A robot integrated joint dual encoder calibration method and system

Abstract

The present invention discloses a calibration method and system for a robot integrated joint dual encoder, the method comprising: rotating the joint at a preset speed, collecting angle data, calculating the first and second angle difference curves based on the joint reduction ratio, analyzing the average angle difference curve, and when the deviation ratio meets the non-disassembly calibration condition, using a pre-trained model to perform intelligent calibration on the dual encoder, otherwise it needs to be disassembled and recalibrated. The present invention can accurately identify and correct the deviation of the encoder without disassembling the joint by implementing key technical means such as data collection at a preset speed, angle difference calculation based on the joint reduction ratio, threshold analysis and intelligent decision-making, thereby greatly improving the accuracy and efficiency of calibration.

Description

Technical Field

[0001] The present invention relates to the technical field of joint module calibration, and in particular to a calibration method and system for a robot integrated joint dual encoder. Background Art

[0002] In the field of industrial automation, precise control of robot integrated joints is one of the key factors to ensure production efficiency and product quality. However, traditional calibration methods often face many challenges, especially in the calibration of encoders for robot joints. Traditional methods usually rely on manual inspection and frequent disassembly and assembly operations, which is not only time-consuming and laborious, but may also cause additional damage to the robot joints, affecting their long-term performance and stability. In addition, the subjectivity and inconsistency of manual calibration also limit the accuracy and efficiency of calibration, increase maintenance costs, and reduce production efficiency.

[0003] The traditional calibration process usually includes the following steps: first, the robot joint needs to be disassembled to facilitate direct contact and adjustment of the encoder; then, the joint angle is measured using standard tools or equipment and compared with the encoder reading to identify the deviation; finally, the encoder parameters are adjusted according to the deviation and the joint is reassembled. This series of operations is not only cumbersome, but also during the frequent disassembly and assembly process, it may cause the internal components of the joint to loosen or be damaged, thus affecting the overall performance of the robot. Summary of the invention

[0004] In order to solve at least one of the technical problems raised above, the present invention provides a calibration of a robot's integrated joint dual encoder. By implementing key technical means such as data acquisition at a preset speed, angle difference calculation based on the joint reduction ratio, threshold analysis and intelligent decision-making, the encoder deviation can be accurately identified and corrected without disassembling the joint, thereby greatly improving the accuracy and efficiency of the calibration.

[0005] On the one hand, the present invention provides a calibration method for a robot integrated joint dual encoder, the method comprising:

[0006] The joint is rotated at a preset speed to collect a standard joint angle curve, a first angle curve output by a first encoder, and a second angle curve output by a second encoder, wherein the first encoder is located at the joint output side and the second encoder is located at the motor side;

[0007] Based on a preset joint reduction ratio, respectively calculating the difference between the standard joint angle curve and the first angle curve and the second angle curve to obtain a first angle difference curve and a second angle difference curve, and obtaining an average angle difference curve according to the first angle difference curve and the second angle difference curve;

[0008] Calculate the proportion of the average angle difference curve that is greater than a preset first threshold. When the proportion is in a preset non-disassembly calibration threshold interval, perform non-disassembly calibration on the first encoder and the second encoder according to the first angle difference curve and the second angle difference curve and use a pre-trained encoder calibration model. When the proportion is greater than the maximum value of the non-disassembly calibration threshold interval, reinstall and calibrate the first encoder and the second encoder.

[0009] Preferably, the method of calculating the difference between the standard joint angle curve and the first angle curve and the second angle curve based on the preset joint reduction ratio to obtain the first angle difference curve and the second angle difference curve comprises:

[0010] The standard joint angle curve is subjected to point-by-point subtraction processing with the first angle curve to obtain a first angle difference curve, the second angle curve is multiplied point by a preset reduction ratio, and the standard joint angle curve and the multiplication result are subjected to point-by-point subtraction processing to obtain a second angle difference curve.

[0011] Preferably, before performing non-disassembly calibration on the first encoder and the second encoder according to the first angle difference curve and the second angle difference curve and using a pre-trained encoder calibration model, the method further includes training the encoder calibration model, specifically including:

[0012] Collect the operating parameters of robot joints at various working speeds;

[0013] A loss function is designed based on the operating parameters and the encoder calibration model based on the neural network model is trained through an optimization function; wherein the operating parameters include at least a standard angle, a joint rotation speed, a first encoder angle, and a second encoder angle.

[0014] Preferably, the neural network model adopts an LSTM network model, and the LSTM network model is an STL-LSTM network model, and the STL-LSTM network model is composed of two layers of STL-LSTM networks, and each layer of STL-LSTM network includes two LSTM networks and one fully connected layer;

[0015] The extracted features are passed into the two LSTM networks in the first layer of the STL-LSTM network, the two LSTM networks are trained respectively, and the outputs of the two LSTM networks are respectively input into the corresponding fully connected layers for fusion to obtain the first fused data;

[0016] The first fusion data is respectively transmitted to two LSTM networks in the second layer STL-LSTM network, the two LSTM networks are trained, and the outputs of the two LSTM networks are respectively input to the corresponding fully connected layers for fusion, and the encoder calibration model is obtained by continuous iteration until the model converges;

[0017] A back-propagation algorithm is used to obtain the gradient corresponding to the optimization function, and a gradient descent method is used to optimize the network parameters of the encoder calibration model.

[0018] Preferably, the standard angle is calculated based on a time period calculation formula, and the time period calculation formula is expressed as:

[0019] ,

[0020] In the formula, Indicates the standard angle, Indicates the measurement time, Indicates the time required for a joint to rotate once.

[0021] Preferably, the non-disassembly calibration of the first encoder and the second encoder according to the first angle difference curve and the second angle difference curve and using a pre-trained encoder calibration model comprises:

[0022] The first angle difference curve and the second angle difference curve are taken as input, and correction coefficients are output through the encoder calibration model. The correction coefficients include a first correction coefficient and a second correction coefficient. The reading of the first encoder is adjusted based on the first correction coefficient, and the reading of the second encoder is adjusted based on the second correction coefficient.

[0023] Preferably, after performing non-disassembly calibration on the first encoder and the second encoder according to the first angle difference curve and the second angle difference curve and using a pre-trained encoder calibration model, the method further includes evaluating the calibration effect, specifically including:

[0024] Performing multiple repeated joint rotation tests using the calibrated first encoder and the second encoder, collecting new first angle curves, second angle curves, and standard joint angle curves, and calculating new first angle difference curves and second angle difference curves;

[0025] Compare the root mean square errors of the new and old angle difference curves to determine whether the root mean square error of the new angle difference curve is lower than the preset calibration effect threshold. If so, the calibration is determined to be successful; if not, re-execute the calibration process.

[0026] In a second aspect, the present invention provides a robot integrated joint dual encoder calibration system, the system comprising:

[0027] A data acquisition module, used to rotate the joint at a preset speed, collect a standard joint angle curve, a first angle curve output by a first encoder, and a second angle curve output by a second encoder, wherein the first encoder is located at the joint output side, and the second encoder is located at the motor side;

[0028] A data comparison module, for calculating the difference between the standard joint angle curve and the first angle curve and the second angle curve respectively based on a preset joint reduction ratio, obtaining a first angle difference curve and a second angle difference curve, and obtaining an average angle difference curve according to the first angle difference curve and the second angle difference curve;

[0029] The encoder calibration module is used to calculate the proportion of the average angle difference curve that is greater than a preset first threshold value. When the proportion is in a preset non-disassembly calibration threshold interval, the first encoder and the second encoder are non-disassembly calibrated according to the first angle difference curve and the second angle difference curve and using a pre-trained encoder calibration model. When the proportion is greater than the maximum value of the non-disassembly calibration threshold interval, the first encoder and the second encoder are reinstalled and calibrated.

[0030] Preferably, the system further comprises a model training module, wherein the model training module comprises:

[0031] A data acquisition unit is used to collect operating parameters of robot joints at various working speeds;

[0032] A model training unit is used to design a loss function based on the operating parameters and train the encoder calibration model based on the neural network model through an optimization function; wherein the operating parameters include at least a standard angle, a joint rotation speed, a first encoder angle, and a second encoder angle.

[0033] Preferably, the system further comprises a calibration assessment module, wherein:

[0034] A difference curve calculation unit is used to perform multiple repeated joint rotation tests using the calibrated first encoder and the second encoder, collect new first angle curves, second angle curves and standard joint angle curves, and calculate the new average angle difference curve;

[0035] The difference curve comparison unit is used to compare the root mean square error of the new average angle difference curve to determine whether the root mean square error is lower than a preset calibration effect threshold. If so, the calibration is determined to be successful; if not, the calibration process is re-executed.

[0036] The beneficial effects of the present invention are:

[0037] (1) The traditional calibration method relies on manual inspection, while the present invention avoids the frequent disassembly and assembly operations in the traditional calibration method, effectively reducing the additional damage risk and high maintenance cost caused by it. By implementing key technical means such as data collection at a preset speed, angle difference calculation based on joint reduction ratio, threshold analysis and intelligent decision-making, it is possible to accurately identify and correct the deviation of the encoder without disassembling the joint, greatly improving the accuracy and efficiency of calibration;

[0038] (2) The neural network-based intelligent calibration model of the present invention can more accurately identify and correct the deviation of the encoder, significantly improving the positioning accuracy of the robot joints and the overall reliability of the system. Furthermore, by training the model under a variety of working conditions, the calibration model has good generalization ability and can automatically adapt to different working environments and speeds, thereby improving the operating stability of the robot joints.

[0039] (3) The present invention can more accurately grasp the actual position of the joint through the standard angle determination method based on the time period calculation formula, thereby improving the calibration accuracy of the encoder and further improving the positioning accuracy of the robot joint. Furthermore, regular calibration based on the time period calculation formula can timely detect and correct the deviation of the encoder, prevent potential operational instability, and enhance the overall stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the background technology, the drawings required for use in the embodiments of the present invention or the background technology will be described below.

[0041] The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and, together with the specification, are used to illustrate the technical solutions disclosed in the present invention.

[0042] Figure 1 A schematic flow chart of a calibration method for a robot integrated joint dual encoder provided by an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of a process for training an encoder calibration model provided by an embodiment of the present invention;

[0044] Figure 3 A schematic diagram of the training data processing flow of the LSTM network model provided in an embodiment of the present invention;

[0045] Figure 4 A schematic diagram of the structure of a robot integrated joint dual encoder calibration system provided by an embodiment of the present invention;

[0046] Figure 5A schematic diagram of the structure of a calibration evaluation module provided in an embodiment of the present invention;

[0047] Figure 6 A schematic diagram of the structure of a model training module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or devices.

[0050] The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can represent including any one or more elements selected from the set consisting of A, B, and C.

[0051] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0052] In addition, in order to better illustrate the present invention, numerous specific details are provided in the following specific embodiments. It should be understood by those skilled in the art that the present invention can be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present invention.

[0053] At present, the traditional joint encoder calibration method relies on manual inspection and frequent disassembly and assembly, which is time-consuming and labor-intensive and can easily damage the joints. This limits the calibration accuracy and efficiency, increases maintenance costs, and affects production efficiency. The disassembly and assembly operations may cause damage to the internal components of the joints, threatening the overall performance of the robot.

[0054] Embodiment 1, on the one hand, the present invention provides a calibration method for a robot integrated joint dual encoder, referring to Figure 1 , methods include:

[0055] S100, rotating the joint at a preset speed, collecting a standard joint angle curve, a first angle curve output by a first encoder, and a second angle curve output by a second encoder, wherein the first encoder is located at the joint output side, and the second encoder is located at the motor side;

[0056] S200, based on a preset joint reduction ratio, respectively calculating the difference between the standard joint angle curve and the first angle curve and the second angle curve to obtain a first angle difference curve and a second angle difference curve, and obtaining an average angle difference curve according to the first angle difference curve and the second angle difference curve;

[0057] S300, calculate the proportion of the average angle difference curve that is greater than a preset first threshold value. When the proportion is in a preset non-disassembly calibration threshold interval, perform non-disassembly calibration on the first encoder and the second encoder according to the first angle difference curve and the second angle difference curve and use a pre-trained encoder calibration model. When the proportion is greater than the maximum value of the non-disassembly calibration threshold interval, reinstall and calibrate the first encoder and the second encoder.

[0058] In this embodiment, a high-precision angle sensor (as the source of the standard joint angle curve), a first encoder (located on the joint output side), and a second encoder (located on the motor side) are used to place the robot arm in the initial position to ensure that the joint is in the zero position state. At the same time, the initial readings of the first encoder and the second encoder are recorded, and a driver is used to drive a joint of the robot arm to rotate at a preset speed (e.g., 10° / s). At the same time, the high-precision angle sensor, the first encoder, and the second encoder are started for data collection. During the entire rotation cycle, the data of the standard joint angle curve, the first angle curve, and the second angle curve are continuously collected. These data should include the angle change of the joint, the timestamp, and any environmental parameters that may affect the accuracy of the measurement. Using the reduction ratio of the joint (e.g., 100:1), the difference between the standard joint angle curve and the first angle curve and the second angle curve is calculated to generate the first angle difference curve and the second angle difference curve. Subsequently, the two are averaged to obtain the average angle difference curve.

[0059] The proportion of points in the average angle difference curve that are greater than the preset first threshold (for example, ±0.5°) is analyzed. If the proportion falls within the preset non-disassembly calibration threshold range (for example, 0%-15%), the non-disassembly calibration process is entered. If the proportion exceeds the upper limit of the non-disassembly calibration threshold range (15%), an alarm is triggered to prompt the operator to perform disassembly calibration, that is, reinstall the encoder and execute the calibration procedure.

[0060] By calling a pre-trained encoder calibration model, the parameters of the first encoder and the second encoder are automatically adjusted to minimize the deviation in the average angle difference curve, wherein the encoder calibration model is constructed based on a neural network model.

[0061] Traditional calibration methods rely on manual inspection, while this embodiment avoids the frequent disassembly and assembly operations in traditional calibration methods, effectively reducing the additional risk of damage and high maintenance costs caused thereby. By implementing key technical means such as data collection at a preset speed, angle difference calculation based on joint reduction ratio, threshold analysis and intelligent decision-making, it is possible to accurately identify and correct the encoder deviation without disassembling the joint, thereby greatly improving the accuracy and efficiency of the calibration.

[0062] Preferably, based on a preset joint reduction ratio, the difference between the standard joint angle curve and the first angle curve and the second angle curve is calculated respectively to obtain the first angle difference curve and the second angle difference curve, including:

[0063] The standard joint angle curve is subjected to point-by-point difference processing with the first angle curve to obtain a first angle difference curve, the second angle curve is multiplied point by a preset reduction ratio, and the standard joint angle curve is subjected to point-by-point difference processing with the multiplication result to obtain a second angle difference curve.

[0064] In this embodiment, the standard joint angle curve is subjected to point-by-point difference processing with the first angle curve, that is, for each moment t, the difference between the standard angle S(t) and the first encoder angle E1(t) is calculated to obtain the first angle difference curve ΔE1(t)=S(t)-E1(t). Taking into account the influence of the joint reduction ratio, the second angle curve E2(t) is first multiplied point by point by the preset reduction ratio R (assuming it is 1:20), and then subjected to point-by-point difference processing with the standard joint angle curve, that is, ΔE2(t)=S(t)-R*E2(t), to obtain the second angle difference curve. Based on the first angle difference curve ΔE1(t) and the second angle difference curve ΔE2(t), the average values ​​of the two at all sampling points are calculated to form the average angle difference curve ΔE_avg(t).

[0065] In this embodiment, a sophisticated angle difference calculation method combined with the influence of the joint reduction ratio can more accurately identify and quantify the encoder deviation, thereby achieving more precise calibration.

[0066] Preferably, before performing non-disassembly calibration on the first encoder and the second encoder using a pre-trained encoder calibration model according to the first angle difference curve and the second angle difference curve, the method further includes training the encoder calibration model, referring to Figure 2 , including:

[0067] S10, collecting the operating parameters of the robot joints at various working speeds;

[0068] S20, designing a loss function based on operating parameters and training an encoder calibration model based on a neural network model through an optimization function; wherein the operating parameters include at least a standard angle, a joint rotation speed, a first encoder angle, and a second encoder angle.

[0069] In this embodiment, at different working speeds of the robot joint (for example, low speed 10rpm, medium speed 50rpm, high speed 100rpm), a series of operating parameters including but not limited to standard angle, joint rotation speed, first encoder angle and second encoder angle are collected using high-precision sensors. The collected data is preprocessed, including data cleaning, outlier detection and removal, and data normalization, to ensure the quality and consistency of the training data.

[0070] Based on the first angle difference curve ΔE1 and the second angle difference curve ΔE2, a loss function is defined that comprehensively considers the encoder accuracy and the actual motion state of the joint. For example, the loss function can include two parts: one is the sum of squares of the difference between the encoder angle and the standard angle, and the other is the sum of squares of the difference between the joint rotation speed and the target speed. Such a loss function can fully reflect the effect of encoder calibration and the dynamic response performance of the joint.

[0071] A neural network model based on deep learning is constructed. The input layer receives multi-dimensional features such as standard angle, joint rotation speed, first encoder angle and second encoder angle, and the output layer predicts the angle difference after encoder calibration. The model is trained using the preprocessed data set, and the loss function is optimized through the back propagation algorithm and gradient descent method, and the network weights are continuously adjusted until the model converges.

[0072] Divide the data set into training set, validation set and test set to ensure the generalization ability of the model and prevent overfitting. Evaluate the performance of the model on the test set, including but not limited to accuracy, recall rate, F1 score and other indicators. Analyze the deviation between the model prediction results and the actual data, fine-tune the model according to the evaluation results, and optimize hyperparameters such as learning rate and number of hidden layers to further improve the prediction accuracy of the model.

[0073] The trained encoder calibration model is applied to the actual robot joint calibration process. When the first angle difference curve ΔE1 and the second angle difference curve ΔE2 are detected to exceed the preset threshold, the model is used to predict the calibration parameters to achieve automatic calibration under non-disassembly conditions. Furthermore, field operation data is regularly collected as a supplementary data source for model iterative training to continuously optimize model performance.

[0074] In this embodiment, the intelligent calibration model based on neural network can more accurately identify and correct the deviation of the encoder, significantly improving the positioning accuracy of the robot joints and the overall reliability of the system. Furthermore, by training the model under a variety of working conditions, the calibration model has good generalization ability, can automatically adapt to different working environments and speeds, and improve the operating stability of the robot joints.

[0075] Preferably, see Figure 3 The neural network model adopts the LSTM network model, which is the STL-LSTM network model. The STL-LSTM network model consists of two layers of STL-LSTM networks, and each layer of STL-LSTM network contains two LSTM networks and one fully connected layer.

[0076] S21, passing the extracted features into the two LSTM networks in the first layer of the STL-LSTM network, training the two LSTM networks respectively, and inputting the outputs of the two LSTM networks into the corresponding fully connected layers for fusion, thereby obtaining the first fused data;

[0077] S22, respectively passing the first fused data into the two LSTM networks in the second-layer STL-LSTM network, training the two LSTM networks, respectively inputting the outputs of the two LSTM networks into the corresponding fully connected layers for fusion, and continuously iterating until the model converges to obtain the encoder calibration model;

[0078] S23, using a back-propagation algorithm to obtain a gradient corresponding to the optimization function, and using a gradient descent method to optimize the network parameters of the encoder calibration model.

[0079] In this embodiment, a two-layer STL-LSTM network structure is adopted, and each layer includes two LSTM networks and a fully connected layer. This structural design can make full use of the characteristics of time series data, capture long-term and short-term dependencies, and improve the prediction accuracy of the model. The preprocessed feature data is input into two LSTM networks respectively, and the two networks are trained independently to learn different aspects of the data respectively. Then, the outputs of the two LSTM networks are sent to their respective fully connected layers for fusion to obtain the first fused data. The first fused data is input into the two LSTM networks of the second layer again to further refine the features and learn. Similarly, the outputs of the two LSTM networks are respectively entered into the fully connected layers for fusion, and the encoder calibration model is finally obtained by continuous iteration until the model converges. The back propagation algorithm is used to calculate the gradient of the loss function with respect to the model parameters, which is the core step of model training. Based on the calculated gradient, the gradient descent method is used to update the model parameters, and the network structure is gradually optimized to minimize the model loss until the model converges.

[0080] Evaluate the performance of the model on an independent test data set, including prediction accuracy, generalization ability, etc. Apply the trained model to the actual encoder calibration process, monitor the encoder data in real time, automatically identify deviations and make corrections, and improve the operating accuracy of the robot joints.

[0081] In this embodiment, by constructing an encoder intelligent calibration model based on the STL-LSTM network, automatic identification and correction of the robot joint encoder deviation is achieved, which has a significant promoting effect on improving the efficiency and reliability of the automated production line.

[0082] Preferably, the standard angle is calculated based on a time period calculation formula, which is expressed as:

[0083] ,

[0084] In the formula, Indicates the standard angle, Indicates the measurement time, Indicates the time required for a joint to rotate once.

[0085] In this embodiment, through the standard angle determination method based on the time period calculation formula, the actual position of the joint can be grasped more accurately, thereby improving the calibration accuracy of the encoder, and then improving the positioning accuracy of the robot joint. Furthermore, regular calibration based on the time period calculation formula can timely detect and correct encoder deviations, prevent potential operational instability, and enhance the overall stability of the system.

[0086] Preferably, the first encoder and the second encoder are non-disassembly calibrated according to the first angle difference curve and the second angle difference curve and using a pre-trained encoder calibration model, including:

[0087] The first angle difference curve and the second angle difference curve are used as input, and correction coefficients are output through the encoder calibration model. The correction coefficients include a first correction coefficient and a second correction coefficient. The reading of the first encoder is adjusted based on the first correction coefficient, and the reading of the second encoder is adjusted based on the second correction coefficient.

[0088] In this embodiment, the first angle difference curve and the second angle difference curve are used as input and passed to the pre-trained encoder calibration model. The encoder calibration model outputs correction coefficients by analyzing the input difference curve, including the first correction coefficient and the second correction coefficient. The first correction coefficient and the second correction coefficient reflect the degree of deviation of the first encoder and the second encoder from the standard angle. The reading of the first encoder is adjusted based on the first correction coefficient. The specific method is to add (or subtract) the first correction coefficient to the subsequent reading to compensate for the previously detected deviation. Similarly, the reading of the second encoder is adjusted based on the second correction coefficient to ensure that its output is closer to the actual standard joint angle.

[0089] After completing the non-disassembly calibration, collect the angle curve again, compare the angle difference curves before and after calibration, and verify the calibration effect. If the new angle difference curve shows that the deviation is significantly reduced, it means that the non-disassembly calibration is successful and the positioning accuracy of the robot joint is improved.

[0090] Preferably, after non-disassembly calibration is performed on the first encoder and the second encoder according to the first angle difference curve and the second angle difference curve using a pre-trained encoder calibration model, the calibration effect is evaluated, specifically including:

[0091] Performing multiple repeated joint rotation tests using the calibrated first encoder and second encoder, collecting new first angle curves, second angle curves and standard joint angle curves, and calculating new first angle difference curves and second angle difference curves;

[0092] Compare the root mean square errors of the new and old angle difference curves to determine whether the root mean square error of the new angle difference curve is lower than the preset calibration effect threshold. If so, the calibration is determined to be successful; if not, re-execute the calibration process.

[0093] In this embodiment, after completing the non-disassembly calibration, the calibrated first encoder and second encoder are immediately used to perform a preliminary joint rotation test, and a new first angle curve, a second angle curve and a standard joint angle curve are collected. The new first angle difference curve and the second angle difference curve are calculated as the basis for preliminary evaluation. In order to ensure the reliability of the calibration effect, multiple repeated joint rotation tests are performed, and the angle curve data needs to be re-collected after each test to calculate a new angle difference curve. The root mean square error (RMSE) of the new and old angle difference curves is compared to evaluate the deviation change between the encoder reading and the standard joint angle before and after calibration. A preset calibration effect threshold is set. If the root mean square error of the new angle difference curve is lower than this threshold, the calibration is determined to be successful; conversely, if the root mean square error is higher than the threshold, it indicates that the calibration effect is not as expected and the calibration process needs to be re-executed.

[0094] If the calibration effect does not meet expectations, record the specific reasons for the calibration failure, such as the deviation type and size, as a reference for subsequent model training and optimization. Adjust the parameters of the encoder calibration model based on the feedback information, or add more training data to improve the accuracy and robustness of the model. Repeat the calibration process until the calibration effect meets the preset threshold to ensure the positioning accuracy and operation stability of the robot joints.

[0095] In this embodiment, the reliability of the calibration effect is ensured through repeated testing and strict effect evaluation, and positioning errors caused by insufficient or excessive calibration are avoided. Furthermore, based on the feedback loop of the calibration effect, the encoder calibration model is continuously optimized, which improves the adaptability and prediction accuracy of the model and further improves the calibration effect.

[0096] Embodiment 2, the present invention provides a robot integrated joint dual encoder calibration system, the system comprising:

[0097] The data acquisition module 100 is used to rotate the joint at a preset speed, collect a standard joint angle curve, a first angle curve output by a first encoder, and a second angle curve output by a second encoder, wherein the first encoder is located at the joint output side and the second encoder is located at the motor side;

[0098] The data comparison module 200 is used to calculate the difference between the standard joint angle curve and the first angle curve and the second angle curve based on a preset joint reduction ratio, obtain the first angle difference curve and the second angle difference curve, and obtain the average angle difference curve according to the first angle difference curve and the second angle difference curve;

[0099] The encoder calibration module 300 is used to calculate the proportion of the average angle difference curve that is greater than a preset first threshold. When the proportion is in a preset non-disassembly calibration threshold interval, the first encoder and the second encoder are non-disassembly calibrated according to the first angle difference curve and the second angle difference curve and using a pre-trained encoder calibration model. When the proportion is greater than the maximum value of the non-disassembly calibration threshold interval, the first encoder and the second encoder are reinstalled and calibrated.

[0100] Preferably, see Figure 5 The system further comprises a calibration assessment module 400, wherein the calibration assessment module 400:

[0101] The difference curve calculation unit 410 is used to perform multiple repeated joint rotation tests using the calibrated first encoder and second encoder, collect new first angle curves, second angle curves and standard joint angle curves, and calculate new average angle difference curves;

[0102] The difference curve comparison unit 420 is used to compare the root mean square error of the new average angle difference curve to determine whether the root mean square error is lower than a preset calibration effect threshold. If so, the calibration is determined to be successful; if not, the calibration process is re-executed.

[0103] Preferably, see Figure 6 The system further includes a model training module 500, and the model training module 500 includes:

[0104] The data acquisition unit 510 is used to collect the operating parameters of the robot joints at various working speeds;

[0105] The model training unit 520 is used to design a loss function based on operating parameters and train an encoder calibration model based on a neural network model through an optimization function; wherein the operating parameters include at least a standard angle, a joint rotation speed, a first encoder angle, and a second encoder angle.

[0106] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features applied herein.

Claims

1. A calibration method for a robot integrated joint dual encoder, characterized in that: The method comprises: The joint is rotated at a preset speed to collect a standard joint angle curve, a first angle curve output by a first encoder, and a second angle curve output by a second encoder, wherein the first encoder is located at the joint output side and the second encoder is located at the motor side; Based on a preset joint reduction ratio, respectively calculating the difference between the standard joint angle curve and the first angle curve and the second angle curve to obtain a first angle difference curve and a second angle difference curve, and obtaining an average angle difference curve according to the first angle difference curve and the second angle difference curve; Calculate the proportion of the average angle difference curve that is greater than a preset first threshold value, and when the proportion is in a preset non-disassembly calibration threshold interval, perform non-disassembly calibration on the first encoder and the second encoder according to the first angle difference curve and the second angle difference curve and using a pre-trained encoder calibration model, and when the proportion is greater than a maximum value of the non-disassembly calibration threshold interval, reinstall and calibrate the first encoder and the second encoder; Before performing non-disassembly calibration on the first encoder and the second encoder according to the first angle difference curve and the second angle difference curve and using a pre-trained encoder calibration model, the method further includes training the encoder calibration model, specifically including: Collect the operating parameters of robot joints at various working speeds; Design a loss function based on the operating parameters and train the encoder calibration model based on the neural network model through an optimization function; wherein the operating parameters at least include a standard angle, a joint rotation speed, a first encoder angle, and a second encoder angle; The neural network model adopts an LSTM network model, and the LSTM network model is an STL-LSTM network model. The STL-LSTM network model is composed of two layers of STL-LSTM networks, and each layer of STL-LSTM network includes two LSTM networks and one fully connected layer; The extracted features are passed into the two LSTM networks in the first layer of the STL-LSTM network, the two LSTM networks are trained respectively, and the outputs of the two LSTM networks are respectively input into the corresponding fully connected layers for fusion to obtain the first fused data; The first fusion data is respectively transmitted to two LSTM networks in the second layer STL-LSTM network, the two LSTM networks are trained, and the outputs of the two LSTM networks are respectively input to the corresponding fully connected layers for fusion, and the encoder calibration model is obtained by continuous iteration until the model converges; A back propagation algorithm is used to obtain a gradient corresponding to the optimization function, and a gradient descent method is used to optimize the network parameters of the encoder calibration model; The standard angle is calculated based on a time period calculation formula, which is expressed as: ， In the formula, j Indicates the standard angle, t Indicates the measurement time, T Indicates the time required for a joint to rotate once; The non-disassembly calibration of the first encoder and the second encoder using a pre-trained encoder calibration model according to the first angle difference curve and the second angle difference curve includes: The first angle difference curve and the second angle difference curve are taken as input, and correction coefficients are output through the encoder calibration model. The correction coefficients include a first correction coefficient and a second correction coefficient. The reading of the first encoder is adjusted based on the first correction coefficient, and the reading of the second encoder is adjusted based on the second correction coefficient.

2. The calibration method of a robot integrated joint dual encoder according to claim 1, characterized in that: The method of calculating the difference between the standard joint angle curve and the first angle curve and the second angle curve based on the preset joint reduction ratio to obtain the first angle difference curve and the second angle difference curve comprises: The standard joint angle curve is subjected to point-by-point subtraction processing with the first angle curve to obtain a first angle difference curve, the second angle curve is multiplied point by a preset reduction ratio, and the standard joint angle curve and the multiplication result are subjected to point-by-point subtraction processing to obtain a second angle difference curve.

3. The calibration method of a robot integrated joint dual encoder according to claim 1, characterized in that: After the first encoder and the second encoder are non-disassembly calibrated according to the first angle difference curve and the second angle difference curve and using a pre-trained encoder calibration model, the calibration effect is evaluated, specifically including: Used to perform multiple repeated joint rotation tests using the calibrated first encoder and the second encoder, collect new first angle curves, second angle curves and standard joint angle curves, and calculate new average angle difference curves; Compare the root mean square error of the new average angle difference curve to determine whether the root mean square error is lower than a preset calibration effect threshold. If so, the calibration is determined to be successful; if not, re-execute the calibration process.

4. A robot integrated joint dual encoder calibration system, using a robot integrated joint dual encoder calibration method as described in any one of claims 1 to 3, characterized in that: The system comprises: A data acquisition module, used to rotate the joint at a preset speed, collect a standard joint angle curve, a first angle curve output by a first encoder, and a second angle curve output by a second encoder, wherein the first encoder is located at the joint output side, and the second encoder is located at the motor side; A data comparison module, for calculating the difference between the standard joint angle curve and the first angle curve and the second angle curve respectively based on a preset joint reduction ratio, obtaining a first angle difference curve and a second angle difference curve, and obtaining an average angle difference curve according to the first angle difference curve and the second angle difference curve; The encoder calibration module is used to calculate the proportion of the average angle difference curve that is greater than a preset first threshold value. When the proportion is in a preset non-disassembly calibration threshold interval, the first encoder and the second encoder are non-disassembly calibrated according to the first angle difference curve and the second angle difference curve and using a pre-trained encoder calibration model. When the proportion is greater than the maximum value of the non-disassembly calibration threshold interval, the first encoder and the second encoder are reinstalled and calibrated.

5. The robot integrated joint dual encoder calibration system according to claim 4, characterized in that: The system also includes a model training module, which includes: A data acquisition unit is used to collect operating parameters of robot joints at various working speeds; A model training unit is used to design a loss function based on the operating parameters and train the encoder calibration model based on the neural network model through an optimization function; wherein the operating parameters include at least a standard angle, a joint rotation speed, a first encoder angle, and a second encoder angle.

6. The robot integrated joint dual encoder calibration system according to claim 4, characterized in that: The system further comprises a calibration assessment module, wherein: A difference curve calculation unit is used to perform multiple repeated joint rotation tests using the calibrated first encoder and the second encoder, collect new first angle curves, second angle curves and standard joint angle curves, and calculate the new average angle difference curve; The difference curve comparison unit is used to compare the root mean square error of the new average angle difference curve to determine whether the root mean square error is lower than a preset calibration effect threshold. If so, the calibration is determined to be successful; if not, the calibration process is re-executed.

Images