A hollow type double-encoder joint module end encoder error identification and compensation method

Abstract

The present application relates to a hollow double-encoder joint module end encoder error identification and compensation method, belonging to the field of robot joint precision control technology. In view of the problem that the end encoder reading deviates from the real end angle due to the hollow shaft cantilever support, the eccentricity of the code disc installation and the vibration transmission of the motor rotor, the motor end angle, the end encoder angle, the q-axis current, the speed and the external reference angle are collected in the calibration stage to establish the end encoder error sequence; the rotor synchronous vibration error is extracted through rotor angle domain frequency analysis, and the remaining q-axis current after stripping the friction current, the angle periodic characteristics and the harmonic prior are combined to enhance the KAN network to predict the residual error of the hollow shaft and the code disc shaking; the two types of compensation amounts are combined online to correct the position closed-loop target value or feedback value, thereby improving the joint end positioning accuracy.

Description

Technical Field

[0001] This invention relates to the field of precision control technology for robot joint modules, specifically to a method for identifying and compensating for end encoder errors in hollow dual encoder joint modules, and more particularly to a method for identifying, modeling, and compensating for end encoder measurement errors caused by hollow shaft cantilever support, end encoder installation deviation, and motor rotor vibration transmitted through bearings or structural coupling. Background Technology

[0002] Robot joint modules typically consist of a motor, reducer, drive controller, encoder, and output structure. To achieve high-precision position control, existing high-performance joint modules usually employ a dual-encoder structure, with a motor-end encoder at the motor end and an end encoder at the reducer output or joint end. The motor-end encoder is mainly used for motor commutation, current loop control, speed loop control, and rotor position feedback, while the end encoder is mainly used to obtain the joint output angle, thereby improving the closed-loop control accuracy of the joint end position.

[0003] In hollow joint modules, a hollow shaft is typically incorporated at the joint output end to accommodate internal wiring, ventilation, weight reduction, or end-mount installation requirements. The hollow shaft serves both as the joint output structure and as a channel for cables, air hoses, or other functional components. To obtain the output angle, an end encoder disc is usually fixed to the rear extension of the hollow shaft, and the signal is read by the end encoder reading head. This structure enables output angle measurement within the hollow joint, but it also makes the end encoder reading susceptible to the assembly condition of the hollow shaft and the code disc.

[0004] First, when the end encoder is fixed to the extended section at the rear of the hollow shaft, the encoder has a certain cantilever characteristic relative to the support position of the joint output end. When the joint end is subjected to radial load, gravity component, or external disturbance, the extended section of the hollow shaft may experience slight deflection, warping, or radial oscillation, causing changes in the coaxiality, clearance, and orientation of the end encoder relative to the end encoder reading head, thereby causing the end encoder angle reading to deviate from the true end angle.

[0005] Secondly, during the pasting, pressing, locking, or other installation processes, it is difficult to ensure that the end encoder disk is completely coaxial with the hollow shaft, which may result in installation eccentricity, installation tilt, end face runout, or radial runout. These installation errors will cause the end encoder reading to change periodically with the output end angle, leading to an absolute positioning error in the joint module that is related to the output end angle.

[0006] Secondly, the hollow shaft and the motor rotor shaft are typically connected via bearings, a housing, or adjacent supporting structures. During rotation, the motor rotor may experience disturbances such as rotor imbalance, assembly eccentricity, electromagnetic force fluctuations, or bearing vibration. These disturbances are transmitted to the hollow shaft and the end encoder via the bearing or structural coupling path, causing the end encoder to produce a slight oscillation synchronized with the motor rotor's rotation. Due to the reduction ratio relationship between the motor rotor angle and the output angle, this type of error may manifest as a high-order error close to the reduction ratio order in the output angle domain, while in the motor rotor angle domain it manifests as a first-order or multi-order rotor synchronization error.

[0007] Existing end-effector control methods for dual-encoder joint modules typically assume the end-effector angle is the true end-effector angle, or only use zero-point calibration, linear correction, and lookup table compensation to handle end-effector errors. These methods struggle to distinguish between different error sources such as hollow shaft cantilever deflection, end-effector code disk installation deviation, and motor rotor vibration transmission. They also fail to utilize the correlation between motor q-axis current, motor speed, motor rotor angle, and end-effector angle to compensate for complex nonlinear measurement errors.

[0008] Therefore, a method for identifying and compensating for end encoder errors in hollow dual encoder joint modules is needed. While retaining the online measurement function of the end encoder, the method identifies the source of the end encoder reading error and compensates it in layers, so that the feedback value or the target value of the position closed loop can be corrected according to the error compensation amount, thereby improving the end positioning accuracy of the hollow dual encoder joint module. Summary of the Invention

[0009] The purpose of this invention is to provide a method for identifying and compensating for end encoder errors in a hollow dual encoder joint module. This method overcomes the problem in existing hollow dual encoder joint modules where the end encoder readings cannot accurately reflect the true end angle due to the influence of the hollow shaft cantilever support, encoder installation deviation, and motor rotor vibration transmission. The method establishes an end encoder error sequence, performs angular domain frequency analysis on the error sequence, extracts the rotor synchronous vibration error component, establishes a speed-current baseline model to extract load-sensitive current characteristics, and uses a harmonic prior-enhanced Kolmogorov-Arnold compensation network to predict the residual error compensation amount. Finally, based on the total error compensation amount, the feedback value or target value of the end position closed-loop control is corrected, thereby improving the end positioning accuracy of the joint module.

[0010] The technical solution of this invention provides a method for identifying and compensating for errors in the end encoder of a hollow dual encoder joint module. The hollow dual encoder joint module includes a motor, a reducer, a motor-end encoder, a hollow shaft, an end code disk disposed on the hollow shaft, and an end encoder for reading the end code disk. The method includes the following steps:

[0011] S1. During the calibration phase, acquire the motor rotor angle output from the encoder at the motor end. The end encoder angle output by the end encoder q-axis current of motor Motor speed and the end reference angle output by an external high-precision angle sensor ;

[0012] S2, based on the angle of the end encoder With the aforementioned end reference angle Establish the end encoder error sequence ;

[0013] S3, the error sequence of the end encoder An angular domain frequency analysis was performed to extract the error components that are synchronized with the rotation of the motor rotor, and a rotor synchronous vibration compensation model was established based on the error components.

[0014] S4. Based on the motor q-axis current under the reference operating conditions. With motor speed Establish a speed-current baseline model, and use the current motor q-axis current as the basis. Subtract from the current motor speed The corresponding speed-dependent baseline current is used to obtain the load-sensitive current characteristic quantity. ;

[0015] S5, the load-sensitive current characteristic quantity The motor rotor angle The angle of the end encoder and the motor speed By inputting a harmonic prior-enhanced Kolmogorov-Arnold compensation network, the residual error compensation caused by hollow shaft cantilever deflection, end encoder installation deviation, and end encoder wobble is obtained. ;

[0016] S6. The rotor synchronous vibration compensation amount output by the rotor synchronous vibration compensation model. With the residual error compensation amount Combined into the total error compensation amount of the end encoder ;

[0017] S7. During the online control phase, based on the total error compensation amount of the end encoder... Correct the feedback or target value of the end position closed-loop control to compensate for the end encoder reading error.

[0018] Further, in step S2, the end encoder error sequence Established according to the following relationship:

[0019] = -

[0020] in, For the end encoder error, For the end encoder angle, This is the end reference angle output by an external high-precision angle sensor.

[0021] Further, in step S3, the angle domain frequency analysis includes output angle domain frequency analysis and rotor angle domain frequency analysis, when the end encoder error sequence When a certain error component in the output angle domain is an error component close to the reduction ratio order of the joint module, and is a first-order or multi-order error component in the rotor angle domain, the error component is identified as the rotor synchronous vibration error caused by the motor rotor vibration being transmitted to the hollow shaft and end encoder through bearings or structural coupling.

[0022] Furthermore, in step S3, the rotor synchronous vibration compensation model is based on the motor rotor angle. The first-order harmonic model, the rotor synchronous vibration compensation amount Obtained according to the following relationship:

[0023]

[0024] in, This represents the amplitude of the rotor's first-order synchronization error. This is the first-order synchronization error phase of the rotor.

[0025] Furthermore, in step S3, the rotor synchronous vibration compensation model is based on the motor rotor angle. The multi-order harmonic model, the rotor synchronous vibration compensation amount Obtained according to the following relationship:

[0026]

[0027] in, The rotor synchronization harmonic order, For the first The amplitude of the first rotor synchronization error. For the first Phase of rotor synchronization error.

[0028] Furthermore, in step S4, the speed-current baseline model is based on the motor q-axis current under no-load, no-load, or reference load conditions. With motor speed The correspondence between them is established, and the load-sensitive current characteristic quantity is... Obtained according to the following relationship:

[0029]

[0030] in, To match the motor speed The corresponding velocity-dependent baseline current.

[0031] Furthermore, the velocity-current baseline model is a Stribeck-type velocity-dependent model.

[0032] Furthermore, in step S5, the harmonic prior enhancement Kolmogorov-Arnold compensation network includes an angle periodic feature encoding layer, a harmonic prior compensation branch, a Kolmogorov-Arnold residual compensation branch, and a compensation amount fusion layer.

[0033] Furthermore, the angle periodic feature encoding layer is used to encode the motor rotor angle. and the end encoder angle Converted to sine and cosine features, wherein the sine and cosine features include at least... , , and .

[0034] Furthermore, the angle periodic feature encoding layer is also used to generate order features related to the joint module deceleration ratio N, the order features including and It is used to characterize the error characteristics of the synchronous vibration of the motor rotor in the output angular domain, which are close to the reduction ratio order.

[0035] Furthermore, the harmonic prior compensation branch is based on the motor rotor angle. Generate an angular error compensation amount that is synchronized with the rotation of the motor rotor. The input of the harmonic prior compensation branch includes... , ... and cos( The rotor synchronous vibration compensation amount is obtained through learnable weights. .

[0036] Furthermore, the Kolmogorov-Arnold residual compensation branch is used to establish the current operating condition characteristics and the residual error compensation amount. The nonlinear mapping relationship between them, whose input includes the load-sensitive current characteristic quantity The motor speed The angle period features processed by the aforementioned angle period feature encoding layer output the residual error compensation amount caused by the hollow shaft cantilever deflection, end encoder mounting eccentricity, end encoder tilt, or change in the gap between the end encoder and the reading head. .

[0037] Furthermore, the connecting edges of the Kolmogorov-Arnold residual compensation branch are provided with learnable univariate functions, which are parameterized using spline functions, sine basis functions, Fourier basis functions, or polynomial basis functions.

[0038] Furthermore, the training labels for the harmonic prior-enhanced Kolmogorov-Arnold compensation network are the error sequences of the end encoder. Or, to deduct the rotor synchronous vibration compensation amount. The residual error sequence after The residual error sequence Obtained according to the following relationship:

[0039] = -

[0040] Further, in step S6, the total error compensation amount of the end encoder Obtained according to the following relationship:

[0041] = +

[0042] Furthermore, in step S7, when the feedback value correction method is used, the compensated end angle estimate is... Obtained according to the following relationship:

[0043] = -

[0044] And based on the target angle and the estimated compensated end angle... The deviation between them is used for closed-loop control of the end position.

[0045] Furthermore, in step S7, when the target value correction method is adopted, the corrected end position closed-loop target value is... Obtained according to the following relationship:

[0046]

[0047] And control the angle of the end encoder Tracking the corrected end-position closed-loop target value .

[0048] Furthermore, the external high-precision angle sensor is only used during the calibration phase to establish the end encoder error sequence. The harmonic prior-enhanced Kolmogorov-Arnold compensation network, trained accordingly, does not participate in the closed-loop control of the end position of the joint module during the online control phase.

[0049] Furthermore, the end encoder is fixed to the rear extension section of the hollow shaft, and the hollow shaft is connected to the motor rotor shaft through a bearing. The method is used to compensate for the measurement error of the end encoder caused by insufficient rigidity of the hollow shaft cantilever support, eccentric installation of the end encoder, tilting of the end encoder installation, and vibration of the motor rotor transmitted through the bearing or structural coupling.

[0050] This invention also provides a hollow dual encoder joint module end encoder error identification and compensation system, comprising:

[0051] The data acquisition module is used to acquire the encoder angle at the motor end. End encoder angle q-axis current of motor Motor speed and external reference angle ;

[0052] An error sequence construction module is used to construct the sequence based on the end encoder angle. and the external reference angle Establish the end encoder error sequence ;

[0053] The frequency analysis module is used to analyze the error sequence of the end encoder. Perform angular domain frequency analysis and extract the error components that are synchronized with the rotation of the motor rotor;

[0054] The rotor synchronous vibration compensation module is used to establish a rotor synchronous vibration compensation model based on the error components that are synchronized with the rotation of the motor rotor, and to output the rotor synchronous vibration compensation amount. ;

[0055] The speed baseline correction module is used to adjust the motor q-axis current based on the reference operating condition. and motor speed Establish a speed-current baseline model, and use the current motor q-axis current as the basis. After subtracting the speed-dependent baseline current, the load-sensitive current characteristic is obtained. ;

[0056] Harmonic prior enhancement Kolmogorov-Arnold compensation network module, used to calculate based on the load-sensitive current characteristics. Motor rotor angle End encoder angle and motor speed Output residual error compensation amount ;

[0057] The compensation fusion module is used to integrate the synchronous vibration compensation of the rotor. and residual error compensation amount Combined into the total error compensation amount of the end encoder ;

[0058] The closed-loop control correction module is used to compensate for the total error of the end encoder. Correct the feedback value or target value of the end position closed-loop control.

[0059] The present invention also provides a joint module controller, including a processor, a memory, and a program stored in the memory and executable by the processor. When the program is executed by the processor, it implements a method for identifying and compensating for the end encoder error of a hollow dual encoder joint module.

[0060] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0061] (1) Based on the structural characteristics of the hollow dual encoder joint module, the present invention divides the measurement error of the end encoder into rotor synchronous vibration error and residual error caused by the shaking of the hollow shaft and the end code disk. The angular domain frequency analysis, harmonic compensation model and harmonic prior enhancement Kolmogorov-Arnold compensation network are used to identify and compensate for the error, thereby improving the pertinence and interpretability of the error compensation.

[0062] (2) By analyzing the frequency in the output angle domain and the frequency in the rotor angle domain, this invention can establish a correspondence between the error in the output angle domain that is close to the reduction ratio order and the first or multiple order error in the motor rotor angle domain, thereby identifying the measurement error caused by the vibration of the motor rotor being transmitted to the hollow shaft and the end encoder through the bearing or structural coupling.

[0063] (3) The present invention corrects the speed baseline of the motor q-axis current by using a speed-current baseline model to obtain the load-sensitive current characteristic quantity, so that the compensation network can use the current characteristics that reflect the external load, gravity component and the stress state of the hollow shaft to improve the prediction ability of nonlinear residual error caused by the cantilever deflection of the hollow shaft and the shaking of the end encoder.

[0064] (4) The present invention adopts a harmonic prior enhancement Kolmogorov-Arnold compensation network, which combines the angle periodic characteristics, harmonic prior compensation branch and Kolmogorov-Arnold residual compensation branch. It can express the periodic error related to the angle and fit the complex nonlinear error caused by structural stress, code disk installation deviation and reading gap change.

[0065] (5) The external high-precision angle sensor in this invention is only used for error sequence establishment and network training during the calibration stage. No external high-precision angle sensor is required for control during the online control stage, which makes it easy to deploy in actual joint module controllers.

[0066] (6) The present invention can achieve closed-loop control compensation of end position by correcting feedback value or target value without changing the basic dual encoder control structure of the joint module. It is suitable for use in hollow collaborative robot joints, industrial robot joints, special operation robotic arm joints and other integrated joint modules that need to improve the positioning accuracy of the output end. Attached Figure Description

[0067] Figure 1 This is a flowchart of the overall method of the present invention;

[0068] Figure 2 This is a flowchart of the angular domain frequency analysis and rotor synchronous vibration compensation of the present invention;

[0069] Figure 3 This is a flowchart of the speed-current baseline correction and load-sensitive current feature extraction of the present invention;

[0070] Figure 4 This is a block diagram of the internal structure of the harmonic prior enhancement Kolmogorov-Arnold compensation network of the present invention;

[0071] Figure 5 This is a flowchart of the online closed-loop control correction process of the present invention. Detailed Implementation

[0072] The technical solution of the present invention will be further described below with reference to the accompanying drawings. It should be understood that the described embodiments are only for explaining the present invention and are not intended to limit the scope of protection of the present invention. All equivalent substitutions, equivalent transformations, or conventional improvements made based on the technical concept of the present invention should fall within the scope of protection of the present invention.

[0073] This invention provides a method for identifying and compensating for errors in the end encoder of a hollow dual-encoder joint module. The method is applicable to hollow joint modules with a motor-end encoder and an end encoder, wherein the motor-end encoder outputs the motor rotor angle, and the end encoder outputs the end encoder angle. During the calibration phase, an external high-precision angle sensor is used to output the end reference angle; during the online control phase, the external high-precision angle sensor does not participate in closed-loop control.

[0074] The core idea of ​​this invention is as follows: taking the measurement error of the end encoder relative to the external reference angle as the modeling object, identifying the error component that is synchronized with the rotation of the motor rotor through angular domain frequency analysis, and establishing a rotor synchronous vibration compensation model; at the same time, extracting the load-sensitive current characteristic quantity through speed-current baseline correction, and then using the harmonic prior enhancement Kolmogorov-Arnold compensation network to predict the residual error compensation quantity that is difficult to analyze and model; finally, combining the rotor synchronous vibration compensation quantity and the residual error compensation quantity into the total error compensation quantity of the end encoder, which is used to correct the feedback value or target value of the end position closed-loop control.

[0075] like Figure 1 As shown, the end encoder error identification and compensation method of the present invention includes a calibration stage and an online control stage.

[0076] During the calibration phase, the hollow dual encoder joint module is controlled to operate according to preset angle ranges, preset speed ranges, and preset load conditions, while simultaneously acquiring the motor rotor angle. End encoder angle q-axis current of motor Motor speed and the end reference angle output by an external high-precision angle sensor The preset operating conditions may include forward motion, reverse motion, different speeds, different angular positions, and different load states, in order to improve the adaptability of the error model to actual operating conditions.

[0077] According to the end encoder angle With end reference angle Establish the end encoder error sequence :

[0078] = -

[0079] in, For the end encoder error, For the end encoder angle, This is the end-effector reference angle output from an external high-precision angle sensor. This error sequence is used to characterize the deviation of the end-effector encoder readings from the true end-effector angle.

[0080] like Figure 2 As shown, the error sequence of the end encoder Angle domain frequency analysis is performed. This angle domain frequency analysis includes output angle domain frequency analysis and rotor angle domain frequency analysis.

[0081] Output angle domain frequency analysis based on end encoder angle Alternatively, the joint output angle can be used as the independent variable to calculate the error sequence of the end encoder. Resampling is performed at uniform output angle intervals, followed by Fourier or order analysis to identify periodic error components related to the output angle. Rotor angle domain frequency analysis is then conducted using the motor rotor angle... As the independent variable, the end encoder error sequence Resampling is performed at uniform rotor angle intervals, followed by Fourier analysis or order analysis to identify error components that are synchronized with the motor rotor rotation.

[0082] When a certain error component exhibits an error component close to the Nth order of the joint module reduction ratio in the output angle domain, and a first-order or multi-order error component in the rotor angle domain, this error component is identified as rotor synchronous vibration error. This identification method can establish a correspondence between higher-order errors in the output angle domain and synchronous errors in the motor rotor angle domain, thereby distinguishing rotor synchronous vibration error from other residual errors.

[0083] In one implementation, the rotor synchronous vibration compensation model is based on the motor rotor angle. First-order harmonic model:

[0084]

[0085] in, This is the rotor synchronous vibration compensation amount. This represents the amplitude of the rotor's first-order synchronization error. This is the first-order synchronization error phase of the rotor.

[0086] In another implementation, when there are multiple order components in the rotor synchronous vibration error, the rotor synchronous vibration compensation model is based on the motor rotor angle. Multi-harmonic model:

[0087]

[0088] in, The rotor synchronization harmonic order, For the first The amplitude of the first rotor synchronization error. For the first Phase of rotor synchronization error.

[0089] and It can be obtained through rotor angle domain frequency analysis, or through least squares fitting or other parameter identification methods.

[0090] like Figure 3 As shown, to extract current characteristics related to external load, gravity components, and joint stress states, this invention establishes a velocity-current baseline model. The velocity-current baseline model is based on the motor q-axis current under reference operating conditions. With motor speed The correspondence between them is established. The reference operating condition can be no load, no external load, or a known reference load condition.

[0091] Under baseline operating conditions, the control joint module operates in both forward and reverse directions at different speeds, and the q-axis current of the motor is collected. With motor speed To establish the correspondence between them, a velocity-current baseline model is established. The velocity-current baseline model is a Stribeck-type velocity-dependent model.

[0092] During online operation or calibration data processing, based on the current motor speed Obtain speed-related baseline current And determined by the current q-axis current of the motor. Subtracting the speed-dependent baseline current yields the load-sensitive current characteristic. :

[0093]

[0094] in, For load-sensitive current characteristic quantity, The current q-axis current of the motor To match the current motor speed The corresponding speed-dependent baseline current. This load-sensitive current characteristic can reflect the current characteristics related to external load, gravity component, joint stress state, and changes in end encoder measurement error.

[0095] like Figure 4As shown, this invention employs a harmonic prior-enhanced Kolmogorov-Arnold compensation network to predict the residual error compensation amount. The harmonic prior-enhanced Kolmogorov-Arnold compensation network, which can be abbreviated as HP-KAN compensation network, includes an input layer, an angle periodic feature encoding layer, a feature concatenation layer, a feature normalization layer, a harmonic prior compensation branch, a Kolmogorov-Arnold residual compensation branch, a compensation amount fusion layer, and an output layer.

[0096] The input layer receives load-sensitive current characteristics. Motor speed Motor rotor angle and end encoder angle .because and These are all angle variables, and when directly input into the network, they may jump at the 0 to 2π periodic boundaries. Therefore, they are converted into sine and cosine features through an angle periodic feature encoding layer.

[0097] The angle periodic feature encoding layer generates at least the following angle periodic features: , , and In one implementation, the angle periodic feature encoding layer also generates order features related to the joint module deceleration ratio N: and The order feature is used to characterize the periodic error features related to the deceleration ratio order.

[0098] The feature concatenation layer is used to concatenate the original input features with the angular periodic features to form the network input vector. The feature normalization layer is used to standardize or normalize the concatenated features to reduce the impact of different units and numerical ranges on the stability of network training.

[0099] The harmonic prior compensation branch is used to express the periodic error component synchronized with the motor rotor angle. This branch includes a harmonic feature generation layer, a harmonic linear weighting layer, and a harmonic summation output layer. The harmonic feature generation layer is based on the motor rotor angle. The following harmonic characteristics are generated: , ... and cos( The harmonic linear weighting layer weights the harmonic features using learnable weights, and the harmonic summation output layer outputs the rotor synchronous vibration compensation amount. .

[0100] The Kolmogorov-Arnold residual compensation branch is used to establish a nonlinear mapping relationship between the current operating condition characteristics and the residual error compensation amount. This branch includes a KAN input mapping layer, at least one KAN layer, at least one node aggregation layer, and a residual output layer.

[0101] The KAN input mapping layer is used to receive... , And the angular periodicity features are used to form the input vector of the KAN residual compensation branch. Learnable unary functions are set on the connection edges of the KAN layer. The learnable univariate function can be parameterized using spline functions, sine basis functions, Fourier basis functions, or polynomial basis functions. The node aggregation layer is used to sum and aggregate the univariate function outputs of each connected edge, for example: in, For input node variables, These are the intermediate node variables after aggregation. Through at least one KAN layer and a node aggregation layer, the network can fit the nonlinear relationship between load-sensitive current characteristics, rotational speed characteristics, and angle period characteristics and residual error.

[0102] In one implementation, the Kolmogorov-Arnold residual compensation branch includes a first KAN layer, a first node aggregation layer, a second KAN layer, a second node aggregation layer, and a residual output layer. The first KAN layer performs a univariate function mapping on the input features, and the first node aggregation layer aggregates the output of the first KAN layer. The second KAN layer performs further nonlinear mapping on the aggregated intermediate features, and the second node aggregation layer performs further aggregation. The residual output layer can be a linear output layer or a fully connected output layer, outputting the residual error compensation amount. .

[0103] During the training phase, the training labels for the HP-KAN compensation network can be the end encoder error sequence. It can also be used to deduct the rotor synchronous vibration compensation amount. The residual error sequence after :

[0104] = -

[0105] in, For residual error, For the end encoder error, This is the rotor synchronous vibration compensation amount. (Used) When used as training labels, rotor synchronous vibration errors with clear frequency characteristics are handled by harmonic prior compensation branch or rotor synchronous vibration compensation model, while complex nonlinear residual errors are handled by Kolmogorov-Arnold residual compensation branch.

[0106] During training, the loss function can be constructed using mean squared error (MSE), mean absolute error (MAE), or a weighted combination of both. After training, the HP-KAN compensation network parameters are stored in the joint module controller, and only the motor rotor angle is used during online operation. End encoder angle q-axis current of motor Motor speed It also enables real-time inference using a velocity-current baseline model, eliminating the need for external high-precision angle sensors.

[0107] The compensation fusion layer is used to incorporate rotor synchronous vibration compensation. With residual error compensation amount Combined into the total error compensation amount of the end encoder :

[0108] = +

[0109] in, This represents the total error compensation amount of the end encoder. This is the rotor synchronous vibration compensation amount. This is the residual error compensation amount.

[0110] like Figure 5 As shown, during the online control phase, the controller collects the motor rotor angle in real time. End encoder angle q-axis current of motor and motor speed Calculate the velocity-dependent baseline current based on the velocity-current baseline model. And obtain the load-sensitive current characteristic. The rotor synchronous vibration compensation amount is obtained based on the rotor synchronous vibration compensation model or the harmonic prior compensation branch. The residual error compensation amount is obtained based on the Kolmogorov-Arnold residual compensation branch. The total error compensation of the end encoder is then obtained from the compensation fusion layer. .

[0111] One online compensation method employs a feedback value correction approach. This method is based on the total error compensation amount of the end encoder. For the end encoder angle After correction, the compensated end angle estimate is obtained. :

[0112] = -

[0113] Subsequently, the end-position closed-loop controller uses the target angle and the compensated end-position angle estimate. The deviation between the two is controlled. This method is equivalent to first correcting the end encoder reading, which has measurement errors, to the true end angle estimate, and then using it as feedback to participate in closed-loop control.

[0114] Another online compensation method employs a target value correction approach. This method is based on the total error compensation amount of the end encoder. For the original end target angle Pre-compensation is performed to obtain the corrected closed-loop target value for the end position. :

[0115] = +

[0116] Subsequently, control the angle of the end encoder. Tracking the corrected end-position closed-loop target value This method is equivalent to pre-adding an error compensation amount at the target end, making the actual end-effector angle of the joint approach the original end-effector target angle. .

[0117] The feedback value correction method and the target value correction method can be used interchangeably. The feedback value correction method is suitable for scenarios where the controller can directly modify the position feedback channel; the target value correction method is suitable for scenarios where the control system cannot easily modify the feedback channel but can modify the position command. Both methods can be based on the total error compensation amount of the end encoder. The closed-loop control of the end position is modified.

[0118] This invention also provides a hollow dual-encoder joint module end encoder error identification and compensation system. The system includes a data acquisition module, an error sequence construction module, an angle domain frequency analysis module, a rotor synchronous vibration compensation module, a speed baseline correction module, a harmonic prior enhancement Kolmogorov-Arnold compensation network module, a compensation amount fusion module, and a closed-loop control correction module. Each module is used to complete data acquisition, error sequence construction, rotor synchronous error identification, rotor synchronous vibration compensation, load-sensitive current characteristic extraction, residual error compensation, total error compensation amount fusion, and feedback value or target value correction, respectively.

[0119] This invention also provides a joint module controller. The joint module controller includes a processor, a memory, and a program stored in the memory and executable by the processor. When executed by the processor, the program implements the aforementioned method for identifying and compensating the encoder error at the end of the hollow dual-encoder joint module. The processor can be a microcontroller, digital signal processor, field-programmable gate array, system-on-a-chip, or other computing unit with real-time control capabilities. The memory can be used to store speed-current baseline model parameters, rotor synchronous vibration compensation model parameters, HP-KAN compensation network parameters, and online control programs.

[0120] The hollow dual encoder joint module can use a permanent magnet synchronous motor, a frameless torque motor, or other rotary motor as the drive source; it can use a harmonic reducer, planetary reducer, cycloidal reducer, or other speed reduction and torque amplification transmission mechanism; the motor end encoder and the end encoder can be a magnetic encoder, photoelectric encoder, inductive encoder, or other angle measuring device. The specific hardware configuration described above does not affect the implementation of the method of this invention.

[0121] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art can modify, substitute, or alter the above embodiments without departing from the principles and spirit of the present invention; all equivalent changes made within the technical concept and scope of protection of the claims of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying and compensating for errors in the end encoder of a hollow dual encoder joint module, wherein the hollow dual encoder joint module includes a motor, a reducer, a motor-end encoder, a hollow shaft, an end code disk disposed on the hollow shaft, and an end encoder for reading the end code disk, characterized in that, The method includes the following steps: S1. During the calibration phase, acquire the motor rotor angle output from the encoder at the motor end. The end encoder angle output by the end encoder q-axis current of motor Motor speed and the end reference angle output by an external high-precision angle sensor ; S2, based on the angle of the end encoder With the aforementioned end reference angle Establish the end encoder error sequence ; S3, the error sequence of the end encoder An angular domain frequency analysis was performed to extract the error components that are synchronized with the rotation of the motor rotor, and a rotor synchronous vibration compensation model was established based on the error components. S4. Establish a rotor synchronous vibration compensation model based on the rotor synchronous vibration error, and obtain the rotor synchronous vibration compensation amount. ; S5. Based on the motor q-axis current under the reference operating conditions. With motor speed Establish a speed-current baseline model, and use the current motor q-axis current as the basis. Subtract from the current motor speed The corresponding speed-dependent baseline current is used to obtain the load-sensitive current characteristic quantity. ; S6, the load-sensitive current characteristic quantity The rotor angle of the motor The angle of the end encoder and the motor speed By inputting a harmonic prior-enhanced Kolmogorov-Arnold compensation network, the residual error compensation caused by the hollow shaft cantilever deflection, end encoder installation deviation, and end encoder wobble is obtained. ; S7. The rotor synchronous vibration compensation amount output by the rotor synchronous vibration compensation model. With the residual error compensation amount Combined into the total error compensation amount of the end encoder ; S8. During the online control phase, based on the total error compensation amount of the end encoder... Correct the feedback or target value of the end position closed-loop control to compensate for the end encoder reading error; S9. Based on the motor rotor angle collected online. q-axis current of motor Motor speed ωm, end encoder angle Using the model parameters saved in step S7, calculate the rotor synchronous vibration compensation amount. and residual error compensation amount The two are then combined into the total error compensation amount of the end encoder. ; S10. Based on the total error compensation amount of the end encoder Correct the feedback or target value of the end position closed-loop control to compensate for end encoder reading errors and improve the end positioning accuracy of the joint module.

2. The method for error identification and compensation of the end encoder of the hollow dual encoder joint module according to claim 1, characterized in that, In step S2, the end encoder error sequence Established according to the following relationship: = - ； in, For the end encoder error, For the end encoder angle, This is the end reference angle output by an external high-precision angle sensor.

3. The method for error identification and compensation of the end encoder of the hollow dual encoder joint module according to claim 1, characterized in that, In step S3, the angle domain frequency analysis includes output angle domain frequency analysis and rotor angle domain frequency analysis, when the end encoder error sequence When a certain error component in the output angle domain is an error component close to the reduction ratio order of the joint module, and is a first-order or multi-order error component in the rotor angle domain, the error component is identified as the rotor synchronous vibration error caused by the motor rotor vibration being transmitted to the hollow shaft and end encoder through bearings or structural coupling.

4. The method for error identification and compensation of the end encoder of the hollow dual encoder joint module according to claim 1, characterized in that, In step S3, the rotor synchronous vibration compensation model is based on the motor rotor angle. The first-order harmonic model, the rotor synchronous vibration compensation amount Obtained according to the following relationship: ； in, This represents the amplitude of the rotor's first-order synchronization error. The rotor's first-order synchronization error phase; in step S3, the rotor synchronization vibration compensation model is based on the motor rotor angle. The multi-order harmonic model, the rotor synchronous vibration compensation amount Obtained according to the following relationship: ； in, The rotor synchronization harmonic order, For the first The amplitude of the first rotor synchronization error. For the first Phase of rotor synchronization error.

5. The method for error identification and compensation of the end encoder of the hollow dual encoder joint module according to claim 1, characterized in that, In step S4, the speed-current baseline model is based on the motor q-axis current under no-load, no-load, or reference load conditions. With motor speed The correspondence between them is established, and the load-sensitive current characteristic quantity is... Obtained according to the following relationship: ； in, To match the motor speed The corresponding velocity-dependent baseline current, wherein the velocity-current baseline model is a Stribeck-type velocity-dependent model.

6. The method for error identification and compensation of the end encoder of the hollow dual encoder joint module according to claim 1, characterized in that, In step S5, the harmonic prior enhancement Kolmogorov-Arnold compensation network includes an angle periodic feature encoding layer, a harmonic prior compensation branch, a Kolmogorov-Arnold residual compensation branch, and a compensation amount fusion layer; the angle periodic feature encoding layer is used to input the motor rotor angle and the end encoder angle Converted to sine and cosine features, wherein the sine and cosine features include at least... , , and The angle periodic feature encoding layer is also used to generate order features related to the joint module deceleration ratio N, the order features including... and It is used to characterize the error characteristics of the synchronous vibration of the motor rotor in the output angular domain, which are close to the reduction ratio order.

7. The method for error identification and compensation of the end encoder of the hollow dual encoder joint module according to claim 6, characterized in that, The harmonic prior compensation branch is based on the motor rotor angle. Generate an angular error compensation amount that is synchronized with the rotation of the motor rotor. The input of the harmonic prior compensation branch includes... , ... and cos( The rotor synchronous vibration compensation amount is obtained through learnable weights. The Kolmogorov-Arnold residual compensation branch is used to establish the current operating condition characteristics and the residual error compensation amount. The nonlinear mapping relationship between them, whose input includes the load-sensitive current characteristic quantity The motor speed The angle period features processed by the aforementioned angle period feature encoding layer output the residual error compensation amount caused by the hollow shaft cantilever deflection, end encoder mounting eccentricity, end encoder tilt, or change in the gap between the end encoder and the reading head. The Kolmogorov-Arnold residual compensation branch has a learnable univariate function set on its connecting edge. The learnable univariate function is parameterized using spline functions, sine basis functions, Fourier basis functions, or polynomial basis functions.

8. The method for error identification and compensation of the end encoder of the hollow dual encoder joint module according to claim 1, characterized in that, The training labels for the harmonic prior-enhanced Kolmogorov-Arnold compensation network are the end encoder error sequences. Or, to deduct the rotor synchronous vibration compensation amount. The residual error sequence after The residual error sequence Obtained according to the following relationship: = - 。 9. The method for error identification and compensation of the end encoder of the hollow dual encoder joint module according to claim 1, characterized in that, In step S6, the total error compensation amount of the end encoder Obtained according to the following relationship: = + ； In step S7, when the feedback value correction method is used, the compensated end angle estimate is... Obtained according to the following relationship: ＝ - ； And based on the target angle and the estimated compensated end angle value The deviation between them is used for end-position closed-loop control; in step S7, when the target value correction method is used, the corrected end-position closed-loop target value is... Obtained according to the following relationship: ； And control the angle of the end encoder Tracking the corrected end-position closed-loop target value .

10. The method for error identification and compensation of the end encoder of the hollow dual encoder joint module according to claim 1, characterized in that, The external high-precision angle sensor is only used during the calibration phase to establish the end encoder error sequence. The harmonic prior-enhanced Kolmogorov-Arnold compensation network, trained accordingly, does not participate in the closed-loop control of the end position of the joint module during the online control phase.

Images