Industrial robot positioning precision compensation method based on double model mapping

Abstract

The application relates to the technical field of industrial robot motion precision compensation and solves the problems of existing original controllers and external compensation controllers. The application uses a double-PSD laser calibration device to identify the kinematic parameters of an industrial robot in a full workspace, establishes a correction model containing parameter errors and a nominal model without errors, and, for a given target pose, solves the nominal joint angle through inverse kinematics of the nominal model, calculates the deviation between the actual reachable pose of the correction model under the nominal joint angle and the target pose, and maps the deviation into a joint space compensation amount; the compensation amount is combined with the nominal joint angle in an incremental superposition manner to synthesize a correction joint angle instruction, which is transmitted to a robot servo driver through an external compensation controller. The absolute positioning precision after compensation is better than 0.15 mm, and the application is suitable for various medium and low-end industrial robots with a closed controller bottom layer parameter interface.

Description

Technical Field

[0001] This invention relates to the field of industrial robot motion accuracy compensation technology, specifically to a method for achieving absolute positioning accuracy compensation of a robot by utilizing the mapping relationship between two models (corrected kinematic model and nominal kinematic model) and an external compensation controller. This method is applicable to multi-axis industrial robots whose controller kinematic parameter reading and writing interfaces are not open. Background Technology

[0002] The absolute positioning accuracy of an industrial robot's end effector is a core indicator determining its ability to perform high-end manufacturing tasks such as precision machining, assembly, welding, grinding, and painting. Technical approaches to improving robot positioning accuracy mainly fall into two categories: error prevention and error compensation. Error compensation has become the mainstream solution due to its low cost and quick results. Its core logic is to use high-precision external measuring equipment to identify parameter deviations in the robot's kinematic model and then correct the kinematic model within the controller.

[0003] Existing robot calibration and compensation technologies have the following main shortcomings:

[0004] First, traditional compensation schemes heavily rely on the robot's native controller's parameter writing permissions. The standard compensation process uses devices such as laser trackers to identify errors such as DH parameter deviations and joint clearances, then directly writes the corrected kinematic parameters into the robot's native controller. However, many economical and low-to-mid-range industrial robot manufacturers have closed the underlying parameter read / write interfaces of their native controllers to protect their core technologies. External devices cannot write corrected parameters, making it difficult to implement error data obtained from offline calibration, and high-precision calibration technology has long remained in the laboratory stage.

[0005] Second, existing external compensation schemes struggle to balance compensation accuracy and motion smoothness. Some external compensation methods only perform simple error interpolation in Cartesian space, failing to fully consider joint space kinematic constraints. The compensated trajectory may conflict with the robot's native joint motion logic, easily leading to end effector jitter and trajectory distortion.

[0006] Existing patent technology 1, publication number: CN109746915B, patent title: A kinematic method for improving the absolute positioning accuracy of industrial robots. Although this method proposes a kinematic compensation method combining nominal inverse kinematics, forward kinematics with errors, and Jacobian inverse solution, its essence is to design three algorithm modules (nominal inverse kinematics module, forward kinematics with geometric errors module, and Jacobian matrix module) within the robot controller and solve the joint space deviation through iteration. This method is essentially still a built-in solution of the controller, which requires modification or expansion of the robot controller's software architecture and cannot be applied to low-end and mid-range robots with closed controller interfaces. At the same time, its iterative process is entirely switching between Cartesian space and joint space, without introducing axis-by-axis amplitude constraints in the joint space, which can easily lead to divergence of compensation when the Jacobian matrix approaches singularity. In addition, existing patent technology 2, publication number: CN117415816B, patent title: A dual-model compensation control method for a six-joint robot. This method achieves compensation by establishing a low-precision fitting model and a high-precision actual model, but it still needs to integrate dual-model calculations within the robot controller, and is essentially still a built-in solution of the controller. None of the above-mentioned solutions have achieved non-intrusive closed-loop compensation for external independent controllers.

[0007] Third, the application scope of existing dual-PSD calibration devices is limited. Currently, dual-PSD (position-sensitive detector) laser calibration devices are mainly used for offline parameter identification before startup and are removed after calibration, failing to form a continuously linked closed-loop architecture with the compensation system. They cannot collect residual positioning errors in real time and iteratively update the compensation amount during continuous robot operation, making it difficult to cope with the dynamic error accumulation introduced by factors such as thermal deformation and load fluctuations during long-term operation. Although some research has proposed dual-PSD calibration systems, they only solve static calibration problems such as zero-position deviation and do not extend dual-PSD to real-time error monitoring and dynamic compensation throughout the entire operation process.

[0008] Fourth, existing compensation methods lack hierarchical error processing and adaptive optimization mechanisms. Traditional compensation methods typically treat all error sources as a single entity and correct them all at once using a single compensation strategy. This fails to distinguish between error components from different sources and time scales, such as static geometric errors, thermal deformation drift errors, and load disturbance errors. Consequently, the compensation strategies lack specificity and adaptability.

[0009] Fifth, existing compensation solutions lack the ability to be deployed rapidly in large batches. Whether it is a built-in controller solution or an external measurement solution, each robot needs to undergo a complete calibration and parameter identification process, which is time-consuming and costly in the scenario of mass production line transformation.

[0010] In summary, there is an urgent need to provide an industrial robot positioning accuracy compensation method based on dual-model mapping to solve the problems existing in the above-mentioned technologies. Summary of the Invention

[0011] This invention addresses the problems of existing industrial robot compensation schemes, which require rewriting the internal kinematic parameters of the robot's native controller, making them unsuitable for low- to mid-range robots with closed parameter interfaces. Furthermore, existing external compensation methods lack joint space motion constraints, easily leading to trajectory jitter; dual PSD calibration devices are only used for offline calibration and cannot form online closed-loop compensation; compensation strategies lack hierarchical error processing and adaptive optimization; and there is a lack of rapid batch deployment capabilities. This invention provides an industrial robot positioning accuracy compensation method based on dual-model mapping.

[0012] A method for compensating the positioning accuracy of industrial robots based on dual-model mapping is implemented by the following steps:

[0013] Step 1: Build a dual PSD laser calibration system and identify kinematic parameters to obtain the optimal estimate of the error of all kinematic parameters;

[0014] Step 2: Construct a dual kinematic model stored in an external compensation controller, including a corrected model and a nominal model; the corrected model is used to implement the robot's compensation control, and the nominal model corresponds to the nominal kinematic model inside the robot's native controller;

[0015] Step 3: Based on the dual kinematic model constructed in Step 2, perform joint space compensation calculations for the mapping of the dual kinematic model to generate corrected joint angle commands;

[0016] Step 4: The external compensation controller performs compensation calculations on the original joint angle command output by the robot's native controller based on the corrected joint angle command obtained in Step 3, and transmits the corrected joint angle command obtained after compensation calculation to the robot servo driver.

[0017] Step 5: The dual PSD laser calibration device in the dual PSD laser calibration system continuously collects the actual pose of the robot end effector online. The external compensation controller compares the actual pose of the robot end effector with the target pose and calculates the residual positioning error. When the positioning error meets the set convergence threshold, the industrial robot positioning accuracy compensation is completed.

[0018] The beneficial effects of this invention are as follows: Without modifying any internal parameters, control program, or interpolation logic of the robot's native controller, the method described in this invention independently establishes a mapping relationship between the corrected kinematic model and the nominal kinematic model through an external compensation controller. This accurately maps Cartesian space positioning deviations to joint space to generate compensation amounts. Simultaneously, it introduces a joint space axis-by-axis constraint threshold, a damped least squares numerical stabilization strategy, and a hierarchical error adaptive processing mechanism to achieve high-precision, high-smoothness, and rapidly deployable, non-intrusive absolute positioning accuracy compensation. It possesses the following technical advantages:

[0019] 1. Zero-Intrusion and Outstanding Versatility: This invention does not modify the robot's native controller's internal kinematic parameters, control program, or interpolation logic throughout the entire process, perfectly adapting to various low-to-mid-range industrial robots with closed underlying parameter interfaces. Unlike existing patent technology 1, which requires adding a nominal inverse kinematics module, an error-based forward kinematics module, and a Jacobian matrix module within the controller, this invention places all compensation calculations entirely within an external compensation controller. The native controller merely acts as the trajectory command issuer, requiring no modification to its internal software architecture. Unlike existing patent technology 2, which requires integrating dual-model calculations within the controller, this invention's dual models run entirely within the external controller, fundamentally solving the industry bottleneck of closed controller interfaces preventing the implementation of calibration results. The external compensation controller can be independently packaged as a hardware module or embedded software. External integration requires no modification to the robot itself, with a single-unit modification cycle of 4-6 hours, suitable for rapid deployment on large-scale production lines.

[0020] 2. High Accuracy of Dual-Model Mapping Compensation: This invention establishes a differential mapping relationship between a corrected kinematic model with parameter errors and an error-free nominal kinematic model, with clear physical meaning and a complete error propagation chain. By using the damped least-squares generalized inverse of the weighted Jacobian matrix, the Cartesian space positioning deviation is accurately mapped to the joint space. After compensation, the absolute positioning accuracy is better than 0.15mm, significantly better than traditional external interpolation schemes (typically 0.3~0.5mm).

[0021] 3. Smooth and jitter-free motion: This invention introduces an axis-by-axis compensation constraint threshold and a damped least-squares numerical stabilization strategy in joint space, effectively avoiding compensation divergence caused by ill-conditioned Jacobian matrices near singular poses. The mechanism of over-limit truncation combined with multiple iterative compensation ensures that the single compensation amount does not exceed the joint motion limit, fundamentally eliminating end effector jitter and trajectory distortion, and guaranteeing the smoothness of high-speed robot motion. Compared with existing technologies such as [Technology 1] that do not introduce axis-by-axis amplitude constraints in joint space, the joint constraint mechanism of this invention fundamentally solves the motion safety problem caused by compensation divergence when the Jacobian matrix is ​​singular.

[0022] 4. Layered Error Processing and Adaptive Optimization: This invention decomposes residual errors into three categories: static geometric errors, thermal deformation drift errors, and load disturbance errors, and configures differentiated compensation strategies and update weights for each. Real-time temperature compensation for thermal deformation and online RLS identification mechanism for load disturbances enable the compensation system to adapt to dynamic operating conditions, resulting in significantly better long-term operational accuracy and stability compared to traditional single-loop compensation schemes.

[0023] 5. Dual PSD online measurement throughout the entire process forms a true closed loop: This invention breaks through the traditional limitation of dual PSD being used only for offline calibration, transforming the dual PSD laser calibration device into a measurement device that is continuously online throughout the entire operation. Through anti-interference measures such as laser modulation and phase-locked amplification, its measurement reliability in the workshop environment is ensured, realizing the functional leap from "removal after calibration" to "online monitoring throughout the entire process", forming a true dynamic closed-loop compensation for the entire operation cycle.

[0024] 6. High efficiency in batch deployment: By modifying the model solidification storage and the reuse mechanism of robot models of the same model, this invention only requires the first robot to be fully calibrated. The remaining robots in the same batch can be directly imported to complete the deployment, improving the efficiency of batch transformation by more than 80% and significantly reducing the overall cost of production line precision upgrade. Attached Figure Description

[0025] Figure 1 This is a flowchart of the industrial robot positioning accuracy compensation method based on dual-model mapping described in this invention.

[0026] Figure 2 This is a block diagram illustrating the principle of the dual PSD laser calibration system in the industrial robot positioning accuracy compensation method based on dual model mapping described in this invention.

[0027] Figure 3 A block diagram illustrating the principle of compensation control for mapping between the nominal model and the modified model;

[0028] Figure 4 This is a block diagram of the hierarchical error separation and adaptive compensation strategy.

[0029] In the diagram: 1. Industrial robot; 2. Laser; 3. Robot native controller; 4. Servo driver; 5. Dual PSD laser calibration device; 6. External compensation controller; 7. Temperature sensor array. Detailed Implementation

[0030] Specific Implementation Method 1: Combination Figures 1 to 4 This embodiment describes an industrial robot positioning accuracy compensation method based on dual-model mapping, which is implemented through the following steps:

[0031] Step S1. The dual PSD laser calibration system is built and kinematic parameters are identified to obtain the optimal estimates of all kinematic parameter errors. The specific implementation process is as follows:

[0032] Step S11. Set up the dual PSD calibration system;

[0033] like Figure 2As shown, a laser 2 is installed on the side of the end joint of the industrial robot 1. The laser beam emitted by the laser 2 is in the same direction as the Z-axis of the end flange coordinate system. Temperature sensors are installed on each joint of the industrial robot 1 to form a temperature sensor array 7, and a communication link is established between the array and the external compensation controller 6 to read the real-time temperature of each joint. A communication link is also established between the external compensation controller 6 and the robot's native controller 3 and servo driver 4 for data reading and signal transmission. That is, the external compensation controller 6 is electrically connected to the servo driver 4, the robot's native controller 3, the temperature sensor array 7, and the dual PSD laser calibration device 5. The dual PSD laser calibration device 5 is horizontally fixed in the working space of the industrial robot 1 to receive the laser beam emitted by the laser 2 for calibration of the industrial robot 1. Before calibration, a high-precision theodolite is used to calibrate the spatial pose of the dual PSD laser calibration device 5 to determine the precise coordinates of the dual PSD photosensitive surfaces in the base coordinate system.

[0034] Step S12. Full workspace calibration pose sampling;

[0035] The industrial robot 1 is controlled to traverse no fewer than 40 sets of spatially dispersed standard calibration poses. The pose selection follows the D-optimal observability optimization criterion to ensure that the calibration poses are uniformly distributed throughout the entire workspace of the industrial robot 1, fully stimulating all kinematic parameters. Each pose is held statically for more than 3 seconds, and the dual PSD laser calibration device 5 synchronously acquires the two-dimensional coordinates of the laser spot illuminating the dual PSD photosensitive surface. At the same time, the external compensation controller 6 records the joint angle values ​​output by the robot's native controller 3.

[0036] Step S13. Three-dimensional reconstruction of the end effector pose based on dual PSDs;

[0037] Based on the dual-PSD robot calibration principle, and according to the two-dimensional coordinates of the light spot on the photosensitive surface of the dual-PSD, as well as the spatial structural relationship between the industrial robot 1 and the dual-PSD laser calibration device 5, the laser beam direction vector and the spatial position of the laser 2 are reconstructed. Combined with the geometric relationship between the end flange and the laser 2, the actual end pose of the end flange center in the base coordinate system, including its three-dimensional position, can be reconstructed. With three-axis attitude .

[0038] Step S14. Identification of kinematic parameter errors;

[0039] A robot kinematic error model is established based on the nominal Denavit-Hartenberg (DH) model. For an n-DOF serial industrial robot, four nominal DH parameters (a, α, d, θ) are introduced for each joint, where a is the link length, α is the link torsional pitch, d is the link offset, and θ is the joint angle. Two additional error parameters, perpendicularity and parallelism, are introduced, resulting in a total of 6n error parameters to be identified. Based on these error parameters, the joint angle values ​​output by the robot's native controller 3 recorded in step S12 and the actual end effector pose reconstructed in step S13 are substituted into the kinematic error model (overdetermined equations).

[0040]

[0041] in, This is the end-effector pose error vector. Let Jacobian matrix be the error parameter. Let be the vector of error parameters to be identified. The overdetermined equations are solved using the Levenberg-Marquardt damped iterative least squares method to obtain the optimal estimates of all error parameters to be identified, i.e., the optimal estimates of all kinematic parameter errors.

[0042] Step S2. Construct a dual kinematic model consisting of a modified kinematic model and a nominal kinematic model;

[0043] Step S21. Construct a modified kinematic model (modified model);

[0044] The optimal estimates of all kinematic parameter errors identified in step S14 are superimposed onto the nominal DH parameters to construct a corrected model containing parameter errors, denoted as... The positive kinematic function of the model. The expression is:

[0045]

[0046] in, Let be the homogeneous transformation matrix of the actual end-flange center position in the base coordinate system. For the first The actual angles of each joint. The corrected model accurately reflects the robot's current true kinematic characteristics, including all identifiable static geometric errors such as DH parameter deviation, link machining and assembly errors, and static offset of joint clearances.

[0047] Step S22. Establish a nominal kinematic model (nominal model);

[0048] An error-free nominal model is constructed using the nominal DH parameters provided by the industrial robot manufacturer, denoted as . The nominal model corresponds to the nominal kinematics model used inside the robot's native controller, and is the theoretical basis for the native controller to calculate inverse kinematics and trajectory planning.

[0049] Step S23. Differentiated storage and retrieval mechanism for dual kinematic models;

[0050] The corrected model and the nominal model are stored in the external compensation controller with independent data structures. The two models can be called independently to perform forward kinematics calculations. The corrected model is only called during the compensation calculation phase, and no model parameters are imported or written to the robot's native controller throughout the process, ensuring zero intrusion into the native controller.

[0051] Step S3. Calculate the joint space compensation amount based on the dual kinematic model mapping, and generate the corrected joint angle command, such as... Figure 3 As shown, the specific process is as follows:

[0052] Step S31. Solve the nominal inverse kinematics of the robot's end-effector pose;

[0053] Given the robot end-effector target pose Using a nominal model inverse kinematic function Solve for the corresponding nominal joint angles :

[0054]

[0055] If multiple solutions exist (e.g., a six-axis robot typically has 8 inverse solutions), select the solution that is closest to the current actual joint angle according to the shortest stroke criterion to ensure trajectory continuity.

[0056] Step S32. Calculation of end-effector pose deviation under nominal joint angle;

[0057] The nominal joint angle obtained in step S31 Substitute into the correction model Calculate the actual reachable pose of the robot's end effector:

[0058]

[0059] Calculate the deviation between the actual reachable pose and the target pose, i.e., the Cartesian space positioning deviation;

[0060]

[0061] in, Positioning bias in Cartesian space, including three-dimensional position bias. With three-axis attitude deviation ,in, These represent the robot end effector in the base coordinate system. Positional deviation along the three axes; These represent the robot end effector in the base coordinate system. Attitude deviations along three axes. The physical meaning of these deviations is: if the robot's native controller outputs according to the nominal model... Driving the joints involves parameter errors between the corrected model and the nominal model, resulting in an actual reachable pose at the end effector that deviates from the target pose; in other words, there is a discrepancy between the two. .

[0062] Step S33. Damped least-squares generalized inverse mapping of the weighted Jacobian matrix;

[0063] Based on the principles of robot kinematics, using the current nominal joint angles The Jacobian matrix J at that location will indicate the Cartesian spatial positioning deviation. Mapping to joint space to obtain joint space error compensation amount ;

[0064]

[0065] in, Jacobian matrix The generalized inverse (Moore-Penrose pseudo-inverse) is calculated. To overcome the ill-conditioned problem of the Jacobian matrix near singular poses, damped least squares (DLS) is used to calculate the generalized inverse. ;

[0066]

[0067] in, Jacobian matrix The transpose of the matrix, The damping coefficient is... The condition number of the Jacobian matrix is ​​adaptively adjusted as the condition number changes: the larger the condition number (the closer it is to singular), the larger the damping coefficient, ensuring numerical stability; when the condition number is small, the damping coefficient approaches zero, degenerating into a standard pseudo-inverse to ensure accuracy.

[0068] In this embodiment, a diagonal weighting matrix is ​​introduced to assign different degrees of compensation to pose deviations in different directions, thereby obtaining the weighted joint space error compensation amount:

[0069]

[0070] in, This is a diagonally weighted matrix, and the weights for position and attitude deviations can be configured independently. For applications where position accuracy is paramount (such as grinding and welding), the position weight coefficient is set to 1.0, and the attitude weight is set to 0.3; for applications that require simultaneous assurance of attitude accuracy (such as assembly), both weights are set to 1.0.

[0071] Step S34. Joint space compensation constraint threshold processing;

[0072] The external compensation controller internally stores the maximum permissible single compensation offset threshold for each joint. The threshold is determined by comprehensively considering the load-bearing capacity of each joint reducer, the torque limit of the servo motor, and the maximum allowable trajectory deviation at the end effector. The typical value is 0.5°~1.5° (depending on the robot model and working conditions).

[0073] The weighted joint space error compensation amount calculated in step S33 The amplitude of the compensation amount for each joint is checked one by one:

[0074]

[0075] in, For the first The amount of compensation for each joint. For the first The maximum permissible single-time compensation offset threshold for each joint. The sign function is represented. The weighted joint space error compensation amount after verification constraint processing is obtained. .

[0076] In this embodiment, after amplitude verification, the Cartesian space positioning deviation after verification constraint is recalculated through forward kinematics. If the positioning deviation is still greater than the allowable threshold, a multi-iteration compensation strategy is adopted to decompose the large deviation into multiple small step-size compensations to avoid the motion change caused by excessive single compensation.

[0077] Step S35. Correct the joint angle command synthesis;

[0078] The weighted joint space error compensation amount after S34 has undergone verification constraint processing. With nominal joint angle Superimpose to generate joint angle correction commands :

[0079]

[0080] Step S4: The external compensation controller executes commands non-intrusively. The specific process is as follows:

[0081] Step S41. External compensation controller hardware architecture;

[0082] The external compensation controller uses an industrial-grade embedded industrial computer equipped with a real-time industrial Ethernet interface, and establishes independent communication channels with the robot's native controller, servo driver, temperature sensor array, and dual PSD laser calibration device.

[0083] Step S42. Trajectory command bypass interception mechanism;

[0084] The raw joint angle commands (or Cartesian trajectory points) output by the robot's native controller are not directly sent to the servo drivers, but are first transmitted to the external compensation controller via a communication channel. The external compensation controller intercepts the raw commands for each cycle at a frequency synchronized with the interpolation cycle of the native controller (typically 1kHz~4kHz).

[0085] Step S43. Real-time calculation and issuance of correction instructions;

[0086] For each cycle of the original command intercepted by the external compensation controller 6, the complete compensation algorithm of step S3 is invoked in real time to calculate the corresponding corrected joint angle command. Then, the joint angle correction command is sent to the robot servo driver 4 for execution via an independent communication channel. The entire compensation calculation is completed within a single interpolation cycle (typically <200μs), without introducing additional trajectory delay.

[0087] In this embodiment, the internal kinematic model, control parameters, and interpolation program of the robot's native controller remain unchanged throughout the entire process. The native controller only acts as the issuer of trajectory commands; the actual commands driving joint movements are calculated and issued by the external compensation controller. The native controller is completely unaware of the existence of the external compensation controller, ensuring zero intrusion, and its internal logic and parameters remain in their factory state.

[0088] Step S5. Real-time correction and iterative optimization of residual positioning error closed loop;

[0089] Step S51. Measurement of residual positioning error;

[0090] After the compensation command is executed, the dual PSD laser calibration device 5 continuously acquires the actual pose of the robot's end effector online (without removal, maintaining a working state throughout the process), and the external compensation controller 6 compares the actual reachable pose of the robot's end effector with the target pose. Compare and calculate the residual positioning error. .

[0091] S52. Convergence determination and iterative compensation;

[0092] like , The convergence threshold (typically 0.15mm) is used to determine that the compensation has converged, and the current target pose compensation is complete.

[0093] like The deviation between the current achievable pose of the robot's end effector and the target pose is taken as the new deviation. Input the data and repeat steps S3-S4 for the second round of compensation iteration. The maximum number of iterations is preset to 5. If convergence is not achieved after exceeding the maximum, a system alarm will be triggered, prompting that kinematic parameter identification needs to be performed again.

[0094] like Figure 4 As shown, in this embodiment, a hierarchical error separation and adaptive compensation strategy is adopted;

[0095] The residual positioning error is compensated by an external compensation controller 6. The error is broken down into three different sources:

[0096] Category 1: Static geometric errors;

[0097] The main cause is the identification of kinematic parameters in the early stage, which manifests as a systematic bias strongly correlated with pose. The corresponding compensation strategy is to reuse the joint space compensation calculation process based on the dual kinematic model mapping in step S3, and then verify the weighted joint space error compensation amount after constraint processing. Calculate and update weights Set to a low refresh rate, updating only once each time a point is switched, without consuming real-time interpolation computing power.

[0098] Category 2: Thermal deformation drift error;

[0099] The primary cause is the thermal expansion of the connecting links due to joint friction during continuous operation of the industrial robot, manifesting as a drift deviation that accumulates slowly over time. The corresponding compensation strategy is to introduce a temperature state variable into the correction model. By monitoring the joint temperature in real time, a temperature-error linear regression model is established to obtain the compensation amount for the joint angle error caused by the thermal deformation error of each joint. Set medium-speed update weight It is adapted to slow thermal drift during long-term continuous operation.

[0100] Category 3: Load disturbance error;

[0101] The main causes are variations in workpiece weight, clamping offsets, and external forces, manifesting as sudden or periodic fluctuations in deviation. The corresponding compensation strategy is to introduce load disturbance variables based on a modified model, and use recursive least squares (RLS) to identify the joint flexibility deformation parameters caused by the load online, thus obtaining the compensation amount for the joint angle error caused by the load disturbance. Set high-speed response weights This helps to suppress sudden fluctuations in error.

[0102] The compensation components for the above three types of errors are: the weighted joint space error compensation amount after verification constraint processing. Compensation amount for joint angle error caused by thermal deformation error Joint angle error compensation amount caused by load disturbance The individual calculations are then linearly superimposed to form the final comprehensive compensation amount. .

[0103] This embodiment also includes step S6. Compensation model storage and batch reuse; the specific implementation process is as follows:

[0104] Step S61. Solidify and store the compensation model;

[0105] After completing all calibration and compensation parameter optimization, the corrected model, nominal model, Jacobian matrix, maximum permissible single compensation offset threshold, corrected joint angle command, and temperature-error linear regression model are all encapsulated into independent compensation model files and stored in the external compensation controller 6.

[0106] Step S62. Rapid batch deployment;

[0107] For industrial robots of the same model and batch, only the first robot needs to be calibrated and compensated using steps S1-S5. After obtaining the compensation model parameters, the model file can be directly copied and imported into the external compensation controller 6 of other industrial robots of the same model to complete the deployment. There is no need to repeat the calibration for each robot, which greatly reduces the time and cost of mass production line transformation.

[0108] Step S63. Regular model updates and adaptive maintenance;

[0109] During equipment operation, the external compensation controller 6 continuously monitors the robot's positioning error (residual positioning error). The statistical characteristics (mean, variance, drift trend) of the residual positioning error are analyzed. When the statistical mean of the residual positioning error exceeds the allowable threshold for 100 consecutive sampling cycles, an incremental model update is automatically triggered: using the current corrected model as the initial value, the error data of the most recent production cycle is collected, and lightweight recalibration is performed using the process of steps S1-S5. Only the error parameters that have changed significantly (mainly parameters related to thermal deformation and wear) are updated, without the need for complete recalibration, thus achieving adaptive accuracy maintenance for long-term operation.

[0110] Specific Implementation Method Two: This implementation method is a verification example of the industrial robot positioning accuracy compensation method based on dual-model mapping described in Specific Implementation Method One. The comparative experiment between the method described in Specific Implementation Method One and the traditional controller built-in dual-model compensation scheme (traditional built-in compensation scheme) is divided into setting experimental conditions; experimental groups, as shown in Table 1; experimental results, as shown in Table 2; and experimental results.

[0111] Experimental conditions: The experimental object was the same six-axis industrial robot (rated load 10kg), and the ambient temperature was 25±2℃. The measuring equipment was a dual PSD laser calibration device (calibrated by a laser interferometer, with a measurement accuracy of ±0.02mm).

[0112] Table 1

[0113]

[0114] Table 2

[0115]

[0116] Based on the above comparison, the following experimental results can be drawn:

[0117] Accuracy Comparison: The average position error after compensation by this invention is 0.09mm, which is better than the 0.26mm of the traditional built-in compensation scheme. The reason is that this invention introduces a weighted matrix and a damped least squares strategy, which can maintain stable mapping accuracy even in the ill-conditioned region of the Jacobian matrix, while the traditional built-in compensation scheme uses a standard Jacobian inverse matrix, which is prone to numerical instability near singular poses.

[0118] Comparison of motion safety: Traditional built-in compensation schemes measured a maximum joint compensation of 2.35° near singular poses, which exceeds the single-time allowable compensation limit of most industrial robots and poses a risk of mechanical collision; the present invention limits the maximum compensation to 0.62° by constraining thresholds for each axis (0.6°~1.2° for each joint), which fundamentally ensures motion safety.

[0119] Long-term stability comparison: Traditional built-in compensation schemes are open-loop compensations, which cannot cope with dynamic errors introduced by thermal deformation and load fluctuations, with an 8-hour accuracy drift of +0.35mm; This invention controls the 8-hour drift to +0.06mm through dual PSD full-process online measurement and layered error adaptive mechanism.

[0120] General Comparison: Traditional built-in compensation schemes require the addition of three algorithm modules (nominal inverse kinematics module, forward kinematics module with error, and Jacobian matrix module) inside the robot controller, requiring the controller to have secondary software development capabilities; all calculations in this invention are completed in an independent external controller, with zero intrusion into the native controller.

[0121] Specific Implementation Method 3: This implementation method uses the industrial robot positioning accuracy compensation method based on dual-model mapping described in Specific Implementation Method 1 to rapidly deploy robots of the same model in batches, illustrating the efficient deployment advantages of this invention in mass production line transformation.

[0122] In this embodiment, a car parts manufacturing company has 12 identical six-axis industrial robots (the same model as in Embodiment 1) used for aluminum alloy welding processes, requiring an absolute positioning accuracy of ≤0.2mm. Traditional solutions require laser tracker calibration and compensation for each robot individually, taking approximately 8 hours per robot, totaling 96 hours for all 12 robots, resulting in significant production line downtime losses.

[0123] The following operations are performed using the method described in Specific Implementation Method 1:

[0124] First complete calibration: Perform the complete process of Implementation Method 1 (steps S1-S5) on the first robot, which takes about 6 hours (including hardware installation, calibration sampling, parameter identification, and verification testing), and generate a compensation model file (about 120KB in size, containing 36 parameters of the actual model, Jacobian mapping parameters, joint constraint thresholds, hierarchical error weights, etc.).

[0125] Batch Model Copying: Import the compensation model file into the external compensation controllers of the remaining 11 robots via USB flash drive or Ethernet. The import time for each robot is less than 5 minutes.

[0126] Quick verification: After importing the model, each robot only needs to run 6 verification points (taking about 3 minutes) to confirm the compensation effect.

[0127] In this embodiment, the total modification time for 12 robots was approximately 6.5 hours, which is 93% more efficient than the traditional solution. Random inspection with a coordinate measuring machine showed that the absolute positioning accuracy of all 12 robots after compensation was consistently within the range of 0.12~0.18mm, meeting the production line process requirements.

[0128] Specific Implementation Method Four: This implementation method uses the industrial robot positioning accuracy compensation method based on dual-model mapping described in Specific Implementation Method One to achieve adaptive compensation under dynamic working conditions, illustrating the adaptive compensation capability of the present invention under dynamic working conditions such as load changes and thermal deformation.

[0129] A robot at a certain grinding station is set up such that the weight of the workpiece varies from 3 to 8 kg. After running continuously for 4 hours, the joint temperature rise can reach 15°C.

[0130] Load disturbance adaptation: The system adopts the recursive least squares (RLS) online identification algorithm to estimate the joint flexibility deformation parameters caused by the load in real time at a frequency of 100Hz. When the workpiece is switched from 3kg to 8kg, the system detects a sudden increase of 0.35mm in pose deviation within 0.5 seconds. The RLS algorithm completes parameter convergence within 1.2 seconds, and the residual error falls back to 0.10mm after compensation.

[0131] Thermal deformation adaptive design: PT100 temperature sensors mounted on each joint upload temperature data at a frequency of 10Hz. The system establishes a temperature-error linear regression model. ,in, Represents the linear regression coefficients. This indicates the change in joint temperature relative to the reference temperature. After 4 hours of continuous operation, joint 2's temperature rose by 14.2℃ and joint 3's temperature rose by 11.5℃. The thermal deformation compensation model automatically increased the compensation by 0.114mm and 0.092mm respectively, taking into account the residual positioning error. The entire process is controlled within 0.15mm.

[0132] Long-term stability: After 30 days of continuous operation, the system monitored that the average static residual error slowly increased from 0.08 mm to 0.13 mm (mainly due to increased joint clearance caused by reducer wear). The system automatically triggered incremental recalibration, collecting error data from the most recent 200 sets of operating trajectories, updating only the joint clearance-related parameters (4 parameters), and the recalibration time was less than 2 minutes. After completion, the residual error recovered to 0.09 mm.

[0133] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0134] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A positioning accuracy compensation method for industrial robots based on dual-model mapping, characterized by: The compensation method is implemented by the following steps: Step 1: Build a dual PSD laser calibration system and identify kinematic parameters to obtain the optimal estimate of the error of all kinematic parameters; The dual PSD laser calibration system includes a robot, a laser, a temperature sensor array, a dual PSD laser calibration device, and an external compensation controller. A laser is installed on the side of the robot's end joint, and the laser beam emitted by the laser is in the same direction as the Z-axis of the end flange coordinate system; temperature sensors are installed on each joint of the robot to form a temperature sensor array, and the external compensation controller is used to read the real-time temperature of each joint; Establish a communication link between the external compensation controller, the robot's native controller, and the servo driver for data reading and signal transmission; A dual PSD laser calibration device is horizontally fixed in the robot's workspace to receive the laser beam emitted by the laser and calibrate the robot. Step 2: Construct a dual kinematic model stored in an external compensation controller, including a corrected model and a nominal model; the corrected model is used to implement the robot's compensation control, and the nominal model corresponds to the nominal kinematic model inside the robot's native controller; Step 3: Based on the dual kinematic model constructed in Step 2, perform joint space compensation calculations for the mapping of the dual kinematic model to generate corrected joint angle commands; Step 4: The external compensation controller performs compensation calculations on the original joint angle command output by the robot's native controller according to the corrected joint angle command described in Step 3, and transmits the corrected joint angle command after compensation calculation to the robot servo driver. Step 5: The dual PSD laser calibration device in the dual PSD laser calibration system continuously collects the actual reachable pose of the robot end effector online. The external compensation controller compares the actual reachable pose with the target pose and calculates the residual positioning error. When the residual positioning error meets the set convergence threshold, the robot positioning accuracy compensation is completed.

2. The industrial robot positioning accuracy compensation method based on dual-model mapping according to claim 1, characterized in that: In step one, the specific process of obtaining the optimal estimates of all kinematic parameter errors is as follows: Step 1: Control the robot to acquire multiple sets of spatially dispersed standard calibration poses. The dual PSD laser calibration device synchronously collects the two-dimensional coordinates of the light spot irradiated by the laser beam on the dual PSD photosensitive surface. At the same time, an external compensation controller records the joint angle values ​​output by the robot's native controller. Step 1 and 2: Based on the dual PSD robot calibration principle, according to the two-dimensional coordinates of the light spot on the dual PSD photosensitive surface and the spatial structural relationship between the robot and the dual PSD laser calibration device, the laser beam direction vector and the spatial position of the laser are reconstructed. Combined with the geometric relationship between the end flange and the laser installation, the actual end pose of the robot end flange center in the base coordinate system is obtained. Step 13: Establish a robot kinematic error model, set the error parameters to be identified, and substitute the joint angle values ​​described in Step 11 and the actual end pose obtained in Step 12 into the kinematic error model to obtain the optimal estimate of all kinematic parameter errors.

3. The industrial robot positioning accuracy compensation method based on dual-model mapping according to claim 2, characterized in that: In step two, the construction process of the dual kinematic model is as follows: Constructing a corrected model: The optimal estimates of all kinematic parameter errors obtained in steps one and three are superimposed onto the nominal DH parameters to construct a corrected model containing parameter errors; the positive kinematic function of the corrected model... The expression is: ； in, Let be the actual end-position homogeneous transformation matrix. For the first The actual angle of each joint; Construct a nominal model: Construct an error-free nominal model using the nominal DH parameters.

4. The industrial robot positioning accuracy compensation method based on dual-model mapping according to claim 3, characterized in that: In step three, the specific process of generating the joint angle correction command is as follows: Step 3:

1. Solve for the corresponding nominal joint angles of the set robot end-effector target pose using the inverse kinematics function of the nominal model; Step 32: Substitute the nominal joint angles obtained in Step 31 into the correction model to calculate the actual reachable pose of the robot end effector; and calculate the deviation between the actual reachable pose and the target pose, i.e., the Cartesian space deviation. Step 3: Based on the principles of robot kinematics, the Cartesian space deviation is mapped to the joint space using the Jacobian matrix at the current nominal joint angle to obtain the joint space error compensation amount; a diagonal weighted matrix is ​​introduced to obtain the weighted joint space error compensation amount. Step 3-4: Perform amplitude verification on each joint compensation amount in the weighted joint space error compensation amount obtained in Step 3-3 to obtain the weighted joint space error compensation amount after verification constraint. Step 35: Superimpose the weighted joint space error compensation amount after verification constraint with the nominal joint angle to generate a corrected joint angle command.

5. The industrial robot positioning accuracy compensation method based on dual-model mapping according to claim 4, characterized in that: In step 3.2, the actual reachable pose of the robot's end effector. : ； in, To correct the positive kinematic function of the model, The nominal joint angle is used; the deviation between the actual reachable pose and the target pose is calculated: ； in, For Cartesian spatial positioning bias, The target pose.

6. The industrial robot positioning accuracy compensation method based on dual-model mapping according to claim 5, characterized in that: The joint space error compensation amount Represented as: ； In the formula, Jacobian matrix The generalized inverse is calculated using the damped least squares method. ; ; In the formula, Jacobian matrix The transpose of the matrix, The damping coefficient is... It is the identity matrix; By introducing a diagonal weighting matrix, different degrees of compensation are assigned to pose deviations in different directions, thus obtaining the weighted joint space error compensation amount. : ； In the formula, W is a diagonal weighted matrix; The weighted joint space error compensation amount Δθ W The amplitude of the compensation amount for each joint is checked one by one: ； In the formula, For the first The amount of compensation for each joint. For the first The maximum permissible single-time compensation offset threshold for each joint. The sign function is used to obtain the weighted joint space error compensation amount Δθ after verification and constraint processing. C .

7. The industrial robot positioning accuracy compensation method based on dual-model mapping according to claim 6, characterized in that: In step five, the external compensation controller compares the actual reachable pose of the robot's end effector with the target pose and calculates the residual positioning error. ; like , The convergence threshold is used to determine if the compensation has converged, and the current target pose compensation is complete. like The deviation between the current achievable pose of the robot's end effector and the target pose is taken as the new deviation. Input, and repeat steps three and four for the second round of compensation iteration.

8. The industrial robot positioning accuracy compensation method based on dual-model mapping according to claim 7, characterized in that: External compensation controller for residual positioning error The errors are classified and broken down into residual static geometric errors, thermal deformation drift errors, and load disturbance errors. The compensation strategy for the static geometric error is as follows: the modified model described in step three is used for compensation to obtain the weighted joint space error compensation amount after verification and constraint processing. The compensation strategy for the thermal deformation drift error is as follows: introduce the temperature state variable into the correction model, establish a temperature-error linear regression model by real-time monitoring of joint temperature, and obtain the compensation amount for the joint angle error caused by the thermal deformation error of each joint. The compensation strategy for the load disturbance error is as follows: a load disturbance variable is introduced into the modified model, and the joint flexibility deformation parameters caused by the load are identified online using the recursive least squares method to obtain the compensation amount for the joint angle error caused by the load disturbance. The weighted joint space error compensation amount after verification and constraint processing, the joint angle error compensation amount caused by thermal deformation error, and the joint angle error compensation amount caused by load disturbance are calculated independently and then linearly superimposed to form the final comprehensive compensation amount.

9. The industrial robot positioning accuracy compensation method based on dual-model mapping according to any one of claims 1-8, characterized in that: The process also includes step six, which involves encapsulating the corrected model, nominal model, Jacobi matrix, maximum permissible single compensation offset threshold, corrected joint angle command, and temperature-error linear regression model into independent compensation model files after completing all calibration and compensation parameter optimization. These files are then stored in an external compensation controller to enable batch deployment, regular updates, and adaptive maintenance of the models.

Images