Planar laser engraving machine positioning method

Abstract

The invention discloses a plane laser engraving machine positioning method, and relates to the technical field of laser engraving, the method comprises the following steps: establishing a combined mapping relation from a reference image coordinate to a drop point coordinate, generating a sampling point set covering a scanning field, and forming a verification data set; calculating an expected drop point coordinate based on a distortion compensation parameter and a visual calibration parameter, carrying out difference on the expected drop point coordinate and an actually measured drop point coordinate to obtain a residual sequence, carrying out partition summarization according to a spatial position, comparing with a precision threshold, and outputting a threshold exceeding region set; respectively applying controlled disturbance recalculation residual statistics to the two types of parameters to form a responsibility quantity, comparing the responsibility quantity with a responsibility threshold, outputting a deviation source label, iteratively updating according to the deviation source label, and packaging a combined parameter packet when stopping; in the operation stage, sampling inspection residual errors are calculated according to sampling inspection stepping and compared with a drift threshold value, and when the sampling inspection residual errors exceed the drift threshold value, instructions are paused and re-calculated; the combination effect of the two types of parameters can be verified in a unified mode, deviation gathering partition is positioned, recalculation attribution is completed, trial and error maintenance time is shortened, and the batch error etching risk is reduced.

Description

Technical Field

[0001] This application relates to the field of laser engraving technology, specifically to a positioning method for a planar laser engraving machine. Background Technology

[0002] Planar laser engraving or marking equipment typically uses a galvanometer scanning head and a field lens to achieve rapid scanning, and completes the engraving of text, QR codes or graphics on the workpiece plane. In order to ensure that the engraved content is positioned relative to the workpiece boundary, hole position or positioning mark, the equipment is usually equipped with a vision unit to acquire workpiece images and extract reference features. Based on the image coordinates of the reference features, a conversion relationship between image coordinates and landing point coordinates is established.

[0003] In engineering implementation, to compensate for the nonlinear error of the scanning field introduced by the galvanometer scanning and the field lens, the equipment is usually configured with distortion compensation parameters to correct the landing point coordinate conversion process; to compensate for the installation posture of the vision unit, lens distortion, and the relative relationship between the vision coordinates and the landing point coordinates, the equipment is usually configured with vision calibration parameters to convert the reference image coordinates into landing point coordinates; the two types of parameters are often generated by different debugging processes and maintained independently, and in field maintenance, it is common to see situations where only the distortion compensation parameters or only the vision calibration parameters are updated.

[0004] Existing solutions generally lack a mechanism for unified verification of the combined effect of two types of parameters within the scanning field. This leads to the accumulation of deviations in local areas of the scanning field after parameter updates, manifested as a decrease in deviation in some areas while an increase in deviation in others. Furthermore, there is a lack of a mechanism for recalculating the source of deviation between distortion compensation parameters and visual calibration parameters. On-site maintenance relies on repeated trial and error, increasing maintenance time and downtime costs, and there is a quality risk of batch mis-engraving caused by undetected parameter drift.

[0005] Therefore, proposing a positioning method for a planar laser engraving machine to solve the difficulties existing in the prior art is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] The purpose of this application is to provide a positioning method for a planar laser engraving machine to address the shortcomings in the prior art.

[0007] To achieve the above objectives, this application provides the following technical solution: A positioning method for a planar laser engraving machine includes: Step S1: Establish a combined mapping relationship between the reference image coordinates output by the vision unit and the landing point coordinates, and obtain a set of sampling points covering the scanning field; collect the reference image coordinates for each sampling point and perform point marking to obtain the measured landing point coordinates, forming a verification dataset bound to the combined mapping relationship; Step S2: Based on the distortion compensation parameters and visual calibration parameters, calculate the expected landing point coordinates from the validation dataset, and differ them with the measured landing point coordinates to obtain the residual sequence; Step S3: Based on the spatial location of the sampling point set, perform partitioning and summarizing on the residual sequence to obtain residual statistics. Compare the residual statistics with the precision threshold. Partitions where the residual statistics exceed the precision threshold are recorded as regions exceeding the threshold. Output the set of regions exceeding the threshold. Step S4: Apply controlled perturbations to the distortion compensation parameters and visual calibration parameters respectively and recalculate the residual statistics to form the responsibility quantity. Compare the responsibility quantity with the responsibility threshold and output the deviation source label. Step S5: Based on the bias source label, perform iterative updates on the distortion compensation parameters or visual calibration parameters. In each round of updates, recalculate S2 to S3 and compare the changes in the set of regions exceeding the threshold with the convergence threshold to determine whether to stop. When stopping, encapsulate the updated distortion compensation parameters and the updated visual calibration parameters into a combined parameter package. During runtime, calculate the sampling residual according to the sampling step and compare it with the drift threshold. When the sampling residual exceeds the drift threshold, generate a pause command and call the verification dataset for recalculation.

[0008] The technical effects and advantages of the planar laser engraving machine positioning method provided in this application are as follows: By establishing a combined mapping relationship between the reference image coordinates output by the vision unit and the landing point coordinates, a sampling point set covering the scanning field is constructed, forming a verification dataset bound to the combined mapping relationship. Based on this, the expected landing point coordinates are calculated based on the distortion compensation parameters and the vision calibration parameters, and the residual sequence is obtained by differencing it with the measured landing point coordinates. Then, the residual sequence is partitioned and summarized according to the spatial location of the sampling point set to obtain the residual statistics, which are compared with the accuracy threshold to output the set of regions exceeding the threshold. This achieves unified quantitative verification of the combined effect of the two types of parameters, enabling the deviation clusters in the center, edge, and corner of the scanning field to be located to specific partitions, avoiding misjudgments caused by relying solely on individual points or local observations, thereby reducing the number of repeated trial and error attempts and parameter tuning time on-site, and reducing the risk of increased deviation in local areas.

[0009] Furthermore, controlled perturbations are applied to the distortion compensation parameters and visual calibration parameters respectively, and the residual statistics are recalculated to form distortion liability and calibration liability. The liability difference is then compared with the liability threshold to output the deviation source label. Subsequently, the update object is selected based on the deviation source label, and the distortion compensation parameters or visual calibration parameters are iteratively updated. In each round of updates, the residual statistics are recalculated, and the stop is determined by comparing the change of the set of regions exceeding the threshold with the convergence threshold. In this way, deviation location is transformed from empirical judgment to recalculated attribution and parameter update path selection based on liability. This can quickly distinguish the main source of liability when the two types of parameters are updated separately or drift together, reduce new deviations introduced by invalid adjustments and misadjustments, improve the interpretability and consistency of the maintenance process, and increase the first-time success rate of parameter updates.

[0010] When stopped, the updated distortion compensation parameters and updated visual calibration parameters are encapsulated into a combined parameter package. During the operation phase, the sampling residual is calculated and compared with the drift threshold according to the sampling step. When the sampling residual exceeds the drift threshold, the instruction is paused and the verification dataset is called for recalculation. The set of regions exceeding the threshold, the label of the deviation source, and the update suggestion are output. This realizes drift detection and risk suppression during the operation phase, which can detect parameter drift and local deviation reproduction in time before or during batch processing, reducing the probability of batch misprinting. At the same time, the versioned record of the combined parameter package improves the traceability and reproducibility of parameter configuration, so that maintenance results can be quickly traced back and reused, comprehensively improving the long-term stability of equipment operation and production line availability. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0012] Figure 1 This is a schematic diagram of a planar laser engraving machine positioning method according to this application; Figure 2 This is a flowchart illustrating the iterative update method for distortion compensation parameters or visual calibration parameters described in this application. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] Please see Figure 1 As shown, this embodiment provides a positioning method for a planar laser engraving machine, including: Step S1: Establish a combined mapping relationship between the reference image coordinates output by the vision unit and the landing point coordinates, and obtain a set of sampling points covering the scanning field; collect the reference image coordinates of each sampling point and perform point marking to obtain the measured landing point coordinates, forming a verification dataset bound to the combined mapping relationship.

[0015] In this embodiment, the method steps for establishing a combined mapping relationship between the reference image coordinates output by the vision unit and the landing point coordinates include: A vision unit is defined as an imaging device used to acquire images and output pixel coordinates. Reference image coordinates are defined as pixel coordinates extracted by the vision unit from reference features. A calibration carrier containing reference features is arranged on the processing plane, and the known coordinates of the reference features in the landing point coordinates are recorded. The landing point coordinates are defined as two-dimensional position coordinates on the processing plane. Images of the calibration carrier are acquired, and multiple sets of reference image coordinates are extracted and paired with known coordinates to form a pairing point set. Least square fitting is performed on the pairing point set to obtain the set of transformation parameters from pixels to landing points, and a distortion correction term is introduced to compensate for the residuals of edge points to obtain a combined mapping relationship. The purpose of establishing the combined mapping relationship is to uniformly convert the measurement results of the vision unit into landing point coordinates usable on the processing plane, so as to facilitate the quantification and traceability of the landing point deviation across the entire scanning field with the same caliber.

[0016] In this embodiment, to avoid the conversion process used to generate the measured landing point coordinates from being coupled with the subsequent parameter update process to be attributed, the combined mapping relationship includes at least a measurement mapping sub-relationship used to convert the coordinates of the dot marker pixels into the measured landing point coordinates. During a verification cycle, the parameter version of the measurement mapping sub-relationship remains frozen and is bound to the version information of the verification dataset, and is not used as the object of iterative update of the distortion compensation parameter and the visual calibration parameter, so that the residual sequence reflects the parameter combination deviation rather than the homologous fitting error.

[0017] To facilitate understanding of the solution process for least squares fitting and distortion correction terms, the following example uses the same calibration carrier with sampling points at the center and corners of the scanning field: The vision unit uses an industrial camera, and the pixel coordinates output by the vision unit are used as the reference image coordinates. The calibration carrier is placed on the processing plane and has five reference features located at the four corners and the center of the scanning field, which are used to establish a correspondence with the corner points and the center point of the sampling point set. The landing point coordinates are based on the processing plane as the coordinate plane, and the unit is millimeters. The known coordinates of the four corner reference features are (0, 0), (200, 0), (0, 200), and (200, 200), and the known coordinates of the center reference feature are (100, 100). The vision unit acquires the image of the calibration carrier and extracts the pixel coordinates of the reference features. For example, the reference image coordinates of the four corner and center reference features are (320, 260), (1680, 255), (325, 1620), (1675, 1610), and (1000, 940), respectively. Each set of reference image coordinates is paired with the corresponding known coordinates to form a paired point set.

[0018] Least square fitting is performed on the paired point set to obtain the set of transformation parameters from pixel to landing point. For example, pixel coordinates are labeled (u, v) and landing point coordinates are labeled (x, y). A set of initial parameters is calculated using a transformation form containing translation and linear terms. Then, the four corner points of the scan field edge are used as edge points, and the edge point residuals are calculated by substituting the initial parameters. For example, the corner points show a directional residual on the order of 0.30 mm. A distortion correction term is constructed based on the edge point residuals and the radial distance of the pixels. The distortion correction term is expressed by a combination of quadratic and quartic terms of the radial distance. The distortion correction term and the transformation form are then incorporated into the least square fitting for another solution, so that the corner point residuals converge to the order of 0.05 mm. The set of transformation parameters obtained from the second fitting and the distortion correction term together constitute a combined mapping relationship. The combined mapping relationship is used to convert any reference image coordinates into landing point coordinates and correspond one-to-one with the spatial position of the sampling point set, providing a unified coordinate basis for the subsequent binding and storage of the validation dataset.

[0019] In this embodiment, the method steps for obtaining the sampling point set covering the scan field include: The scanning field boundary is determined, and the scanning field is defined as the effective processing area that the galvanometer allows to scan. A sampling layout is generated according to a layered strategy. The sampling layout includes central grid points, edge equidistant points, and corner densification points, and each sampling point is assigned a sampling point identifier. The sampling point identifier is bound to the nominal landing point coordinates to form a sampling point set. The nominal landing point coordinates are used to drive the point marking and undertake residual calculation. The purpose is to provide spatially uniform and boundary-covered verification input so that the combined mapping relationship has verifiable samples in each region of the scanning field.

[0020] To facilitate understanding of the process of generating the sampling layout and binding the nominal landing point coordinates using the layered strategy, the following example is provided.

[0021] The scanning field boundary is defined as a rectangular area on the processing plane based on the reach of the galvanometer. The coordinate range of the landing point is set to 0 to 200 mm horizontally and 0 to 200 mm vertically. The boundary is consistent with the known coordinates (0, 0), (200, 0), (0, 200), and (200, 200) of the four corner reference features of the aforementioned calibration carrier, so that the sampling point set and the combination mapping relationship are in the same landing point coordinate frame. When generating the sampling layout according to the layer strategy, the central grid points are arranged at equal intervals. For example, a combination of horizontal 50, 100, 150 mm and vertical 50, 100, 150 mm is taken to form nine central grid points, which are denoted as P1 to P9 respectively. The edge equidistant points are arranged at fixed intervals along the four boundaries. For example, one point is taken every 50 mm to obtain (0, 50), (0, 10... (0, 150), (200, 50), (200, 100), (200, 150), (50, 0), (100, 0), (150, 0), (50, 200), (100, 200), (150, 200) and assign sampling point identifiers P10 to P21 respectively; corner densification points add neighboring points centered on the four corners, for example, add (10, 0) and (0, 10) near (0, 0), add (190, 0) and (200, 10) near (200, 0), add (0, 190) and (10, 200) near (0, 200), add (190, 200) and (200, 190) near (200, 200) and assign sampling point identifiers P22 to P29 respectively.

[0022] Each sampling point identifier is bound to its corresponding nominal landing point coordinates. For example, sampling point identifier P5 is bound to nominal landing point coordinates (100, 100), sampling point identifier P22 is bound to nominal landing point coordinates (10, 0), and sampling point identifier P29 is bound to nominal landing point coordinates (200, 190), forming a sampling point set. The nominal landing point coordinates are used as input for galvanometer scanning commands in the subsequent dot-mapping stage and are paired with the measured landing point coordinates in the residual calculation stage to serve as the nominal reference for the combined residual sequence. This enables the coverage verification of the center, edge, and corner regions of the scanning field, reducing the risk of missing local area deviations.

[0023] In this embodiment, the method steps for acquiring reference image coordinates for each sampling point and performing point marking to obtain the measured landing point coordinates, forming a verification dataset bound to the combined mapping relationship, include: The nominal landing point coordinates of the sampling point set are read sequentially, and the galvanometer is driven to perform point marking to generate point markers. The vision unit acquires an image including the reference features and point markers, extracts the reference image coordinates and the pixel coordinates of the point markers. Based on the combined mapping relationship, the pixel coordinates of the point markers are converted into measured landing point coordinates. The sampling point identifiers, reference image coordinates, nominal landing point coordinates, measured landing point coordinates and combined mapping relationship version information are collected and stored to form a verification dataset bound to the combined mapping relationship. The purpose is to construct a recalculated set of verification samples for version difference pre-verification and consistency recalculation. The conversion of the measured landing point coordinates preferably uses the landing point coordinate results output by the measurement mapping sub-relation under the current frozen version. When the combined mapping relationship contains multiple sub-relations, at least the sub-relation used to generate the measured landing point coordinates is independent of the update objects in steps S4 to S5 to ensure that the responsibility quantity calculation has recalculation and interpretability.

[0024] A validation dataset is formed that is bound to the combination mapping relationship. A specific example is as follows: The scanning field boundary is defined as a rectangular area on the processing plane based on the reach of the galvanometer. The landing point coordinate range is set to 0 to 200 mm horizontally and 0 to 200 mm vertically. The four corners of the boundary are consistent with the known coordinates (0, 0), (200, 0), (0, 200), (200, 200) of the four corner reference features of the aforementioned calibration carrier. When generating the sampling layout according to the layered strategy, the central grid points are arranged at equal intervals, taking a combination of 50, 100, 150 mm horizontally and 50, 100, 150 mm vertically to form nine points. These points are assigned sampling point labels P1 to P9 and bound to nominal landing point coordinates (50, 50) to (150, 150) respectively. The edge equidistant points are taken along the four boundaries at 50 mm intervals to obtain (0, 50), (0, 100), (0, 150), (200, 50), (200, 100), (200, 150), (5 (0,0), (100,0), (150,0), (50,200), (100,200), (150,200), and sequentially assign sampling point identifiers P10 to P21 and complete the binding of sampling point identifiers with nominal landing point coordinates; for corner densification points, add nearest neighbor points near the four corners, set (10,0) and (0,10), (190,0) and (200,10), (0,190) and (10,200), (190,200) and (200,190), and sequentially assign sampling point identifiers P22 to P29 and bind nominal landing point coordinates; the sampling point set formed in this way simultaneously covers the central area, edge area and corner area, and the nominal landing point coordinates are directly used as galvanometer command input in subsequent point marking, and are paired with the measured landing point coordinates in residual calculation as nominal reference, so that the combined mapping relationship has verifiable samples in each area of ​​the scanning field.

[0025] Step S2: Based on the distortion compensation parameters and visual calibration parameters, the expected landing point coordinates are calculated from the validation dataset and then differ from the measured landing point coordinates to obtain the residual sequence.

[0026] In this embodiment, the method for calculating the expected landing point coordinates from the validation dataset based on distortion compensation parameters and visual calibration parameters includes: Distortion compensation parameters are defined as a set of parameters used to convert nominal landing point coordinates to compensated landing point coordinates. The method for obtaining distortion compensation parameters includes arranging a dot matrix target on the processing plane, performing point marking according to the nominal coordinates of the dot matrix and measuring the corresponding landing point coordinates, performing a fitting operation based on the paired point set of nominal coordinates and landing point coordinates to obtain the parameter set used for coordinate conversion, and recording version information. Visual calibration parameters are defined as a set of parameters used to convert reference image coordinates to landing point coordinates. The method for obtaining visual calibration parameters includes arranging a calibration carrier containing reference features, recording the known coordinates of the reference features, acquiring images to extract reference image coordinates, and converting the reference image coordinates based on the reference image coordinates and the known coordinates. The system performs fitting operations to obtain a parameter set and records version information. It then reads the validation dataset, which includes sampling point identifiers, nominal landing point coordinates, baseline image coordinates, measured landing point coordinates, and parameter version information. For each sampling point, the baseline image coordinates are converted to baseline landing point coordinates using visual calibration parameters, and the nominal landing point coordinates are converted to compensated landing point coordinates using distortion compensation parameters. The expected landing point coordinates are calculated based on the combination of baseline and compensated landing point coordinates and output in association with the sampling point identifier. The purpose is to output the predicted landing point corresponding to the parameter combination under the same landing point coordinate caliber, which serves as a basis for difference between the predicted and measured landing point coordinates.

[0027] Furthermore, distortion compensation parameters are used to pre-compensate the nominal landing point coordinates of the galvanometer scan to counteract the spatial nonlinearity introduced by the scanning system and the field lens; visual calibration parameters are used to convert the reference image coordinates into reference landing point coordinates to reflect the workpiece posture and clamping offset; the above two types of parameters are output under the same landing point coordinate caliber and participate in the calculation of the desired landing point coordinates, but their acquisition process, version management and iterative updates are distinguished to avoid duplicate compensation or mixed use of coordinate calibers.

[0028] To facilitate understanding the joint calculation process of distortion compensation parameters and visual calibration parameters on the validation dataset, a specific example is shown below: A dot matrix target is arranged on the processing plane. The nominal coordinates of the dot matrix are selected as (0, 0), (100, 0), (200, 0), (0, 100), (100, 100), (200, 100), (0, 200), (100, 200), (200, 200), and dots are marked sequentially. The visual unit converts each dot mark into landing point coordinates to obtain a set of landing point coordinates. For example, the nominal coordinate (200, 200) corresponds to the measured landing point coordinates (199.70, 200.25), and the nominal coordinate (0, 200) corresponds to the measured landing point coordinates (0.20, 199.80). The nominal coordinates and landing point coordinates are paired to form a set of points for fitting. The distortion compensation parameters are obtained and the version information is recorded. The distortion compensation parameters are used to convert the nominal landing point coordinates into compensated landing point coordinates.

[0029] Simultaneously, a calibration carrier containing reference features is arranged. The known coordinates of the reference features follow the aforementioned four corners and center, namely (0, 0), (200, 0), (0, 200), (200, 200), and (100, 100). Images of the calibration carrier are acquired and reference image coordinates are extracted. For example, the coordinates of the five reference images are (320, 260), (1680, 255), (325, 1620), (1675, 1610), and (1000, 940). A pairing point set is formed by the reference image coordinates and the known coordinates, and a fitting operation is performed to obtain visual calibration parameters and record version information. The visual calibration parameters are used to convert the reference image coordinates into landing point coordinates.

[0030] When reading the validation dataset, sampling point identifier P22 is selected as an example. The nominal landing point coordinates of P22 bound to the aforementioned sampling point set are (10, 0). The reference image coordinates recorded in the validation dataset are (990, 945), and the measured landing point coordinates recorded in the validation dataset are (9.92, 0.18), along with distortion compensation parameter version information and visual calibration parameter version information. The reference image coordinates (990, 945) are converted to reference landing point coordinates using the visual calibration parameters, for example, resulting in (10.05, 0.18). 2) The nominal landing point coordinates (10, 0) are converted into compensated landing point coordinates using distortion compensation parameters, for example, (9.97, 0.05); the expected landing point coordinates are calculated based on the combination relationship between the reference landing point coordinates and the compensated landing point coordinates, for example, (10.02, 0.10) is obtained by using the reference landing point coordinates as the position anchor and superimposing the offset of the compensated landing point coordinates; the expected landing point coordinates are associated with the sampling point identifier P22 and output, which is used to perform differential analysis with the measured landing point coordinates (9.92, 0.18) to form a residual sequence.

[0031] Specifically, the method for calculating the desired landing point coordinates based on the combined relationship between the baseline landing point coordinates and the compensated landing point coordinates includes the following steps: Read the nominal landing point coordinates, reference landing point coordinates, and compensated landing point coordinates corresponding to the sampling point identifiers in the verification dataset. The nominal landing point coordinates are the landing point coordinates determined by the sampling layout. The reference landing point coordinates are the landing point coordinates obtained by converting the reference image coordinates through visual calibration parameters. The compensated landing point coordinates are the landing point coordinates obtained by converting the nominal landing point coordinates through distortion compensation parameters. Calculate the compensation offset, which is defined as the difference between the coordinate components of the compensated landing point coordinates and the nominal landing point coordinates. The compensation offset is used to characterize the correction magnitude and direction introduced by distortion compensation at the current sampling point position. Using the reference landing point coordinates as the position reference, the compensation offset is superimposed on the reference landing point coordinates. The superposition operation is performed separately for each coordinate component. The superposition result is defined as the expected landing point coordinates, which are used to simultaneously reflect the visual measurement alignment result and the distortion compensation correction result.

[0032] The scanning field boundary is read and boundary constraints are applied to the desired landing point coordinates. The boundary constraints include truncating or projecting the coordinate components that exceed the scanning field boundary so that the desired landing point coordinates fall into the effective processing area of ​​the scanning field. The desired landing point coordinates after boundary constraints are associated with the sampling point identifier and output, which is used to form a residual sequence by difference with the measured landing point coordinates.

[0033] Specifically, the method for obtaining a parameter set for coordinate transformation by performing a fitting operation based on a pair of points with nominal coordinates and landing point coordinates includes... For each point on the dot matrix target, the nominal coordinates of the dot matrix are recorded, and the corresponding landing point coordinates are measured. A one-to-one pairing of the nominal coordinates and landing point coordinates is performed to form a paired point set. Data preprocessing is performed to unify the origin, unit, and axis conventions for the nominal and landing point coordinates. Missing values, duplicate points, and obvious out-of-bounds points are removed. Edge points are individually marked for weighting. The coordinate transformation model and parameter set are determined, and a mapping form from nominal coordinates to compensated landing point coordinates is selected. The mapping form must include at least linear terms such as translation, rotation, and scaling, and superimposed with higher-order terms to characterize the nonlinear distortion of the scanning field. The undetermined coefficients in the mapping form are defined as the parameter set. The residual and target quantity are constructed, and each nominal coordinate in the paired point set is substituted into the coordinate transformation model to obtain the predicted landing point coordinates. The difference between the coordinate components of the predicted landing point coordinates and the paired landing point coordinates is used as the residual. The sum of squared residual values ​​is used as the target value, and preset amplitude weights are assigned to edge points to enhance boundary fitting constraints. The parameter set is solved, with initial parameters obtained by first obtaining linear terms. Based on the initial parameters, least squares are performed on all undetermined coefficients. When higher-order terms are used, an iterative update method is adopted, and the iteration stopping condition is set to the target value decrease amplitude not exceeding a preset threshold or the parameter change amplitude not exceeding a preset threshold. Robust constraints and refitting are performed, and the pairing point set is screened for consistency according to the residual value. Pairing points with residual values ​​exceeding the outlier threshold are downweighted or removed. Least squares are repeated on the screened pairing point set to obtain a stable parameter set. The conversion results from nominal coordinates to compensation landing point coordinates are generated from the final parameter set, the residual distribution is statistically analyzed, and a verification record is formed. The parameter set and version information are stored together as distortion compensation parameters.

[0034] Specifically, methods for obtaining a parameter set by performing fitting operations based on reference image coordinates and known coordinates include: A calibration carrier containing reference features is set up, and the known coordinates of each reference feature are recorded. The vision unit acquires images of the calibration carrier and extracts the reference image coordinates for each reference feature. The reference image coordinates are paired one-to-one with the known coordinates to form a set of paired points. Coordinate and extraction preprocessing is performed, including sub-pixel localization, duplicate detection, and consistency filtering of the reference image coordinates. The units and coordinate axis conventions of the known coordinates are standardized. Occlusion, extraction failures, and out-of-bounds paired points are removed. A conversion model and parameter set are determined, and a mapping form from the reference image coordinates to the landing point coordinates is selected. The mapping form must include at least linear terms of translation, rotation, and scaling, and a perspective term to cover projection changes introduced by the camera pose. When lens distortion needs to be included, a distortion term with the radial distance of the image as the independent variable is added to the mapping form. The undetermined coefficients in the mapping form are defined as a parameter set. Residuals and target quantities are constructed, and each reference image coordinate is substituted into the mapping. The predicted landing point coordinates are obtained in the following manner; the difference between the coordinate components of the predicted landing point coordinates and the corresponding known coordinates is used as the residual; the sum of squares of the residual values ​​is used as the target value, and weights are set for the corner and edge reference features to improve the extrapolation stability; the parameter set is solved, and the initial parameters are obtained by using linear terms; the parameter set is obtained by performing least squares solution based on the initial parameters; when perspective or distortion terms are included, iterative solution is used, and the iteration stopping condition is set to the target value decrease rate not exceeding a preset threshold or the parameter change rate not exceeding a preset threshold; robust constraints and multiple fitting are performed, and the pairing point set is screened for consistency according to the residual value, and the pairing points with residual values ​​exceeding the outlier threshold are downweighted or removed; the selected pairing point set is solved repeatedly to obtain a stable parameter set; the pairing point set is back-substituted for verification using the parameter set, and the residual statistics and version information are output; the parameter set and version information are stored together as visual calibration parameters.

[0035] In this embodiment, the method steps for obtaining the residual sequence by differencing the measured landing point coordinates include: For each sampling point identifier, the expected landing point coordinates and the measured landing point coordinates are read; a difference operation is performed on the two according to their coordinate components to obtain the residual vector, and the residual value is calculated as a deviation measure; the sampling point identifier, residual vector, and residual value are arranged in the sampling point order to form a residual sequence, and written into the verification record to support the calculation of residual statistics; the purpose is to quantify the deviation of the expected landing point coordinates from the measured landing point coordinates, providing input for partitioned summarization and attribution calculation; a specific example is as follows: For sampling point P22, the expected landing point coordinates are read from the validation dataset as (10.02, 0.10), and the measured landing point coordinates are read as (9.92, 0.18). A difference operation is performed on the coordinate components to obtain the residual vector. The horizontal component of the residual vector is 10.02 minus 9.92, which equals 0.10, and the vertical component is 0.10 minus 0.18, which equals -0.08. When calculating the residual value, the sum of squares of the two components of the residual vector is calculated and the square root is taken. The sum of squares is 0.10 multiplied by 0.10 plus 0.08 multiplied by 0.08, which equals 0.0164. The square root is 0.1281. The residual value is used to characterize the deviation magnitude of sampling point P22.

[0036] To form the residual sequence, data from adjacent sampling point identifiers is read in the sampling point order. For example, sampling point identifier P23 reads the expected landing point coordinates as (0.08, 10.01) and the measured landing point coordinates as (0.15, 9.93), resulting in a residual vector of -0.07, 0.08, and a residual value of 0.1063. Sampling point identifier P5 reads the expected landing point coordinates as (100.03, 100.02) and the measured landing point coordinates as (99.98, 100.06), resulting in a residual vector of 0.05, -0.04, and a residual value of 0.0640. Sampling point identifiers P22, P23, and P5 are combined with their corresponding residual vectors and residual values ​​in the sampling point order to form the residual sequence. This residual sequence is then written into the verification record for subsequent partitioning and summarizing to obtain residual statistics, which are then compared with the accuracy threshold.

[0037] Step S3: Based on the spatial location of the sampling point set, perform partitioning and summarizing on the residual sequence to obtain residual statistics. Compare the residual statistics with the precision threshold. Partitions where the residual statistics exceed the precision threshold are recorded as regions exceeding the threshold. Output the set of regions exceeding the threshold.

[0038] In this embodiment, the method steps for obtaining residual statistics by performing partitioning and summarizing the residual sequence based on the spatial location of the sampling point set include: The process involves retrieving a set of sampling points and defining their spatial locations as the nominal coordinates corresponding to the sampling point identifiers. An index table is then created to map the sampling point identifiers to their spatial locations. The residual sequence is parsed, and the sampling point identifiers, residual vectors, and residual values ​​are extracted. Each residual record is mapped to its corresponding spatial location based on the index table. Partitioning rules are determined, dividing the scan field into central, edge, and corner partitions. These rules are expressed using polygonal boundaries or coordinate intervals and bound to partition identifiers. Based on these rules, residual records are partitioned and merged. Within each partition, the residual value set is summarized, and residual statistics are calculated. These statistics include at least the root mean square value and quantile values ​​to account for both overall and extreme deviations. A correspondence table between partition identifiers and residual statistics is output, serving as input for accuracy threshold comparison. The goal is to merge the residual sequence into different partitions of the scan field based on their spatial locations, generating a quantified result that can be used to locate areas of concentrated deviation.

[0039] To facilitate understanding of the mapping from sampling point identifiers to spatial locations, partitioning and merging, and the calculation of residual statistics, the following example is provided: The sampling point set is retrieved and the nominal landing point coordinates bound to the sampling point identifier are used as spatial locations to establish an index table. The index table includes (100, 100) for P5, (10, 0) for P22, (0, 10) for P23, and (0, 50) for P10. When parsing the residual sequence, the residual value of sampling point identifier P22 (0.1281), the residual value of sampling point identifier P23 (0.1063), the residual value of sampling point identifier P5 (0.0640), and the residual value of sampling point identifier P10 (0.0900) are read and mapped to the spatial locations (10, 0), (0, 10), (100, 100), and (0, 50) respectively according to the index table.

[0040] When determining the zoning rules, the scanning field boundary is used to divide the area into a central zone, an edge zone, and a corner zone. The central zone is defined as a coordinate range of 50 to 150 mm horizontally and 50 to 150 mm vertically. The edge zone is defined as the area within a range of 0 to 200 mm horizontally and 0 to 200 mm vertically, excluding the central zone and the corner zone. The corner zone is defined as the encrypted area near the four corners. The lower left corner of the corner zone is represented by a coordinate range of 0 to 20 mm horizontally and 0 to 20 mm vertically. The other corners are determined according to symmetrical intervals. The central zone, edge zone, and corner zone are respectively bound with zoning identifiers R1, R2, and R3.

[0041] When performing partition merging according to the partitioning rules, the spatial location (100, 100) corresponding to P5 is assigned to partition identifier R1, the spatial location (0, 50) corresponding to P10 is assigned to partition identifier R2, and the spatial locations (10, 0) corresponding to P22 and (0, 10) corresponding to P23 are assigned to partition identifier R3. The summation of the residual values ​​for partition identifier R1 yields 0.0640, the root mean square value is calculated to be 0.0640, and the quantile value is taken as 0.0640. The summation of the residual values ​​for partition identifier R2 yields 0.0900, and the root mean square value is calculated to be 0.0900. The value is 0.0900, and the quantile value is taken as 0.0900. The residual values ​​of the partition identifier R3 are summarized to obtain 0.1281 and 0.1063. The root mean square value is calculated as the square root of the sum of squares. The sum of squares is 0.1281 multiplied by 0.1281 plus 0.1063 multiplied by 0.1063 and then divided by 2, which equals 0.01384. The square root is 0.1176, and the larger residual value of 0.1281 is taken as the quantile value. The partition identifiers R1, R2, and R3 are written into the corresponding root mean square value and quantile value respectively into the corresponding table. The corresponding table is used as the input for the precision threshold comparison.

[0042] In this embodiment, the method steps of comparing residual statistics with a precision threshold, recording partitions where residual statistics exceed the precision threshold as threshold-exceeding regions, and outputting a set of threshold-exceeding regions include: A precision threshold is defined as the criterion for determining the allowable deviation of a partition. The precision threshold is determined based on the equipment's target precision, process tolerance, and historical residual distribution, and is stored in association with the partition identifier. For each partition identifier, the residual statistic and the precision threshold are retrieved and compared using the same statistical caliber. Partitions with residual statistics greater than the precision threshold are identified as exceeding the threshold. The partition identifiers of all exceeding the threshold regions are collected to form an over-threshold region set, and the corresponding set of sampling point identifiers and residual statistic values ​​are recorded simultaneously. The over-threshold region set is used to handle the recalculation of residual statistics for controlled parameter disturbances. Its purpose is to use a unified judgment criterion to screen out spatial partitions that require attribution and updating, narrowing the calculation range of parameter disturbances and improving maintenance efficiency. A specific example is as follows: When defining the accuracy threshold as the criterion for determining the allowable deviation of a partition, the determination idea is implemented as a threshold table with the same caliber as the residual statistics, and stored in association with the partition identifier. In the threshold table, the partition identifier R1 corresponds to an accuracy threshold of 0.08, the partition identifier R2 corresponds to an accuracy threshold of 0.10, and the partition identifier R3 corresponds to an accuracy threshold of 0.11. The threshold value is given an upper limit constraint based on the target accuracy of the equipment and the process tolerance, and more stringent or more lenient determination criteria are configured in combination with the diagonal part of the historical residual distribution to ensure that the determination caliber remains traceable and consistent in different partitions.

[0043] When retrieving the residual statistics and precision threshold for each partition identifier, the same statistical caliber is selected for comparison, with the root mean square value chosen as the comparison object. The root mean square value of partition identifier R1 is 0.0640, which is less than the precision threshold of 0.08, so partition identifier R1 is not considered to be outside the threshold region. The root mean square value of partition identifier R2 is 0.0900, which is less than the precision threshold of 0.10, so partition identifier R2 is not considered to be outside the threshold region. The root mean square value of partition identifier R3 is 0.1176, which is greater than the precision threshold of 0.11, so partition identifier R3 is considered to be outside the threshold region.

[0044] When the partition identifiers of each region exceeding the threshold are aggregated to form the over-threshold region set, the over-threshold region set includes the partition identifier R3, and simultaneously records the sampling point identifier set and residual statistic value corresponding to the partition identifier R3. The sampling point identifier set follows the aforementioned partition merging result as P22 and P23, and the residual statistic value is recorded as the root mean square value of 0.1176 and the quantile value of 0.1281. The over-threshold region set is used to carry out the subsequent processing flow of applying controlled perturbations to the distortion compensation parameters and visual calibration parameters and recalculating the residual statistics, thereby limiting the perturbation calculation to the spatial range corresponding to the partition identifier R3.

[0045] Step S4: Apply controlled perturbations to the distortion compensation parameters and visual calibration parameters respectively, recalculate the residual statistics to form the responsibility quantity, compare the responsibility quantity with the responsibility threshold, and output the deviation source label.

[0046] In this embodiment, the method steps for applying controlled perturbations to the distortion compensation parameters and visual calibration parameters and recalculating the residual statistics to form the liability quantity include: The table corresponding to the set of regions exceeding the threshold, partition identifiers, and residual statistics is retrieved. The nominal landing point coordinates, reference image coordinates, measured landing point coordinates, and parameter version information from the validation dataset are also retrieved to form the recalculation input set. A controlled perturbation sequence for distortion compensation parameters is generated, consisting of a preset perturbation step size, perturbation direction, and perturbation boundary. The controlled perturbation sequence is applied to the distortion compensation parameters item by item. While keeping the visual calibration parameters unchanged, the expected landing point coordinates are recalculated according to S2, and the residual sequence is reconstructed. Then, the residual statistics for the set of regions exceeding the threshold are recalculated according to S3. The distortion response is calculated for each partition identifier. The response quantity is defined as the absolute value or ratio of the difference between the recalculated residual statistic and the original residual statistic, and the mean or maximum value is taken according to the perturbation sequence to obtain the distortion responsibility quantity. A controlled perturbation sequence of visual calibration parameters is generated and applied to the visual calibration parameters. Under the condition of keeping the distortion compensation parameters unchanged, the recalculated input set is reused, and the residual statistics are recalculated according to steps S2 and S3 to obtain the calibration responsibility quantity. The distortion responsibility quantity and the calibration responsibility quantity are summarized according to the partition identifier to form a responsibility quantity set. The purpose is to construct a parameter-sensitive response within the limit range of the overthreshold region set, and to characterize the responsibility contribution of the deviation source by the change of the residual statistic.

[0047] In this embodiment, when the difference ratio is used as the distortion response quantity or the calibration response quantity, a lower limit protection is introduced for the denominator or the absolute value of the difference is preferred to suppress the amplification of the original residual statistics when they are small. At the same time, for each controlled disturbance dimension, a positive disturbance and a negative disturbance are constructed and the residual statistics are recalculated. The update direction is selected to be the disturbance direction that makes the target quantity decrease, so that the update direction can be directly derived from the calculation result and can be consistently reproduced in the recalculation.

[0048] To facilitate understanding the construction of controlled disturbance sequences, the process of recalculating residual statistics, and the formation of the responsibility set, the following example is provided: The set of regions exceeding the threshold is retrieved to obtain the partition identifier R3. The original residual statistic R3 is obtained by retrieving the partition identifier and the residual statistic correspondence table, and the root mean square value of the residual statistic is taken as 0.1176. The nominal landing point coordinates, reference image coordinates, measured landing point coordinates, and parameter version information corresponding to P22 and P23 are read from the validation dataset and compiled to form the recalculation input set. When generating the controlled perturbation sequence of distortion compensation parameters, the preset perturbation step size is one percent of the value range of distortion compensation parameters, the perturbation direction is positive and negative, and the perturbation boundary is two percent. Based on this, three perturbation values ​​are formed, namely negative one percent, zero, and positive one percent, and are applied to the distortion compensation parameters one by one. Taking positive one percent as an example, while maintaining visual calibration... Under the condition of unchanged parameters, the input set is reused, the expected landing point coordinates are recalculated according to S2 and the residual sequence is reconstructed, and then the residual statistics are recalculated only for the partition identifier R3 according to S3, and the root mean square value of the recalculated residual statistics of R3 is 0.1300; taking the negative one percent as an example, the root mean square value of the recalculated residual statistics of R3 is 0.1100 in the same way; when calculating the distortion response for the partition identifier R3, the distortion response is taken as the absolute value of the difference between the recalculated residual statistics and the original residual statistics. The absolute value of the difference for the positive one percent is 0.0124, and the absolute value of the difference for the negative one percent is 0.0076; the maximum value of the perturbation sequence is taken as the distortion liability, and the distortion liability of R3 is 0.0124.

[0049] When generating the controlled perturbation sequence for the visual calibration parameters, the preset perturbation step size is one percent of the adjustable range of the visual calibration parameters, the perturbation direction is either positive or negative, and the perturbation boundary is two percent. Based on this, three perturbation values—negative one percent, zero, and positive one percent—are generated and applied to the visual calibration parameters one by one. While keeping the distortion compensation parameters unchanged, the input set is reused, and the R3 residual statistics are recalculated according to S2 and S3. For example, positive one percent corresponds to an R3 root mean square value of 0.1600, negative one percent corresponds to a negative one percent, and negative one percent corresponds to a negative one percent. One percent corresponds to the R3 root mean square value of 0.1500; the absolute value of the difference is used to calculate the calibration response, with positive one percent corresponding to 0.0424 and negative one percent corresponding to 0.0324. The maximum value is taken as the calibration responsibility quantity according to the disturbance sequence, resulting in an R3 calibration responsibility quantity of 0.0424; the R3 distortion responsibility quantity of 0.0124 and the R3 calibration responsibility quantity of 0.0424 are summarized according to the partition identifier to form a responsibility quantity set, which is used for subsequent responsibility threshold comparison and deviation source label output.

[0050] In this embodiment, the method steps for comparing the responsibility amount with the responsibility threshold and outputting the deviation source label include: A responsibility threshold is defined as the criterion for determining responsibility discrepancies. The responsibility threshold is determined based on repeated measurement dispersion, recalculation noise level, and responsibility separation, and is stored in association with the partition identifier. For each partition identifier, the distorted responsibility quantity and the calibrated responsibility quantity are retrieved, and the responsibility discrepancy is calculated. The responsibility discrepancy is the absolute value of the difference between the distorted responsibility quantity and the calibrated responsibility quantity. The responsibility discrepancy is compared with the responsibility threshold. If the responsibility discrepancy is greater than the responsibility threshold, a deviation source label is output. The deviation source label indicates whether the deviation source points to the distortion compensation parameter or the visual calibration parameter according to the magnitude of the responsibility quantity. The deviation source label is bound to the set of over-threshold regions, the partition identifier, and the set of responsibility quantities. This is used to select the iterative update path for the distortion compensation parameter or the visual calibration parameter in S5. The deviation source label includes at least one pointing to the distortion compensation parameter and one pointing to the visual calibration parameter. The purpose is to transform the set of responsibility quantities into attribution conclusions that can be used for parameter iterative updates, reducing the trial-and-error range and improving maintainability and recalculability.

[0051] To facilitate understanding of the partition association of responsibility thresholds, the calculation of responsibility variance, and the output rules for deviation source labels, the following example is provided: When defining the responsibility threshold as the criterion for determining responsibility differences, the responsibility threshold is implemented as a threshold table associated with the partition identifier and stored. In the threshold table, the partition identifier R3 corresponds to the responsibility threshold of 0.0100. The determination of the responsibility threshold is based on the repeated measurement dispersion and the recalculation noise level to give a lower limit of difference, and the severity of the determination is adjusted in combination with the responsibility separation degree, so that the attribution conclusion is output only when the responsibility difference reaches the lower limit of difference.

[0052] When retrieving the distortion liability and calibration liability for partition identifier R3, the distortion liability is read as 0.0124 and the calibration liability is read as 0.0424. The liability difference is calculated as the absolute value of the difference between the distortion liability and the calibration liability, which is 0.0424 minus 0.0124 equals 0.0300. The liability difference of 0.0300 is compared with the liability threshold of 0.0100. Since the liability difference is greater than the threshold, a deviation source label is triggered. The deviation source label indicates the source of the deviation according to the magnitude of the liability, pointing to the visual calibration parameter, because the calibration liability is greater than the distortion liability.

[0053] During the recording phase, the deviation source label is bound and stored with the set of over-threshold regions, the partition identifier R3, and the set of responsibility quantities. The set of responsibility quantities includes the R3 distortion responsibility quantity of 0.0124 and the R3 calibration responsibility quantity of 0.0424. The responsibility threshold of 0.0100 and the responsibility difference quantity of 0.0300 used to trigger the comparison are recorded simultaneously. This is used to support the iterative update path selection for visual calibration parameters in S5, thereby transforming the set of responsibility quantities into attribution conclusions that can be used for parameter updates, reducing the number of trial and error rounds on site and retaining the basis for recalculation.

[0054] Step S5: Based on the bias source label, perform iterative updates on the distortion compensation parameters or visual calibration parameters. In each round of updates, recalculate steps S2 to S3, and compare the changes in the set of regions exceeding the threshold with the convergence threshold to determine whether to stop. When stopping, encapsulate the updated distortion compensation parameters and the updated visual calibration parameters into a combined parameter package. During runtime, calculate the sampling residual according to the sampling step and compare it with the drift threshold. When the sampling residual exceeds the drift threshold, pause the instruction and call the verification dataset for recalculation.

[0055] In this embodiment, refer to Figure 2 As shown, the method steps for iteratively updating distortion compensation parameters or visual calibration parameters based on the bias source label, recalculating S2 to S3 in each round of updates, and comparing the change in the set of out-of-threshold regions with the convergence threshold to determine whether to stop include: Read the deviation source labels, responsibility set, out-of-threshold region set, validation dataset, and current parameter version information; determine the update object as a distortion compensation parameter or visual calibration parameter based on the deviation source labels, and record the parameters that have not been updated as fixed parameters; construct a constrained candidate update set, which consists of multiple values ​​of the update object in a controlled perturbation sequence. The controlled perturbation sequence is limited by the update step size, update direction, and update boundary. The update direction is selected from the direction that reduces the responsibility in step S4, and the update boundary limits the parameter change range; perform recalculation and evaluation on each candidate update set, reusing the validation dataset, calculate the expected landing point coordinates and generate the residual sequence according to step S2, and then calculate the residual statistics and output the out-of-threshold region set according to step S3; select the best candidate based on the target value, which is a weighted average of the out-of-threshold region set size and the residual statistics, used to simultaneously compress the out-of-threshold region set and reduce the residual statistics; select the candidate with the smallest target value as... This round of updates the results and generates new parameter version information; it calculates the change in the set of regions exceeding the threshold, taking the symmetrical difference between the current set of regions exceeding the threshold and the previous set, and simultaneously calculates the change in the residual statistics, taking the decrease in the root mean square value; it compares the change in the set of regions exceeding the threshold and the change in the residual statistics with the convergence threshold, which is defined as the stopping criterion, based on the allowable change in residuals and the allowable region jitter; when the change in the set of regions exceeding the threshold is not greater than the convergence threshold and the change in the residual statistics is not less than the convergence threshold, it is recorded as a convergence round; when the number of convergence rounds reaches the preset consecutive round number threshold, a stop flag is output; when the target quantity increases relative to the previous round and the increase exceeds the rollback threshold, a rollback point is triggered and the update step size is reduced before entering the candidate update set for evaluation; the reason is to convert the deviation source label into an executable parameter update action and use the change trend of the set of regions exceeding the threshold to give the stopping criterion.

[0056] In this embodiment, the convergence threshold is preferably set as a convergence threshold group, which includes at least a first convergence threshold for limiting the number of symmetrical differences in the set of over-threshold regions, and a second convergence threshold for limiting the improvement magnitude of the residual statistics. When the region jitter satisfies the first convergence threshold and the improvement magnitude satisfies the second convergence threshold, and the preset number of rounds is met continuously, a stop flag is output to avoid ambiguity in the criteria caused by different dimensional indicators sharing a single threshold.

[0057] To facilitate understanding of the bias source label-driven candidate update set evaluation, objective quantity selection, and convergence determination process, an example is provided below: The deviation source label is read to obtain the visual calibration parameter. Simultaneously, the responsibility set, the out-of-threshold region set, the validation dataset, and the current parameter version information are read, with the current parameter version information denoted as V0. Based on this, the update object is determined to be the visual calibration parameter, and the distortion compensation parameter is denoted as the fixed parameter. When constructing the constrained candidate update set, the controlled perturbation sequence uses an update step size of 1%, an update direction that decreases the calibration responsibility in S4, and an update boundary of 2%, forming three candidate values ​​corresponding to visual calibration parameter version V0 minus 1%, visual calibration parameter version V0, and visual calibration parameter version V0 plus 1%, respectively. These three candidate values, along with the fixed parameter, constitute the candidate update set.

[0058] When performing recalculation evaluation on each candidate update set, the validation dataset is reused and the expected landing point coordinates are calculated according to S2 to generate a residual sequence. Then, the residual statistics are calculated according to S3 and the set of out-of-threshold regions is output. Taking visual calibration parameter version V0 minus one percent as an example, the recalculated R3 root mean square value of the partition identifier is 0.1050, and the set of out-of-threshold regions is empty. Taking visual calibration parameter version V0 as an example, the recalculated R3 root mean square value of the partition identifier is 0.1176, and the set of out-of-threshold regions includes R3. Taking visual calibration parameter version V0 plus one percent as an example, the recalculated R3 root mean square value of the partition identifier is 0.1300, and the set of out-of-threshold regions includes R3. When selecting the best target quantity, the target quantity is the weighted sum of the size of the set of regions exceeding the threshold and the root mean square value. The weights are the same. Then the three candidate target quantities are 0 plus 0.1050, 1 plus 0.1176, and 1 plus 0.1300, respectively. The minimum target quantity corresponds to visual calibration parameter version V0 minus one percent. It is selected as the result of this round of updates and a new parameter version information V1 is generated.

[0059] When calculating the change in the set of regions exceeding the threshold, the set of regions exceeding the threshold in the previous round is denoted as including R3, and the set of regions exceeding the threshold in the current round is empty, with a symmetric difference quantity of 1. Simultaneously, the change in the residual statistic is calculated by taking the root mean square decrease, which is 0.1176 minus 0.1050 equals 0.0126. When comparing the change in the set of regions exceeding the threshold and the change in the residual statistic with the convergence threshold, the convergence threshold allows for a region jitter of 1 and an allowable residual change of 0.0050. Therefore, the current round satisfies the condition that the symmetric difference quantity is no greater than 1 and the decrease... If the decrease is not less than 0.0050, it is recorded as a convergence round. When the threshold for consecutive rounds is 2, the next round of candidate update set is constructed starting from V1 and the evaluation is repeated. If the next round of the out-of-threshold region set is still empty and the root mean square value decrease is not less than 0.0050, the stop flag is output after two consecutive convergence rounds. If the target value increases relative to the previous round and the increase exceeds the rollback threshold of 0.0200, the rollback point is triggered to roll back to the parameter version information V0 and the update step size is reduced to five per thousand before entering the candidate update set evaluation.

[0060] In this embodiment, the method steps for encapsulating the updated distortion compensation parameters and the updated visual calibration parameters into a combined parameter package upon stopping include: The combined parameter package is defined as a collection record containing version information of distortion compensation parameters and version information of visual calibration parameters. The collection record also includes sampling point set identifiers, validation dataset identifiers, and stop round identifiers. The distortion compensation parameters and visual calibration parameters corresponding to the stop flag are written into the same collection record, along with a verification summary. The verification summary includes the set of out-of-threshold regions for the stop round, residual statistics, and convergence round counts, used to reproduce the stop judgment criteria. An effective condition field is written to the combined parameter package, including the scan field boundary and sampling step, used to limit the applicable scope of the combined parameter package. The purpose is to solidify traceable parameter combinations and ensure consistency in the calculation of expected landing point coordinates.

[0061] To facilitate understanding of the field composition of the combined parameter package, the method of writing collection records, and the limiting role of the effective condition fields, the following example is provided: When defining the combined parameter package as a set record, the fields of the set record are set as distortion compensation parameter version information, visual calibration parameter version information, sampling point set identifier, verification dataset identifier, and stop round identifier. The distortion compensation parameter remains a fixed parameter during the aforementioned iteration process, and the distortion compensation parameter version information is still D0. The visual calibration parameter obtains a new parameter version information V1 through the candidate update set, and outputs a stop flag when the consecutive round number threshold is reached. The stop round identifier is denoted as T3. The sampling point set identifier follows the aforementioned sampling point set generation result and is denoted as Samp200. The verification dataset identifier follows the aforementioned record result and is denoted as ValSet01. These constitute the primary key fields of the combined parameter package set record.

[0062] When writing the distortion compensation parameters and visual calibration parameters corresponding to the stop flag into the same set of records, the distortion compensation parameter version information D0 and the visual calibration parameter version information V1 are written, along with a verification summary. The verification summary includes the set of out-of-threshold regions corresponding to the stop round T3, the residual statistics, and the convergence round count. The out-of-threshold region set is empty, the residual statistics are recorded as the root mean square value of partition identifier R3 (0.1050) and the quantile value (0.1100), and the convergence round count is recorded as 2. This is used to reproduce the stopping judgment criteria and support recalculation and verification.

[0063] When writing the effective condition field to the combined parameter package, the effective condition field includes the scan field boundary and the sampling step. The scan field boundary follows the aforementioned landing point coordinate range of 0 to 200 mm in the horizontal direction and 0 to 200 mm in the vertical direction. The sampling step follows the sampling interval criterion of the operation stage, which is set to perform a sampling inspection once every 100 processing cycles. The set record limits the applicable scope of the combined parameter package with the scan field boundary and the sampling step, so that the same combined parameter package can be called under the same boundary and the same sampling step when calculating the expected landing point coordinates in the future.

[0064] In this embodiment, the steps of calculating the sampling residual at runtime according to the sampling step and comparing it with the drift threshold, generating a pause command when the sampling residual exceeds the drift threshold, and calling the method of recalculating the verification dataset include: The sampling step is defined as the sampling interval criterion. The sampling interval criterion is established based on the processing cycle and the sampling coverage. The sampling time is extracted from the processing sequence and the corresponding nominal landing point coordinates are extracted according to the sampling interval criterion.

[0065] At the sampling inspection time, the reference image coordinates and the actual landing point coordinates are collected. The expected landing point coordinates are calculated using the combined parameter package according to S2. The sampling inspection residual is obtained by performing a difference between the expected landing point coordinates and the actual landing point coordinates, and the sampling inspection residual record is formed according to the residual value.

[0066] The drift threshold is defined as the criterion for determining the allowable range of the sampling residual. The determination idea is based on the allowable risk level and the distribution of historical sampling residuals. The sampling residual is compared with the drift threshold. When the sampling residual is greater than the drift threshold, a pause command is generated and the version information of the combined parameter package and the sampling time information are locked. The pause command is used to pause the scanning command sequence sent to the galvanometer and laser emission control and its corresponding light output enable, so that the current workpiece's writing action stops in a controllable state.

[0067] In this embodiment, after generating the pause command, in addition to locking the version information of the combined parameter package and the sampling time information, the current workpiece identifier and processing sequence position are also recorded; during the pause, the verification dataset is recalculated and the set of out-of-threshold regions and the deviation source label are output. When the recalculation passes, the pause is lifted and the light output enable is restored. When the recalculation fails, the pause is maintained or the system is switched to safe mode and a maintenance prompt is output. Thus, the drift detection, risk handling and recovery conditions are written into an executable process to reduce the risk of batch mis-engraving.

[0068] The verification dataset is called for recalculation. The recalculation process reuses steps S2 to S4 to obtain the set of regions exceeding the threshold and the labels of the sources of deviation, and outputs the updated object and update suggestions. The update suggestions include the update step size and update boundary, which are used to continue the iterative update process. Its purpose is to capture parameter drift and trigger the recalculation closed loop during the processing and running phase, thereby reducing the risk of batch deviation.

[0069] To facilitate understanding of the connection between sampling time, sampling residual calculation, drift threshold determination, and triggering recalculation process, the following explanation uses the aforementioned combined parameter package, including distortion compensation parameter version information D0 and visual calibration parameter version information V1, with the scanning field boundary being 0 to 200 mm horizontally and 0 to 200 mm vertically, and the sampling point identifier including P22, as follows: When the sampling step is defined as the sampling interval criterion, the sampling interval criterion is set according to the processing cycle and sampling coverage to sample once every 100 processing cycles. The sampling time corresponding to the 100th processing cycle is extracted from the processing sequence, and the nominal landing point coordinates corresponding to the 100th processing instruction are read. The nominal landing point coordinates are 10 and 0, which are consistent with the nominal landing point coordinates of the sampling point identifier P22, so that the spatial position of the sampling point is comparable to that of the aforementioned verification dataset.

[0070] When acquiring the reference image coordinates and the measured landing point coordinates at the sampling inspection time, the vision unit acquires the reference image coordinates as (990, 945) and simultaneously measures the measured landing point coordinates as (9.70, 0.40). Using the combined parameter package, the distortion compensation parameter version information D0 and the vision calibration parameter version information V1 are called, and the expected landing point coordinates are calculated according to S2. For example, the expected landing point coordinates are calculated to be (10.02, 0.10). The sampling inspection residual is obtained by performing a difference between the expected landing point coordinates and the measured landing point coordinates. The sampling inspection residual vector is 0.32, -0.30, and the sampling inspection residual value is 0.4380. The sampling inspection time, nominal landing point coordinates, reference image coordinates, expected landing point coordinates, measured landing point coordinates, and sampling inspection residual value are written into the sampling inspection residual record.

[0071] When defining the drift threshold as the criterion for determining the allowable range of sampling residuals, the drift threshold is set to 0.20 based on the historical sampling residual distribution and the allowable risk level. The sampling residual value of 0.4380 is compared with the drift threshold of 0.20. If the sampling residual value is greater than the drift threshold, a pause command is generated and the combined parameter package version information D0 and V1, as well as the sampling time information, are locked to maintain the consistency of the recalculation input.

[0072] When calling the validation dataset for recalculation, the recalculation process reuses S2 to S4, recalculates the residual statistics of the validation dataset and outputs the set of regions exceeding the threshold, then applies controlled perturbations to form the responsibility quantity and outputs the bias source label; for example, the recalculated set of regions exceeding the threshold is re-included with the partition identifier R3 and the bias source label points to the distortion compensation parameter; the updated object is output synchronously as the distortion compensation parameter, and an update suggestion is output. The update suggestion includes an update step size of 0.5% and an update boundary of 1%, which are used to continue the iterative update process and restore the usable state of the combined parameter package.

[0073] Specifically, the steps for generating update suggestions include: Read the deviation source label, responsibility set, over-threshold region set, residual statistics, validation dataset and current parameter version information; the output fields of the update suggestion are defined as update object pointer, update step size and update boundary, and the update object pointer is taken as the distortion compensation parameter or visual calibration parameter.

[0074] The target of the update is determined based on the deviation source label, and the other parameter is recorded as a fixed parameter; when the deviation source label is an undetermined label, the target of the update is set to the alternating update order, and the update step size is set to the preset minimum step size.

[0075] Determine the update direction, with the corresponding responsibility quantity of the update object as the target decrease amount, and construct two controlled disturbances: forward probing and reverse probing, with the disturbance amplitude taking a preset minimum step size; under the condition that the fixed parameters remain unchanged, recalculate the trial residual statistics and the trial over-threshold region set according to steps S2 to S3 respectively, and select the one that reduces the target quantity as the update direction; the target quantity is composed of the over-threshold region set size and the residual statistics weighted together.

[0076] Determine the update step size, generate a step size sequence starting from the preset minimum step size, and increase or decrease the step size sequence by a factor; apply perturbation to each item of the step size sequence along the update direction and recalculate the target quantity according to S2 to S3, filter the step sizes where the target quantity decreases and the decrease exceeds the improvement threshold, and select the step size with the smallest value as the update step size; when none of the step size sequences trigger the improvement threshold, set the update step size to the preset minimum step size and output a review prompt.

[0077] Determine the update boundary, read the value range of the update object and set the boundary coefficient. The boundary coefficient is determined based on the repeated measurement dispersion and recalculation noise level. The update boundary is obtained by multiplying the update step size by the boundary coefficient. The update boundary is compared with the range boundary and the smaller one is taken as the final update boundary to limit the update amplitude of a single round.

[0078] Output update suggestions, writing the update object, update direction, update step size, and update boundary into the update suggestion record, along with a recalculation basis field. This field includes the target quantity decrease obtained from trial recalculation, changes in the set of regions exceeding the threshold, and the corresponding residual statistics. These are used for constructing candidate update sets and determining rollback in the subsequent iterative update process. A specific example is shown below: The deviation source label is read to obtain the visual calibration parameter pointed to by the deviation source label. The responsibility set is read to obtain the distortion responsibility value of 0.0124 and the calibration responsibility value of 0.0424 corresponding to the partition identifier R3. The out-of-threshold region set is read to obtain the out-of-threshold region set containing the partition identifier R3. The residual statistics are read to obtain the root mean square value of the partition identifier R3 of 0.1176. The validation dataset is read to obtain the nominal landing point coordinates, reference image coordinates and measured landing point coordinates corresponding to the sampling point identifiers P22 and P23. The current parameter version information is read to obtain the distortion compensation parameter version information D0 and the visual calibration parameter version information V0. The output fields of the update suggestion are set to update object pointer, update step size and update boundary, where the update object pointer is limited to the distortion compensation parameter or the visual calibration parameter.

[0079] The target of the update is determined based on the deviation source label as the visual calibration parameter, and the distortion compensation parameter is recorded as a fixed parameter. Since the deviation source label is not an undetermined label, the update step size is not directly output using the preset minimum step size, but instead enters the direction and step size determination process.

[0080] When determining the update direction, a calibration responsibility amount of 0.0424 is used as the target decrease amount. Two controlled perturbations, forward and reverse, are constructed, with the perturbation amplitude set to 1% of the preset minimum step size. Under the condition that the distortion compensation parameter version information D0 remains unchanged, 1% forward and 1% reverse are applied to the visual calibration parameter version information V0 respectively, and recalculated according to steps S2 to S3. In the recalculation results, the forward trial yields a root mean square value of 0.1300 for the partition identifier R3, and the set of regions exceeding the threshold contains R3. The reverse trial yields a root mean square value of 0.1050 for the partition identifier R3, and the set of regions exceeding the threshold is empty. The target amount is the weighted sum of the size of the set of regions exceeding the threshold and the root mean square value, with the weights being equal. Therefore, the target amount for the forward trial is 1 plus 0.1300, and the target amount for the reverse trial is 0 plus 0.1050. The term representing the decrease in the target amount is the reverse trial, and the update direction is determined to be reverse.

[0081] When determining the update step size, a step size sequence of 1%, 0.5%, and 0.25% is generated starting from the preset minimum step size of 1%. Perturbations are applied item by item along the reverse update direction, and the target quantity is recalculated according to steps S2 to S3. The recalculated target quantity is 0 plus 0.1050 for 1%, 0.1100 for 0.5%, and 1 plus 0.1140 for 0.25%. When the improvement threshold is 0.0050, the decrease of 1% relative to the original target quantity 1 plus 0.1176 is 1.0126 and exceeds the improvement threshold; the decrease of 0.5% is 1.0076 and exceeds the improvement threshold; and the decrease of 0.25% is 0.0036 and does not exceed the improvement threshold. After filtering, the step size with the smallest value that triggers the improvement threshold is selected as 0.5%, and the update step size is determined to be 0.5%.

[0082] When determining the update boundary, the range boundary of the visual calibration parameter is set to 2%, the boundary coefficient is 4, and the update boundary is obtained by multiplying the update step size by the boundary coefficient to get 2%. The update boundary of 2% is compared with the range boundary of 2%, and the final update boundary is set to 2%, which is used to limit the update amplitude of a single round.

[0083] When outputting update suggestions, the update suggestion record is written to the visual calibration parameter as the update object, the update direction is reverse, the update step size is 0.5%, the update boundary is 2%, and it is written to the recalculation basis field. The recalculation basis field records the decrease of the reverse trial target quantity, the change of the set of out-of-threshold regions from containing R3 to empty, and the decrease of the root mean square value of the partition identifier R3 from 0.1176 to 0.1100, which are used to continue the construction of candidate update sets and rollback determination.

[0084] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A positioning method for a planar laser engraving machine, characterized in that, include: Step S1: Establish a combined mapping relationship between the reference image coordinates output by the vision unit and the landing point coordinates to obtain the sampling point set covering the scanning field; The coordinates of the reference image are collected for each sampling point and the points are marked to obtain the coordinates of the measured landing point, forming a verification dataset bound to the combined mapping relationship; Step S2: Based on the distortion compensation parameters and visual calibration parameters, calculate the expected landing point coordinates from the validation dataset, and differ them with the measured landing point coordinates to obtain the residual sequence; Step S3: Based on the spatial location of the sampling point set, perform partitioning and summarizing on the residual sequence to obtain residual statistics. Compare the residual statistics with the precision threshold. Partitions where the residual statistics exceed the precision threshold are recorded as regions exceeding the threshold. Output the set of regions exceeding the threshold. Step S4: Apply controlled perturbations to the distortion compensation parameters and visual calibration parameters respectively and recalculate the residual statistics to form the responsibility quantity. Compare the responsibility quantity with the responsibility threshold and output the deviation source label. Step S5: Based on the bias source label, perform iterative updates on the distortion compensation parameters or visual calibration parameters. In each round of updates, recalculate S2 to S3 and compare the changes in the set of regions exceeding the threshold with the convergence threshold to determine whether to stop. When stopping, encapsulate the updated distortion compensation parameters and the updated visual calibration parameters into a combined parameter package. During runtime, calculate the sampling residual according to the sampling step and compare it with the drift threshold. When the sampling residual exceeds the drift threshold, generate a pause command and call the verification dataset for recalculation.

2. The positioning method for a planar laser engraving machine according to claim 1, characterized in that, Methods for creating validation datasets include: The scanning field boundary is determined and a sampling layout is generated according to a layered strategy. The landing point coordinates determined by the sampling layout are defined as nominal landing point coordinates and bound with the sampling point identifier to form a sampling point set. The nominal landing point coordinates are obtained to drive the galvanometer to mark the points. The visual unit acquires an image including the reference features and the markings and extracts the reference image coordinates and the pixel coordinates of the markings. Based on the combined mapping relationship, the pixel coordinates of the markings are converted into the measured landing point coordinates. The sampling point identifier, reference image coordinates, nominal landing point coordinates, measured landing point coordinates and combined mapping relationship version information are collected and stored to form a verification dataset.

3. The positioning method for a planar laser engraving machine according to claim 2, characterized in that, Methods for calculating the expected landing point coordinates include: Obtain the sampling point identifiers, nominal landing point coordinates, and reference image coordinates of the validation dataset; convert the reference image coordinates into reference landing point coordinates using visual calibration parameters, and convert the nominal landing point coordinates into compensated landing point coordinates using distortion compensation parameters; the compensation offset is the compensated landing point coordinates minus the nominal landing point coordinates; the desired landing point coordinates are the reference landing point coordinates plus the compensation offset.

4. The positioning method for a planar laser engraving machine according to claim 3, characterized in that, Methods for obtaining residual sequences include: For each sampling point identifier, obtain the expected landing point coordinates and the measured landing point coordinates; perform a difference operation on the two according to the coordinate components to obtain the residual vector, and calculate the residual value as the deviation measure; arrange the sampling point identifier, residual vector and residual value according to the sampling point order to form a residual sequence.

5. The positioning method for a planar laser engraving machine according to claim 4, characterized in that, Methods for obtaining residual statistics include: The spatial location is the nominal landing point coordinate bound to the sampling point identifier; an index table is established to map the sampling point identifier to the spatial location, and the residual sequence is parsed to obtain the sampling point identifier, residual vector, and residual value; the residual value is mapped to the spatial location according to the index table, and the center partition, edge partition, and corner partition are divided based on the scanning field boundary and the partition identifier is bound; the residual value set is merged according to the partition identifier and the residual statistics are calculated, including the root mean square value and the quantile value.

6. The positioning method for a planar laser engraving machine according to claim 5, characterized in that, Methods for outputting the set of out-of-threshold regions include: For each partition identifier, retrieve the residual statistics and precision threshold, and perform a size comparison according to the same statistical caliber; partition identifiers with residual statistics greater than the precision threshold are determined to be regions exceeding the threshold; collect all partition identifiers exceeding the threshold to form a set of regions exceeding the threshold.

7. The positioning method for a planar laser engraving machine according to claim 6, characterized in that, Methods for determining liability include: Retrieve the table corresponding to the set of regions exceeding the threshold, the partition identifier, and the residual statistics. Extract the nominal landing point coordinates, the reference image coordinates, the measured landing point coordinates, and the parameter version information from the validation dataset to form a recalculation input set. Apply a controlled perturbation sequence defined by a preset perturbation step size, perturbation direction, and perturbation boundary to the distortion compensation parameters, while keeping the visual calibration parameters unchanged. Recalculate the residual statistics corresponding to the set of regions exceeding the threshold according to steps S2 and S3. Generate the distortion liability quantity based on the absolute value or ratio of the difference between the recalculated residual statistics and the original residual statistics. Apply a controlled perturbation sequence to the visual calibration parameters, while keeping the distortion compensation parameters unchanged, and recalculate the residual statistics to generate the calibration liability quantity. Summarize the distortion liability quantity and the calibration liability quantity according to the partition identifier to form a liability quantity set.

8. The positioning method for a planar laser engraving machine according to claim 7, characterized in that, Methods for outputting deviation source labels include: The liability threshold is used as the criterion for determining liability differences and is stored in association with the partition identifier. For each partition identifier, the distortion liability quantity and the calibration liability quantity are retrieved and the liability difference quantity is calculated. The liability difference quantity is the absolute value of the difference between the distortion liability quantity and the calibration liability quantity. The liability difference quantity is compared with the liability threshold. If the liability difference quantity is greater than the liability threshold, a deviation source label is output. The deviation source label indicates the deviation source to the distortion compensation parameter or the visual calibration parameter according to the relationship between the distortion liability quantity and the calibration liability quantity.

9. A positioning method for a planar laser engraving machine according to claim 8, characterized in that, Methods for determining whether to stop by comparing the changes in the set of out-of-threshold regions with a convergence threshold include: Read the bias source labels, the set of regions exceeding the threshold, and the validation dataset; select the distortion compensation parameter or visual calibration parameter according to the bias source label and perform the update, while keeping the other parameter unchanged; generate a candidate update set according to the update step size, update direction, and update boundary; recalculate item by item according to steps S2 to S3, and obtain the target quantity by weighting it with the size of the set of regions exceeding the threshold and the residual statistics, and select the smallest term of the target quantity to generate a new parameter version; compare the number of symmetric differences and the decrease of the root mean square of the set of regions exceeding the threshold with the convergence threshold respectively; when the number of symmetric differences is not greater than the convergence threshold and the decrease of the root mean square is not less than the convergence threshold, it is recorded as the convergence round, and when the number of convergence rounds reaches the preset consecutive round number threshold, output a stop flag.

10. A positioning method for a planar laser engraving machine according to claim 9, characterized in that, The methods for generating pause instructions and calling verification dataset recalculation include: The sampling step is based on the sampling interval criterion. The sampling time is extracted from the processing sequence according to the sampling interval criterion, and the corresponding nominal landing point coordinates are extracted. At the sampling time, the reference image coordinates and the measured landing point coordinates are collected, and the expected landing point coordinates are calculated using the combined parameter package according to step S2. The sampling residual is obtained by the difference between the expected landing point coordinates and the measured landing point coordinates, and a sampling residual record is formed. When the sampling residual is greater than the drift threshold, a pause command is generated, and the version information of the combined parameter package and the sampling time information are locked. The validation dataset is called to recalculate and reuse the output of the over-threshold region set and the deviation source label from steps S2 to S4.

Images