Automatic deviation correction control method and system for axle box body cleaning and polishing station

Abstract

The invention discloses an axle box body cleaning and polishing station automatic deviation rectification control method and system, and relates to the technical field of industrial visual inspection and automatic control, and the method comprises the steps: recognizing the model of a workpiece through an RFID reader-writer, and loading corresponding standard template parameters; acquiring a characteristic region image of the axle box body through a visual detection unit; extracting a reference surface contour line and a positioning hole circle center coordinate by adopting edge detection and Hough circle transformation; matching the extracted features with a standard template to calculate translation deviation and rotation deviation; the deviation is converted into an executing mechanism compensation correction amount through coordinate transformation; according to the method, the problems of secondary positioning accumulative errors, tool tolerance superposition, poor model switching adaptability and the like in multi-station circulation of the axle box body are solved, the deviation correction precision is superior to + / -0.2 mm and + / -0.1 degree, and the time consumed for single deviation correction is less than 3 seconds.

Description

Technical Field

[0001] This invention relates to the field of industrial vision inspection and automation control technology, specifically to an automatic deviation correction control method and system for the shaft box cleaning and grinding station. Background Technology

[0002] The axle box is a key component of the bogie in rail transit vehicles, and its machining accuracy directly affects the safety and stability of vehicle operation. Cleaning and grinding are two important processes in the production of axle boxes. Cleaning removes oil, chips, and scale from the surface of the raw or semi-finished product, while grinding performs precision machining on the key surfaces of the axle box to meet assembly accuracy requirements. Modern rail transit vehicle manufacturers typically use automated production lines to achieve multi-station processing of axle boxes, with workpieces transferred between multiple processing stations such as cleaning and grinding stations via tooling pallets.

[0003] In the prior art, Chinese invention patent application CN116071348A discloses a visual inspection method and related device for workpiece surface inspection. This method detects the workpiece grinding quality by constructing separate side grinding inspection models and external curved surface grinding inspection models, and adjusts the parameters of the workpiece grinding equipment based on the inspection results. This prior art mainly focuses on the visual inspection of grinding quality and the adjustment of equipment parameters, using a deep learning model to identify and locate grinding defects. However, this prior art has the following shortcomings: First, this method focuses on quality inspection after grinding and does not address the positioning correction problem during the workpiece's transfer between workstations, failing to solve the cumulative error caused by secondary positioning; second, this method uses a single workpiece template image for feature point matching, without considering the feature differences between different workpiece models, requiring manual template switching when the production line needs to produce multiple workpiece models; third, the detection results of this method are used to adjust the force parameters of the grinding equipment, failing to form a closed-loop control with the workstation positioning system, and thus cannot achieve active posture correction before processing.

[0004] In actual production, the following technical problems arise when axle housings are transferred between multiple workstations. First, there is the problem of cumulative errors in secondary positioning. When a workpiece is transferred from one workstation to the next, due to the limited repeatability of the transfer mechanism and the clearance between the tooling and the workpiece, each transfer introduces new positioning errors. These errors accumulate gradually during the multi-station transfer process, eventually leading to a significant deviation of the machining trajectory relative to the actual position of the workpiece. Second, there is the problem of cumulative tolerances in tooling fixtures. Different batches of tooling trays have manufacturing tolerances, and even tooling trays from the same batch will experience wear and deformation after long-term use. The cumulative effect of these tolerances leads to differences in the workpiece positioning datum on different toolings, resulting in uneven grinding and other quality issues. Third, there is the problem of poor adaptability of traditional mechanical positioning methods. Existing mechanical limit positioning methods are designed for specific workpiece models. When the production line needs to switch to producing different models of axle housings, manual replacement of positioning tooling or adjustment of the limit mechanism is required, resulting in low changeover efficiency and susceptibility to human error.

[0005] Therefore, there is an urgent need for a method that can automatically detect workpiece position deviation at the station entrance, calculate compensation amount in real time, and drive the actuator to complete the correction action, so as to solve the above-mentioned technical problems and improve the positioning accuracy and production efficiency of multi-station machining of shaft box. Summary of the Invention

[0006] The purpose of this invention is to provide an automatic correction control method and system for the cleaning and grinding station of axle box, so as to solve the technical problems of secondary positioning cumulative error, tooling and fixture tolerance superposition, and poor adaptability of traditional mechanical positioning methods during the transfer of axle box between cleaning and grinding stations.

[0007] To achieve the above objectives, this invention provides an automatic deviation correction control method for a shaft box cleaning and grinding station, comprising: a model identification step, wherein an RFID reader deployed at the station entrance reads RFID tag data on the tooling tray, determines the workpiece model information of the current shaft box based on the RFID tag data, and loads standard template parameters corresponding to the workpiece model information from a multi-model template library; an image acquisition step, wherein a vision detection unit deployed at the cleaning station entrance or grinding station entrance vertically captures images of the feature areas of the shaft box on the tooling, thereby acquiring feature area images; a feature extraction step, wherein edge detection processing is performed on the feature area images to extract the reference surface contour lines of the shaft box, and Hough circle transformation processing is performed on the feature area images to identify the positioning holes of the shaft box and obtain the center coordinates of the positioning holes; and deviation calculation. The process involves several steps: First, matching and calculating the coordinates of the reference plane contour line and the center of the positioning hole with the standard template parameters to determine the X-direction translational deviation, Y-direction translational deviation, and Z-axis rotational deviation of the current axis box relative to the standard pose. Second, in the coordinate transformation step, based on the transformation relationship between the coordinate system of the current workstation and the coordinate system of the servo platform or the robot base coordinate system, converting the translational and rotational deviations into compensation correction amounts for the servo platform or robot end effector. Third, in the closed-loop correction step, driving the servo platform or robot end effector to perform correction actions based on the compensation correction amounts, collecting real-time position feedback data of the actuator, calculating the residual deviation between the current pose and the target pose, and feeding back the residual deviation to the coordinate transformation step for iterative correction until the residual deviation is less than or equal to the preset correction accuracy threshold.

[0008] Preferably, the preset correction accuracy threshold includes a position accuracy threshold and an angle accuracy threshold. The position accuracy threshold ranges from 0.1 mm to 0.3 mm, and the angle accuracy threshold ranges from 0.05° to 0.15°.

[0009] Preferably, the vision inspection unit includes an industrial camera and a ring light source, wherein the industrial camera has a resolution of not less than 5 million pixels and the ring light source has a color temperature range of 5500K to 6500K.

[0010] To achieve the above objectives, the present invention also provides an automatic deviation correction control system for a shaft box cleaning and grinding station, comprising: a model identification unit, used to read RFID tag data on the tooling tray through an RFID reader / writer deployed at the station entrance, determine the workpiece model information of the current shaft box based on the RFID tag data, and load standard template parameters corresponding to the workpiece model information from a multi-model template library; a visual inspection unit, used to take a vertical downward photograph of the feature area of ​​the shaft box on the tooling at the cleaning station entrance or grinding station entrance, and acquire the feature area image; and a feature extraction unit, used to perform edge detection processing on the feature area image to extract the reference surface contour line of the shaft box, and perform Hough circle transform on the feature area image. The system identifies the positioning holes of the shaft box and obtains the coordinates of the center of the positioning holes; the template matching unit is used to match and calculate the coordinates of the reference surface contour line and the center of the positioning holes with the standard template parameters to determine the translational and rotational deviations of the current shaft box relative to the standard pose; the coordinate transformation unit is used to convert the translational and rotational deviations into compensation correction amounts according to the transformation relationship between the coordinate system of the current workstation and the coordinate system of the servo platform or the robot base coordinate system; the closed-loop control unit is used to drive the actuator to perform correction actions according to the compensation correction amounts, collect the real-time position feedback data of the actuator, calculate the residual deviation, and when the residual deviation is greater than the preset correction accuracy threshold, feed the residual deviation back to the coordinate transformation unit for iterative correction.

[0011] The beneficial effects of this invention are as follows:

[0012] First, by deploying vision inspection units at the entrances of the cleaning and grinding stations, the workpiece's positional deviation can be detected in real time before it enters the processing area. Automatic correction is then achieved through closed-loop control, eliminating the impact of accumulated secondary positioning errors on processing accuracy. The correction accuracy is better than ±0.2mm and ±0.1°.

[0013] Secondly, by building a multi-model template library and using RFID to automatically identify the current workpiece model, the system can automatically load the corresponding template parameters when the workpiece arrives, enabling mixed-line production of multiple workpiece models without manual intervention, thus improving the flexibility and changeover efficiency of the production line.

[0014] Third, by forming a closed-loop control between the visual inspection results and the servo platform or robot actuator, a complete correction closed loop of detection, calculation, execution and feedback is realized. The time for a single correction is less than 3 seconds, ensuring the consistency of the processing benchmark and the stability of the processing quality at each station. Attached Figure Description

[0015] Figure 1 This is a flowchart of the automatic deviation correction control method for the axle box cleaning and grinding station provided in an embodiment of the present invention;

[0016] Figure 2This is an architecture diagram of the automatic deviation correction control system for the shaft box cleaning and grinding station provided in an embodiment of the present invention. Detailed Implementation

[0017] Please refer to the attached document. Figures 1-2 The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] See Figure 1 This invention provides an automatic deviation correction control method for axle box cleaning and grinding stations. This method is applied to an automated production line for axle boxes in rail transit vehicles to solve the positioning deviation problem that occurs when the axle box moves between the cleaning and grinding stations. In this embodiment, the production line includes cleaning and grinding stations arranged sequentially. The axle box is carried by a tooling pallet and transferred between the stations by a conveying mechanism. The steps of this method are described in detail below.

[0019] Step S1: Model Identification Step

[0020] In one embodiment of the present invention, the model identification step reads the RFID tag data on the tooling tray by an RFID reader deployed at the workstation entrance, determines the workpiece model information of the current shaft box based on the RFID tag data, and loads the standard template parameters corresponding to the workpiece model information from a multi-model template library.

[0021] Specifically, when the tooling pallet carrying the shaft housing arrives at the entrance of the cleaning or grinding station, the RFID reader installed at the station entrance automatically senses and reads the RFID electronic tag fixed to the side of the tooling pallet. This RFID electronic tag stores information such as the workpiece model code, batch number, and tooling pallet number for the current shaft housing. In this embodiment, the RFID reader uses ultra-high frequency RFID technology, operating at a frequency of 920MHz to 925MHz, with a reading distance of 500mm to 1000mm. It can complete non-contact data reading during the movement of the tooling pallet, with a reading time of less than 100ms.

[0022] The system retrieves and loads the corresponding standard template parameters from a pre-established multi-model template library based on the read workpiece model code. The multi-model template library is a structured parameter database that stores the standard pose parameters of all producible axle box models on the production line. In a preferred embodiment of the invention, the standard template parameters include a standard coordinate sequence of the datum plane contour line, standard center coordinates of the positioning holes, standard radii of the positioning holes, and a matrix of relative positional relationships of feature points. The standard coordinate sequence of the datum plane contour line records the pixel coordinates of the axle box's datum plane boundary in the standard pose, stored as a discrete point sequence with a point spacing of no more than 2 pixels. The standard center coordinates of the positioning holes record the pixel coordinates of the center of each positioning hole in the standard pose; typically, an axle box has 2 to 4 positioning holes. The standard radius of the positioning holes records the theoretical radius value of each positioning hole, used for parameter constraints during subsequent Hough circle transformation. The matrix of relative positional relationships of feature points records the relative distances and angular relationships between feature points, used for matching verification and anomaly detection.

[0023] The core idea of ​​the model-adaptive template loading algorithm proposed in this invention is to establish a mapping relationship between RFID model codes and template parameter sets, thereby achieving automatic template switching upon workpiece arrival. Let the currently read RFID model code be... The template collection stored in the multi-model template library is Each template Includes model code and the corresponding parameter set The template loading process first searches the template library for templates that meet the requirements. template If the search is successful, the parameter set will be... The template is loaded into the system's working memory for use in subsequent steps. If the retrieval fails, an unknown model alarm is triggered. In this embodiment, the template library supports storing template parameters for no fewer than 50 different models of axle box bodies, and the template retrieval and loading time is less than 50ms.

[0024] By employing an RFID-driven dynamic loading mechanism for multiple template models, this invention solves the problem of manual adjustment required during model switching in traditional mechanical positioning methods, enabling automated production of multiple workpiece models on mixed lines. When the production planning and scheduling system arranges for the alternating production of different models of shaft boxes, this system can automatically identify and switch the corresponding positioning parameters without stopping the line for model changes, significantly improving production efficiency.

[0025] Step S2: Image Acquisition Step

[0026] In one embodiment of the present invention, the image acquisition step involves taking a vertical downward photograph of the feature area of ​​the tooling shaft housing by a vision detection unit deployed at the entrance of the cleaning station or the entrance of the grinding station, thereby obtaining an image of the feature area.

[0027] The hardware components of the vision inspection unit include an industrial camera, a lens, a ring light source, and a mounting bracket. In this embodiment, the industrial camera uses a CMOS sensor with a resolution of 2592×1944 pixels (approximately 5 megapixels), a frame rate of 15fps, and a pixel size of 2.2μm×2.2μm. The lens is a fixed-focus industrial lens with a focal length of 16mm, an aperture of F2.8, and a working distance of 400mm to 600mm. With this configuration, the actual physical size corresponding to a single pixel is approximately 0.1mm, which meets the inspection requirement of ±0.2mm correction accuracy. The ring light source uses a white LED array with a color temperature range of 5500K to 6500K. Uniform ring illumination can eliminate specular reflection interference from the workpiece surface and enhance the contrast of edge features. The mounting bracket adopts a gantry structure, fixing the camera and light source above the workstation entrance, ensuring that the camera's optical axis is perpendicular to the plane of the tooling tray.

[0028] When the conveyor transports the tooling pallet to the preset shooting position, the position sensor installed on the conveyor sends a trigger signal, and the vision detection unit starts image acquisition according to the trigger signal. The original image acquired by the camera is an 8-bit grayscale image with an image size of 2592×1944 pixels. In this embodiment, the feature region refers to a local area on the shaft box body that includes the reference surface boundary and the positioning hole. The position of this region in the image is pre-calibrated and determined according to the workpiece model and the camera installation position. Based on the feature region position information in the standard template parameters loaded in step S1, the system crops the region of interest image containing the feature region from the original image. The size of the cropped image is typically between 1024×1024 pixels and 1600×1200 pixels.

[0029] In a preferred embodiment of the present invention, to improve image quality and the stability of subsequent feature extraction, the image acquisition step further includes an image preprocessing process. The image preprocessing process first performs histogram equalization on the original image to enhance the overall contrast; then, bilateral filtering is used for noise reduction. Bilateral filtering can smooth noise while preserving edge information. The filter window size is set to 9×9 pixels, the spatial domain standard deviation is set to 75, and the grayscale domain standard deviation is set to 75; finally, grayscale normalization is performed on the filtered image, mapping the grayscale values ​​to a standard range of 0 to 255.

[0030] Step S3: Feature Extraction Step

[0031] In one embodiment of the present invention, the feature extraction step performs edge detection processing on the feature region image to extract the reference surface contour line of the axle box body, and performs Hough circle transformation processing on the feature region image to identify the positioning hole of the axle box body and obtain the center coordinates of the positioning hole.

[0032] The specific process for extracting the baseline contour is as follows. First, the Canny edge detection algorithm is used to extract edges from the preprocessed feature region image. The Canny edge detection algorithm includes four steps: Gaussian filtering, gradient calculation, non-maximum suppression, and double thresholding. In this embodiment, the kernel size of the Gaussian filter is set to 5×5, and the standard deviation is set to 1.4; the gradient calculation uses the Sobel operator to calculate the gradient components in the x and y directions respectively; non-maximum suppression refines the edges along the gradient direction; the high threshold of the double thresholding is set to 100, and the low threshold is set to 50, with a ratio of approximately 2:1. After Canny edge detection, the output is a binary edge image, with edge pixel values ​​of 255 and non-edge pixel values ​​of 0.

[0033] Then, straight line fitting is performed on the edge pixels belonging to the reference plane boundary region in the binary edge image. The reference plane boundary region is determined based on the preset region of interest range in the standard template parameters loaded in step S1. Within this region, the least squares method is used to perform straight line fitting on the edge pixels to obtain the straight line equation parameters of the reference plane contour. Let the set of edge pixels be... The equation of the line is The goal of the least squares method is to minimize the sum of squared fitting errors:

[0034] ,

[0035] in: This is the sum of squared fitting errors; This represents the total number of edge pixels. For the first The coordinates of the edge pixels; The slope of the line; It is the line intercept.

[0036] Through the To each and By taking the partial derivative and setting it to zero, we can obtain the optimal slope and intercept.

[0037] ,

[0038] ,

[0039] When the reference surface of the axle box contains multiple boundary lines, a straight line fitting is performed on each boundary line to obtain the set of straight line equation parameters for the reference surface contour.

[0040] The Hough circle transform algorithm is used to extract the center coordinates of the positioning holes. The Hough circle transform is a classic algorithm for detecting circular targets in images. Its basic principle is to map a circle in image space to a point in parameter space, and then determine the circle's parameters by finding accumulated peak values ​​in the parameter space. For circles in an image... Its center coordinates are , radius is In this embodiment, the gradient-based Hough circle transform method is adopted. This method first uses the gradient direction information of the edge points to determine the possible center positions of the circle, then calculates the cumulative radius value at the candidate center, and finally determines the parameters of the circle by threshold judgment.

[0041] The parameters for the Hough circle transform are set as follows: the minimum center distance is set to 1 / 8 of the image's shorter side size to avoid repeated detection of the same positioning hole; the edge detection threshold is set to 100; the center accumulation threshold is set to 50; the minimum radius is set to 0.8 times the standard radius of the positioning hole in the standard template parameters; and the maximum radius is set to 1.2 times the standard radius. By constraining the radius range, interference circles that are not positioning holes can be effectively filtered out.

[0042] After Hough circle transform, the output is a list of parameters for the detected circles, including the center coordinates for each circle. and radius In this embodiment, a typical axle box has 2 to 4 positioning holes. The system performs preliminary verification against a standard template based on the number and position distribution of the detected circles, eliminating false detections and missed detections.

[0043] Step S4: Deviation Calculation Step

[0044] In one embodiment of the present invention, the deviation calculation step matches and calculates the coordinates of the reference surface contour line and the center of the positioning hole with the standard template parameters to determine the X-direction translational deviation, Y-direction translational deviation and Z-axis rotational deviation of the current shaft box relative to the standard pose.

[0045] The core idea of ​​the multi-feature fusion deviation calculation algorithm proposed in this invention is to comprehensively utilize the features of the reference surface contour line and the positioning hole features, and calculate the workpiece's pose deviation through a two-stage coarse and fine matching strategy. Compared with methods that use only a single feature, multi-feature fusion can improve the robustness and accuracy of matching.

[0046] The first stage of deviation calculation is a coarse estimate of the rotational deviation based on the reference plane profile. Let the slope of the currently detected reference plane profile be... The slope of the reference plane contour line in the standard template is The initial rotational deviation estimate is:

[0047] ,

[0048] in: This is the initial rotational deviation estimate, in radians; The slope of the currently detected reference plane profile; The slope of the reference plane outline in the standard template; It is the arctangent function.

[0049] The second stage of deviation calculation is the translational deviation calculation based on the coordinates of the positioning hole center. First, the coordinates of the currently detected positioning hole center are rotated and compensated according to the initial rotational deviation estimate, transforming them to the same orientation as the standard template. Let the currently detected... The coordinates of the center of each positioning hole are: The coordinates after rotation compensation are:

[0050] ,

[0051] ,

[0052] in: For the rotationally compensated first Coordinates of the center of each positioning hole; The coordinates of the rotation center are usually taken from the center of the image or the reference origin defined in the standard template.

[0053] Then, calculate the positional difference between the compensated coordinates of the positioning hole center and the coordinates of the standard positioning hole center. Let the first... The coordinates of the center of each positioning hole are: Then the first The translational deviation corresponding to each positioning hole is:

[0054] ,

[0055] ,

[0056] For those with For the shaft box body with each positioning hole, the overall translational deviation is calculated using a weighted average:

[0057] ,

[0058] ,

[0059] in: and These represent the combined translational deviations in the X and Y directions, respectively. For the first The weighting coefficient of each positioning hole is usually determined based on the detection confidence level of the positioning hole or its deviation from the standard radius; This represents the total number of positioning holes.

[0060] The third stage of the deviation calculation is the fine-tuning of the rotation deviation. Based on the matching residuals of multiple feature points, an iterative nearest-point algorithm is used to fine-tune the rotation deviation. Let the current estimated rotation deviation value be... The estimated translation deviation value is The feature point set includes sampling points on the center of the positioning hole and the contour line of the reference surface. The goal of iterative optimization is to minimize the sum of squared distances between the transformed current feature points and the feature points of the standard template.

[0061] ,

[0062] in: For the matching error function; This represents the total number of feature points; It is a homogeneous transformation function that includes rotation and translation; The currently detected number Coordinates of one feature point; For the first in the standard template Coordinates of one feature point; Let be the Euclidean distance norm.

[0063] The objective function is optimized using gradient descent or the Gauss-Newton method, and iteratively updated. , and The iteration continues until convergence or the maximum number of iterations is reached. In this embodiment, the iteration termination condition is set to the parameter change between two adjacent iterations being less than [a certain value]. Or the number of iterations exceeds 20.

[0064] The final output deviation includes the translational deviation in the X direction. Y-direction translational deviation and rotational deviation around the Z-axis These three deviations describe the pose deviation of the current axle box relative to the standard pose within the tooling plane. It should be noted that, since the camera is mounted vertically downwards and the axle box is placed on a horizontal tooling tray, this method only calculates the three degrees of freedom deviations (X translation, Y translation, Z rotation) in the plane, and does not involve the calculation of deviations in the height direction and tilt angle.

[0065] Step S5: Coordinate Transformation Step

[0066] In one embodiment of the present invention, the coordinate transformation step converts the X-direction translational deviation, Y-direction translational deviation, and rotational deviation around the Z-axis into compensation correction amounts for the servo platform or robot end effector based on the transformation relationship between the coordinate system of the current workstation and the coordinate system of the servo platform or the robot base coordinate system.

[0067] In multi-station automated production lines, the visual coordinate system of the vision inspection unit is usually not coincident with the coordinate system of the servo platform or the robot base coordinate system that performs the correction action. Coordinate transformation is required to convert the deviation obtained by vision inspection into compensation instructions that can be executed by the actuator.

[0068] The adaptive coordinate transformation algorithm proposed in this invention employs a combination of pre-calibration and online compensation. The pre-calibration stage is completed during system installation and debugging, using a calibration board to measure and determine the homogeneous transformation matrix from the visual coordinate system to the servo platform coordinate system (or the robot's base coordinate system). Let the visual coordinate system be... The servo platform coordinate system is The homogeneous transformation matrix between the two is:

[0069] ,

[0070] in: This is the homogeneous transformation matrix from the visual coordinate system to the servo platform coordinate system; This represents the rotation angle of the visual coordinate system relative to the servo platform coordinate system. and These are the translation components in the X and Y directions, respectively.

[0071] The deviation vector in the visual coordinate system is ,in and The translational deviation calculated in step S4 is represented in the visual coordinate system. Multiplying the deviation vector by the homogeneous transformation matrix on the left yields the deviation vector in the servo platform coordinate system:

[0072] ,

[0073] The calculation yields:

[0074] ,

[0075] ,

[0076] in: and These represent the compensation correction amounts in the X and Y directions, respectively, in the servo platform coordinate system.

[0077] Regarding rotational deviation, since the Z-axis direction is the same in both the visual coordinate system and the servo platform coordinate system (both perpendicular to the tooling plane and upwards), the rotational deviation around the Z-axis has the same value in both coordinate systems, that is:

[0078] ,

[0079] in: This is the rotation compensation correction amount in the servo platform coordinate system; The rotational deviation around the Z-axis is calculated in step S4.

[0080] The online compensation phase is used to compensate for coordinate system transformation parameter drift caused by factors such as temperature changes and mechanical wear. In a preferred embodiment of the invention, the system periodically performs online calibration using standard calibration blocks, updating the parameters of the homogeneous transformation matrix by comparing the deviation between the visually detected values ​​and known standard values. The online calibration cycle can be set to once per shift or once per week, depending on the stability of the production environment.

[0081] The output of the coordinate transformation step is the compensation correction amount in the servo platform coordinate system (or robot base coordinate system). This compensation correction is directly used to drive the actuator to perform corrective actions.

[0082] Step S6: Closed-loop correction steps

[0083] In one embodiment of the present invention, the closed-loop correction step drives the servo platform or robot end effector to perform correction actions according to the compensation correction amount, collects real-time position feedback data of the actuator, calculates the residual deviation between the current pose and the target pose, and when the residual deviation is greater than the preset correction accuracy threshold, the residual deviation is fed back to the coordinate transformation step for iterative correction until the residual deviation is less than or equal to the preset correction accuracy threshold.

[0084] The closed-loop iterative correction control algorithm proposed in this invention adopts an incremental correction strategy and an iterative convergence mechanism based on residual deviation feedback to ensure that the correction accuracy meets the preset requirements while avoiding overshoot and oscillation.

[0085] Incremental correction strategy refers to a strategy where the amount of correction performed in each step is not directly calculated using the compensation adjustment amount, but rather the product of the compensation adjustment amount and a preset proportional coefficient. Let the first step be... The compensation correction amount calculated in the next iteration is: The actual amount of correction implemented is:

[0086] ,

[0087] in: For the first The actual amount of correction performed in each iteration; The preset scaling factor ranges from 0.6 to 0.9. In this embodiment, the preset scaling factor is set to 0.8. Using a scaling factor less than 1 can avoid overshoot caused by system errors or detection noise, thus improving the stability of the correction process.

[0088] The actuator drives the tooling pallet or workpiece to adjust its position and orientation according to the correction command. In this embodiment, the actuator uses a three-axis servo platform, including an X-axis linear module, a Y-axis linear module, and a θ-axis rotary table. The X-axis and Y-axis linear modules have a travel of ±50mm, a positioning accuracy of ±0.01mm, and a maximum moving speed of 100mm / s. The θ-axis rotary table has a rotation range of ±5°, a positioning accuracy of ±0.005°, and a maximum rotation speed of 10° / s. All three axes of the servo platform are equipped with high-precision linear scales or encoders for real-time feedback of the current position.

[0089] After the correction action is completed, the system collects real-time position feedback data from the actuator and calculates the residual deviation between the current pose and the target pose. The residual deviation can be calculated in two ways: one is based on encoder feedback from the actuator, using the difference between the correction command and the actual execution amount as the residual deviation; the other is based on secondary visual detection, where images are re-acquired after the correction action, feature extraction and deviation calculation are performed, and the newly calculated deviation value is used as the residual deviation. In a preferred embodiment of the invention, the second method based on secondary visual detection is used. This method can detect error sources other than the actuator's motion error, resulting in higher correction accuracy.

[0090] Let the first The residual error calculated after the second iteration is: The preset correction accuracy threshold includes the position accuracy threshold. and angle accuracy threshold The iteration termination condition is:

[0091] ,

[0092] When the above conditions are met, the correction process ends, the system outputs a correction completion flag, and the workpiece enters the subsequent processing steps. When the above conditions are not met, the residual deviation is fed back to step S5 for coordinate transformation to obtain a new compensation correction amount, and the next iteration of correction is performed.

[0093] In this embodiment, the position accuracy threshold is set to 0.2 mm, and the angle accuracy threshold is set to 0.1°. Under this accuracy requirement, the typical time for a single correction (including image acquisition, feature extraction, deviation calculation, coordinate transformation, and action execution) is 1.5 to 2 seconds. In most cases, the accuracy requirement can be met after 1 to 2 iterations, and the total correction time is less than 3 seconds.

[0094] To prevent infinite iterations under abnormal conditions, the closed-loop correction step also has a maximum iteration limit. When the number of correction iterations exceeds the preset maximum number of iterations (set to 5 in this embodiment) and the residual deviation is still greater than the preset correction accuracy threshold, the system determines that the correction is abnormal, generates a correction abnormality alarm signal, and suspends the processing flow of the current station, awaiting manual intervention. Abnormal situations may be caused by severe workpiece deformation, tooling damage, or vision system failure.

[0095] Through the above six steps, the present invention realizes automatic correction control of the shaft box body during the transfer process between cleaning and grinding stations, effectively eliminating the influence of cumulative errors in secondary positioning and the superposition of tooling and fixture tolerances, and ensuring the consistency of processing benchmarks at each station.

[0096] In another embodiment of the present invention, the application effect of the method of the present invention on an actual production line was tested and verified. The test was conducted on an axle box cleaning and grinding production line of a rail transit vehicle manufacturing company. This production line has one cleaning station and one grinding station, producing three types of axle boxes. The test sample consisted of 100 axle boxes of each type, totaling 300 pieces. The test indicators included the success rate of correction, the average correction time, the position correction accuracy, and the angle correction accuracy.

[0097] Test results show that under mixed-line production conditions for three types of shaft box bodies, the correction success rate of the method of this invention is 99.7% (only one piece failed correction due to severe surface corrosion leading to feature extraction failure), with an average correction time of 2.3 seconds. Regarding position correction accuracy, the mean value of the translational deviation in the X direction is 0.08 mm, and the standard deviation is 0.04 mm; the mean value of the translational deviation in the Y direction is 0.09 mm, and the standard deviation is 0.05 mm. Regarding angular correction accuracy, the mean value of the rotational deviation around the Z-axis is 0.05°, and the standard deviation is 0.02°. All corrected workpiece pose deviations are less than the preset correction accuracy thresholds (position ±0.2 mm, angle ±0.1°), meeting the positioning accuracy requirements of subsequent processing steps.

[0098] Compared with traditional mechanical positioning methods, the method of the present invention eliminates the need for manual adjustment of positioning fixtures during model switching, reducing the model changeover time from approximately 30 minutes to less than 1 second with automatic template switching by the system. In terms of secondary positioning accuracy, the cumulative error of secondary positioning in traditional methods can reach ±1mm to ±2mm, while the method of the present invention reduces this error to within ±0.2mm, which is an improvement of approximately 10 times.

[0099] See Figure 2 The present invention also provides an automatic deviation correction control system for the axle box cleaning and grinding station, which is used to execute the automatic deviation correction control method for the axle box cleaning and grinding station described in the above method embodiments.

[0100] The automatic deviation correction control system for the shaft box cleaning and grinding station includes a model identification unit 1, a vision inspection unit 2, a feature extraction unit 3, a template matching unit 4, a coordinate transformation unit 5, and a closed-loop control unit 6. The functions and connections of each unit are explained below.

[0101] The model identification unit 1 reads RFID tag data from the tooling tray using an RFID reader deployed at the workstation entrance, determines the workpiece model information of the current shaft box based on the RFID tag data, and loads the standard template parameters corresponding to the workpiece model information from a multi-model template library. The hardware components of the model identification unit 1 include an RFID reader and an antenna. In this embodiment, the RFID reader uses ultra-high frequency RFID technology, operating at a frequency of 920MHz to 925MHz, and supports the ISO18000-6C protocol. The output of the model identification unit 1 is the workpiece model information and the corresponding standard template parameters, which is sent to the template matching unit 4.

[0102] The vision inspection unit 2 is used to capture images of the feature areas of the shaft housing on the tooling from a vertical downward angle at the entrance of the cleaning or grinding station, thereby acquiring feature area images. As described in the aforementioned method embodiments, the hardware components of the vision inspection unit 2 include an industrial camera, a lens, a ring light source, and a mounting bracket. The output of the vision inspection unit 2 is a feature area image, which is sent to the feature extraction unit 3.

[0103] Feature extraction unit 3 performs edge detection processing on the feature region image to extract the reference surface contour line of the axle box body, and performs Hough circle transform processing on the feature region image to identify the positioning holes of the axle box body and obtain the center coordinates of the positioning holes. Feature extraction unit 3 can be implemented using an industrial computer or an embedded image processor, and has built-in edge detection algorithm module and Hough circle transform algorithm module. The input of feature extraction unit 3 is the feature region image from vision detection unit 2, and the output is the reference surface contour line parameters and the center coordinates of the positioning holes. This output is sent to template matching unit 4.

[0104] Template matching unit 4 is used to match and calculate the coordinates of the reference surface contour line and the center of the positioning hole with the standard template parameters to determine the X-direction translational deviation, Y-direction translational deviation, and Z-axis rotational deviation of the current shaft box relative to the standard pose. As described in the aforementioned method embodiment, template matching unit 4 uses a multi-feature fusion deviation calculation algorithm to achieve accurate deviation calculation. The input of template matching unit 4 includes feature parameters from feature extraction unit 3 and standard template parameters from model recognition unit 1, and the output is translational deviation and rotational deviation, which is sent to coordinate transformation unit 5.

[0105] The coordinate transformation unit 5 is used to convert translational and rotational deviations into compensation correction amounts based on the transformation relationship between the coordinate system of the current workstation and the coordinate system of the servo platform or the robot base coordinate system. As described in the aforementioned method embodiments, the coordinate transformation unit 5 has pre-calibrated homogeneous transformation matrix parameters and supports online calibration updates. The input of the coordinate transformation unit 5 is the deviation amount from the template matching unit 4, and the output is the compensation correction amount, which is sent to the closed-loop control unit 6. Simultaneously, the coordinate transformation unit 5 also receives residual deviations fed back from the closed-loop control unit 6 for coordinate transformation during iterative correction.

[0106] The closed-loop control unit 6 drives the actuator to perform correction actions based on the compensation correction amount, collects real-time position feedback data of the actuator, calculates the residual deviation, and when the residual deviation exceeds a preset correction accuracy threshold, feeds the residual deviation back to the coordinate transformation unit 5 for iterative correction. The hardware components of the closed-loop control unit 6 include a motion controller, a servo driver, and an actuator. In this embodiment, the motion controller uses a programmable logic controller or a dedicated motion control card, supporting multi-axis linkage control; the actuator uses a three-axis servo platform, including an X-axis linear module, a Y-axis linear module, and a θ-axis rotary table. The closed-loop control unit 6 also establishes a data communication connection with the vision detection unit 2 to trigger secondary vision detection after the correction action is completed to calculate the residual deviation.

[0107] The data flow and control flow relationships between the various units of the system of this invention are as follows: After the tooling pallet arrives at the workstation entrance, the model identification unit 1 first reads the RFID tag data and loads the standard template parameters; then, the vision inspection unit 2 acquires the image of the feature area and sends it to the feature extraction unit 3; the feature extraction unit 3 extracts the feature parameters and sends them to the template matching unit 4; the template matching unit 4 calculates the deviation based on the standard template parameters and sends it to the coordinate transformation unit 5; the coordinate transformation unit 5 converts the deviation into a compensation correction amount and sends it to the closed-loop control unit 6; the closed-loop control unit 6 drives the actuator to complete the correction action and determines whether iterative correction is needed. If iterative correction is needed, the closed-loop control unit 6 triggers the vision inspection unit 2 to reacquire the image and feeds back the residual deviation to the coordinate transformation unit 5, forming closed-loop control.

[0108] In a preferred embodiment of the present invention, the system further includes a system controller and a human-machine interface. The system controller is used to coordinate the working sequence of each unit, manage a multi-model template library, record correction logs, and handle abnormal situations. The human-machine interface is used to display the system operating status, real-time correction data, and historical statistical information, and supports operators to configure parameters and perform manual intervention operations.

[0109] This invention's system can be deployed at the entrance of each workstation requiring precise positioning on the shaft box production line. Multiple workstation correction systems can operate independently or be centrally managed via a host computer or industrial Ethernet. Through this system, the shaft box can automatically correct its alignment before entering the cleaning or grinding workstation, ensuring consistent processing standards at each workstation, thereby improving overall processing quality and production efficiency.

[0110] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. 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 modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. An automatic deviation correction control method for the cleaning and grinding station of the shaft box, characterized in that, include: The model identification step involves reading the RFID tag data on the tooling tray using an RFID reader deployed at the workstation entrance, determining the workpiece model information of the current shaft box based on the RFID tag data, and loading the standard template parameters corresponding to the workpiece model information from a multi-model template library. In the image acquisition step, a vision inspection unit deployed at the entrance of the cleaning station or the entrance of the grinding station takes a vertical downward image of the feature area of ​​the upper shaft box of the tooling to obtain the feature area image; The feature extraction step involves performing edge detection processing on the feature region image to extract the reference surface contour line of the axle box body, and performing Hough circle transform processing on the feature region image to identify the positioning holes of the axle box body and obtain the center coordinates of the positioning holes. The deviation calculation step involves matching the coordinates of the reference surface contour line and the center of the positioning hole with the standard template parameters to determine the X-direction translational deviation, Y-direction translational deviation, and rotational deviation around the Z-axis of the current axle box relative to the standard pose. The coordinate transformation step involves converting the X-direction translation deviation, the Y-direction translation deviation, and the rotation deviation around the Z-axis into compensation correction amounts for the servo platform or robot end effector, based on the transformation relationship between the coordinate system of the current workstation and the coordinate system of the servo platform or robot base coordinate system. The closed-loop correction step involves driving the servo platform or robot end effector to perform correction actions based on the compensation correction amount, collecting real-time position feedback data of the actuator, calculating the residual deviation between the current pose and the target pose, and feeding back the residual deviation to the coordinate transformation step for iterative correction when the residual deviation is greater than the preset correction accuracy threshold, until the residual deviation is less than or equal to the preset correction accuracy threshold.

2. The automatic deviation correction control method for the shaft box cleaning and grinding station according to claim 1, characterized in that, The preset correction accuracy threshold includes a position accuracy threshold and an angle accuracy threshold. The position accuracy threshold ranges from 0.1 mm to 0.3 mm, and the angle accuracy threshold ranges from 0.05° to 0.15°.

3. The automatic deviation correction control method for the shaft box cleaning and grinding station according to claim 1, characterized in that, The visual inspection unit includes an industrial camera and a ring light source. The industrial camera has a resolution of no less than 5 million pixels, and the ring light source has a color temperature range of 5500K to 6500K.

4. The automatic deviation correction control method for the shaft box cleaning and grinding station according to claim 1, characterized in that, The standard template parameters include the standard coordinate sequence of the reference plane contour line, the standard center coordinates of the positioning hole, the standard radius of the positioning hole, and the matrix of relative positional relationships of feature points.

5. The automatic deviation correction control method for the shaft box cleaning and grinding station according to claim 1, characterized in that, In the feature extraction step, the edge detection processing of the feature region image includes: performing Gaussian filtering denoising processing on the feature region image, performing edge detection on the denoised image using the Canny operator to obtain an edge image, and performing line fitting on the edge pixels in the edge image to obtain the reference surface contour line.

6. The automatic deviation correction control method for the shaft box cleaning and grinding station according to claim 1, characterized in that, In the deviation calculation step, matching the reference surface contour line and the center coordinates of the positioning hole with the standard template parameters includes: calculating the angular deviation between the reference surface contour line and the standard reference surface contour line as an initial rotation deviation estimate; performing rotation compensation on the current center coordinates of the positioning hole based on the initial rotation deviation estimate; calculating the position difference between the compensated center coordinates of the positioning hole and the standard center coordinates of the positioning hole as a translation deviation; and performing fine correction on the rotation deviation based on the matching residuals of multiple feature points.

7. The automatic deviation correction control method for the shaft box cleaning and grinding station according to claim 1, characterized in that, In the coordinate transformation step, the coordinate transformation based on the transformation relationship between the coordinate system of the current workstation and the coordinate system of the servo platform includes: obtaining the homogeneous transformation matrix from the pre-calibrated visual coordinate system to the servo platform coordinate system, and multiplying the deviation vector in the visual coordinate system by the homogeneous transformation matrix to obtain the compensation correction amount in the servo platform coordinate system.

8. The automatic deviation correction control method for the shaft box cleaning and grinding station according to claim 1, characterized in that, In the closed-loop correction step, an incremental correction strategy is adopted. The amount of each correction action is the product of the current calculated compensation correction amount and the preset proportional coefficient. The preset proportional coefficient ranges from 0.6 to 0.

9.

9. The automatic deviation correction control method for the shaft box cleaning and grinding station according to claim 1, characterized in that, The closed-loop correction step further includes: when the number of correction iterations exceeds the preset maximum number of iterations and the residual deviation is still greater than the preset correction accuracy threshold, generating a correction anomaly alarm signal and pausing the processing flow of the current workstation.

10. An automatic deviation correction control system for the axle box cleaning and grinding station, used to implement the automatic deviation correction control method for the axle box cleaning and grinding station as described in any one of claims 1-9, characterized in that, include: The model identification unit is used to read the RFID tag data on the tooling tray through the RFID reader deployed at the workstation entrance, determine the workpiece model information of the current shaft box based on the RFID tag data, and load the standard template parameters corresponding to the workpiece model information from the multi-model template library. The visual inspection unit is used to take vertical downward images of the feature areas of the tooling shaft box at the entrance of the cleaning station or the entrance of the grinding station to obtain images of the feature areas. The feature extraction unit is used to perform edge detection processing on the feature region image to extract the reference surface contour line of the axle box body, and to perform Hough circle transformation processing on the feature region image to identify the positioning holes of the axle box body and obtain the center coordinates of the positioning holes. The template matching unit is used to match and calculate the coordinates of the reference surface contour line and the center of the positioning hole with the standard template parameters to determine the X-direction translational deviation, Y-direction translational deviation and Z-axis rotational deviation of the current shaft box relative to the standard pose. The coordinate transformation unit is used to convert the translational deviation and the rotational deviation into compensation correction amounts based on the transformation relationship between the coordinate system of the current workstation and the coordinate system of the servo platform or the robot base coordinate system. The closed-loop control unit is used to drive the actuator to perform correction actions according to the compensation correction amount, collect the real-time position feedback data of the actuator, calculate the residual deviation, and when the residual deviation is greater than the preset correction accuracy threshold, feed the residual deviation back to the coordinate transformation unit for iterative correction.

Images