the temperature-corrected identification parameters, representing the final identification result obtained after nonlinear temperature coupling correction. Correspondingly, they can be the corrected external load torque, the corrected equivalent moment of inertia, or the corrected impact disturbance.

[0038] These are the original identification parameters of the model, representing the initial identification results obtained by solving the load identification model online; correspondingly, they can be the original external load torque, the original equivalent moment of inertia, or the original impact disturbance.

[0039] It is a nonlinear temperature correction coefficient, a dimensionless constant, pre-calibrated based on the full-temperature operating characteristics of the joint module, used to characterize the nonlinear characteristics of the influence of temperature on the identification parameters.

[0040] The difference between the real-time temperature and the reference temperature represents the deviation of the current joint module temperature from the standard operating temperature, expressed in °C.

[0041] The rated reference operating temperature of the module indicates the preset standard operating temperature point of the joint module, in °C.

[0042] After completing the identification parameter correction calculation, the following results are obtained: .

[0043] After temperature correction, the identification parameters are sent to the S2-2 load condition classification process. The system sets four threshold ranges based on... The system comprehensively judges the corresponding external load torque, equivalent moment of inertia, and impact disturbance, and outputs four load condition levels: light load, rated load, heavy load, and impact disturbance.

[0044] The corrected identification parameters are output synchronously with the load condition level. One data stream is sent to the adaptive compensation generation unit corresponding to S3, and the other data stream is used to match the running error data returned by S3 to achieve iterative optimization of the load identification model coefficients.

[0045] The adaptive compensation generation unit receives the identification parameters and load condition level from S2, and starts the gain smoothing control process of S3-1. The unit has a pre-stored gain set, and a continuous nonlinear mapping rule is established between different gain parameters and the condition level within the set. The system retrieves the corresponding control gain according to the current load condition level. When the condition level changes, interpolation is used to complete the continuous transition of the gain parameter, avoiding the joint jitter phenomenon caused by sudden gain changes.

[0046] The unit synchronously generates inertia feedforward compensation terms and disturbance observation compensation terms, and simultaneously starts the S3-2 thermal error dynamic compensation process. The unit continuously reads the real-time temperature data transmitted by S1, calculates the temperature gradient parameters based on the temperature difference between adjacent sampling periods, and dynamically adjusts the thermal error compensation coefficient based on the temperature gradient value, and generates real-time thermal error compensation terms in combination with the joint thermal deformation characteristics.

[0047] The adaptive compensation generation unit transmits all compensation parameters to the dual-loop instruction synthesis drive unit corresponding to S4. The unit executes the S4-1 compensation weight fusion process, enabling the normalized weight adaptive fusion algorithm. It dynamically allocates the superposition weights of the inertia feedforward compensation term, disturbance observation compensation term, and thermal error compensation term based on the current disturbance level and load size, maintaining the total weight value within a fixed normalized range. After weight allocation, the unit superimposes the control gain and various compensation parameters onto the current loop control command and the torque loop control command.

[0048] Once the instruction is generated, it is output to the joint module driver hardware, which drives the joint module to produce the corresponding motion output.

[0049] During operation, the joint module continuously outputs real-time operation feedback data. The dual-ring instruction synthesis drive unit executes the S4-2 real-time iterative control process, collects the real-time operation feedback data, and sends the feedback data back to the S1 data preprocessing unit to start a new round of complete calculation process.

[0050] The dual-ring instruction synthesis drive unit synchronously extracts the actual motion error of the module as the running error data and transmits the running error data to the S2 load parameter identification unit. The load parameter identification unit compares the predicted output of the identification model with the actual running error and continuously iterates and adjusts the internal coefficients of the load identification model to gradually reduce the identification calculation deviation.

[0051] During the execution of the complete operation process, the various units maintain continuous bidirectional data interaction; the load parameter identification unit sends the operating condition level signal to the data preprocessing unit to realize the adjustment of the acquisition strategy, and the adaptive compensation generation unit sends back the running error data to the load parameter identification unit to complete the identification model optimization.

[0052] When the joint module is subjected to sudden external forces that cause impact disturbances, S1 automatically increases the sampling frame rate, and the filtering model adaptively increases. Improve signal stability; S2 temperature correction model synchronously compensates for identification drift caused by temperature, and the working condition classification module determines that it has entered the impact disturbance working condition; S3 switch to the control gain adapted to the impact working condition, and at the same time increase the superposition weight of the disturbance observation compensation term; S4 complete the fusion of multiple compensation terms and update the control command, and the whole process is executed synchronously to complete the adaptive regulation under load change.

[0053] Example 2 This embodiment is based on the control method described in Embodiment 1. A test platform is built to verify the computation process. The test platform is equipped with a computation processing unit, a signal acquisition unit, a joint module drive unit, and a sensor acquisition unit. The sensor acquisition unit is deployed on the robot joint module body and is used to output raw operating data such as phase current, rotational speed, position error, and temperature. The signal acquisition unit receives the output signals from the sensor acquisition unit and transmits them to the computation processing unit. The computation processing unit has a built-in program that implements all processes S1 to S4 described in Embodiment 1. The computation processing unit outputs control commands to the joint module drive unit.

[0054] Three typical test conditions were set during the test: the first condition was constant light load uniform motion; the second condition was rated load reciprocating motion; and the third condition was variable load with instantaneous impact disturbance. The internal program of the processing unit strictly followed the steps described in Example 1. The program encapsulates a weighted recursive adaptive sliding window filtering module, a temperature deviation calibration module, a load identification calculation module, a nonlinear temperature correction module, a condition classification module, a gain smoothing scheduling module, a thermal error compensation calculation module, and a weight fusion module.

[0055] The program executes the corresponding filtering, identification, correction, and compensation calculations in the logical order of steps S1 to S4 described above.

[0056] The testing process continuously records key intermediate variables, including the filtered output. Parameters identified before correction Corrected identification parameters Load condition level, values ​​of various compensation items, and final output control commands.

[0057] Example data table of test data collection records: As shown in the table, when the rotational speed increases from 22.4 r / min to 61.1 r / min, the program automatically increases the speed. When the real-time temperature rises, relatively It produces a significant correction offset; when an instantaneous impact disturbance is applied externally, the system determines that the operating condition level has switched to impact disturbance and adjusts various compensation parameters synchronously.

[0058] During the sampling period when the instantaneous impact disturbance occurs, the processing unit receives the impact disturbance condition signal transmitted by S2, increases the signal acquisition frame rate, and adjusts the filtering window. Synchronous adjustment suppresses instantaneous noise; S3 dynamically increases the weight of disturbance observation compensation terms, and the control commands generated after fusion quickly offset motion deviations caused by external shocks.

[0059] The test platform continuously adjusts the ambient temperature, covering the entire range from low-temperature startup to high-temperature steady state of the module, and the nonlinear temperature coupling correction model continues to play its role.

[0060] If the temperature correction calculation corresponding to S2-1 is removed The deviation from the actual load parameters of the module continues to increase; after enabling the exponential temperature correction model, the identification deviation remains within a stable range under different ambient temperatures.

[0061] The computing unit continuously sends operational error data to the load identification module. As the sampling period continues to accumulate, the coefficients within the load identification model are continuously iterated and adjusted. The difference between the actual load parameters of the module gradually decreases.

[0062] When the test conditions smoothly transition from heavy load to rated load and light load, the S3 gain smoothing control module enables interpolation calculation to control the continuous change of gain parameters without producing abrupt changes, and there is no obvious jitter during the movement of the joint module.

[0063] The S1-2 temperature deviation calibration continuously corrects the raw phase current and speed data collected under high temperature conditions, eliminates the original measurement deviations introduced by sensor temperature drift, ensures the reliability of standardized operating data, and provides a stable input data source for subsequent load identification calculations.

[0064] Example 3 This embodiment applies the robot joint module load adaptive control method described in Embodiment 1 to the joint control system of a multi-degree-of-freedom robotic arm. Each joint of the robotic arm is independently equipped with a signal acquisition unit and a drive unit. Each set of calculation programs independently executes the complete control process described in Embodiment 1. There is no parameter coupling between adjacent joint calculation units. Each joint independently completes load identification and adaptive compensation based on its own force state.

[0065] The robotic arm performs grasping and handling operations, which can be divided into three typical stages: unloaded movement, grasping and moving the workpiece under load, and the instantaneous impact caused by the workpiece contacting an external obstacle. During the unloaded movement stage, each joint module of the robotic arm is in a light-load condition. After grasping the workpiece, it switches to the rated load or heavy-load condition. The moment the workpiece contacts an obstacle, it enters the impact and disturbance condition.

[0066] The robotic arm control system independently allocates computing resources to each joint. Each joint's S1 process independently collects its corresponding phase current, rotational speed, position error, and temperature data, and independently performs weighted recursive adaptive sliding window filtering and temperature deviation calibration calculations.

[0067] Different joints rotate at different speeds, and each joint's computational unit calculates its own speed independently. Numerical values, the filter window lengths of adjacent joints do not need to be kept uniform.

[0068] Standardized operational data for each joint are fed into their respective load identification models for independent solution. And through the exponential temperature correction model: get The load condition level is determined independently. When the end joint grasps a heavy object and bears a large load, the end joint is determined to be under heavy load. When the joint near the base bears a relatively small load, it is determined to be under rated load. The two sets of joints independently call up their respective matched control gains to generate independent inertia feedforward compensation terms, disturbance observation compensation terms, and thermal error compensation terms.

[0069] When the workpiece being transported comes into contact with a fixed obstacle, generating a momentary impact, the end effector quickly identifies the impact disturbance. The normalized weighted adaptive fusion algorithm corresponding to S4 increases the superposition weight of the disturbance observation compensation term, generating a compensated control command to suppress the position fluctuation caused by the impact. Simultaneously, the end effector continuously transmits real-time operational feedback data back to its own S1 preprocessing unit, continuously iterating the entire calculation process. The operational error data generated by the end effector is continuously transmitted back to its own load identification unit, continuously optimizing the identification model coefficients.

[0070] During the operation of a multi-degree-of-freedom robotic arm, the joints continuously heat up. After a long period of operation, the joint temperature continues to rise. The S2-1 nonlinear temperature coupling correction model continuously corrects the identification parameters to offset the identification drift caused by the temperature rise. The S3-2 thermal error dynamic compensation process continuously updates the thermal error compensation items based on the real-time temperature gradient to compensate for the position control deviation caused by joint thermal deformation.

[0071] The robotic arm performs reciprocating transport tasks, continuously performing tens of thousands of cycles of motion. The load state of each joint changes periodically, and the entire adaptive control method continuously updates the parameters.

[0072] The complete data link for a single operation cycle is as follows: joint sensors collect raw operating data → S1 filtering and temperature calibration outputs standardized operating data → S2 load identification, temperature correction, and operating condition classification → S3 gain matching and multi-compensation term generation → S4 weight fusion outputs dual-loop control commands to drive the joint → operating feedback data is sent back to S1, and operating errors are sent back to S2. The entire data link is executed continuously in a loop.

[0073] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A load adaptive control method for a robot joint module, characterized in that, include: S1: Synchronously collect phase current, rotation speed, position error and temperature operation data of robot joint module, use weighted recursive adaptive sliding window filtering algorithm combined with rotation speed for noise reduction processing, perform temperature deviation calibration, and output standardized operation data; S2: Based on the standardized operating data, the external load torque, equivalent moment of inertia and impact disturbance are solved online. A nonlinear temperature coupling correction factor is introduced to correct the identification results. The load condition level is determined based on the corrected identification parameters. S3: Based on the load condition level and the identified parameters, match the control gain and generate inertia feedforward compensation term, disturbance observation compensation term and thermal error compensation term; S4: The control gain and compensation terms are superimposed on the current loop and torque loop control commands to drive the robot joint module to run, and the real-time feedback data is sent back to step S1 to form a closed loop.

2. The load adaptive control method for a robot joint module according to claim 1, characterized in that, In step S1, a weighted recursive adaptive sliding window filtering algorithm is used in conjunction with the rotation speed for noise reduction. Specifically, this includes: matching the dynamic window length by linear mapping based on the real-time rotation speed value of the joint module, configuring the time-series weighting coefficients monotonically increasing according to the sampling time sequence, and performing weighted average calculation on the original collected data.

3. The load adaptive control method for a robot joint module according to claim 1, characterized in that, The temperature deviation calibration in step S1 specifically includes: comparing the real-time temperature with a preset temperature threshold; when the real-time temperature exceeds the preset threshold range, calling the data correction model to perform deviation correction on the phase current data and rotation speed data.

4. The load adaptive control method for a robot joint module according to claim 1, characterized in that, In step S2, a nonlinear temperature coupling correction factor is introduced to correct the identification results. Specifically, this includes obtaining the temperature-corrected identification parameters based on the original model identification parameters, the nonlinear temperature correction coefficient, the difference between the real-time temperature and the reference temperature, and the module's rated reference operating temperature using the exponential calculation principle.

5. The load adaptive control method for a robot joint module according to claim 1, characterized in that, In step S2, the load condition level is determined based on the corrected identification parameters. Specifically, the load condition is determined as light load, rated load, heavy load, or impact disturbance based on the corrected external load torque, equivalent moment of inertia, and impact disturbance values.

6. A load adaptive control system for a robot joint module, characterized in that, include: The data preprocessing module is used to synchronously collect phase current, rotation speed, position error and temperature operation data of robot joint modules, perform noise reduction filtering and temperature deviation calibration on the raw operation data, and output standardized operation data. The load identification module, connected to the data preprocessing module, is used to solve the external load torque, equivalent moment of inertia and impact disturbance online based on the standardized operating data, introduce a nonlinear temperature coupling correction factor to correct the identification results, and determine the load condition level based on the corrected identification parameters. An adaptive compensation module, connected to the load identification module, is used to match the control gain according to the load condition level and identification parameters, and generate an inertia feedforward compensation term, a disturbance observation compensation term, and a thermal error compensation term. The instruction-driven module, connected to the adaptive compensation module, is used to superimpose the control gain, inertia feedforward compensation term, disturbance observation compensation term and thermal error compensation term onto the current loop and torque loop control commands to drive the robot joint module to run. The instruction-driven module also sends real-time operation feedback data back to the data preprocessing module, forming a closed loop.

7. The load adaptive control system for a robot joint module according to claim 6, characterized in that, The data preprocessing module is further configured to: employ a weighted recursive adaptive sliding window filtering algorithm, linearly map and match the dynamic window length according to the real-time rotation speed value of the joint module, and perform noise reduction processing by configuring the time-series weighting coefficients monotonically increasing according to the sampling time sequence; The real-time temperature is compared with the preset temperature threshold. When the real-time temperature exceeds the preset threshold range, deviation correction is performed on the phase current data and speed data.

8. The load adaptive control system for a robot joint module according to claim 6, characterized in that, The load identification module also includes: based on the original identification parameters of the model, the nonlinear temperature correction coefficient, the difference between the real-time temperature and the reference temperature, and the rated reference operating temperature of the module, the temperature-corrected identification parameters are obtained using the exponential calculation principle. Based on the corrected values ​​of external load torque, equivalent moment of inertia, and impact disturbance, the load condition is determined as light load, rated load, heavy load, or impact disturbance.

9. The load adaptive control system for a robot joint module according to claim 6, characterized in that, The adaptive compensation module also includes: Establish a continuous nonlinear mapping rule between load conditions and control gain, and complete the continuous and smooth transition of control gain through interpolation calculation when the load condition level changes. Calculate the temperature gradient parameters of real-time temperature data, dynamically adjust the thermal error compensation coefficient based on the temperature gradient value, and generate a real-time thermal error compensation term.

10. A load adaptive control system for a robot joint module according to claim 6, characterized in that, The instruction-driven module further includes: using a normalized weight adaptive fusion algorithm to allocate superimposed weights for inertia feedforward compensation, disturbance observation compensation, and thermal error compensation based on the current disturbance level and load size.

Images