A method and system for detecting the center distance of a curved hole by using two lasers

Abstract

The application discloses a kind of double laser datum detection methods and systems of hole center distance, belong to the technical field of precision geometric quantity detection, to solve the problem that the actual space axis of curved hole is difficult to measure and the center distance of different attitude curved hole is susceptible to datum transmission, workpiece attitude and clamping offset, the application is collected by double laser unit aperture profile and layered inner wall point, attitude compensation is carried out under the unified fixture datum, different variance robust section fitting and curvature constraint space axis fitting, and the center distance uncertainty is calculated by propagating covariance, according to the adaptive complex scanning of leading error and incremental fusion, the technical effects of obtaining the real center distance of curved hole, effective measurement section and credible detection state are realized.

Description

Technical Field

[0001] This invention relates to the field of precision geometric measurement technology, and in particular to a method and system for detecting the center distance of curved holes using dual laser references. Background Technology

[0002] Curved holes are widely found in components with flow channels, lubrication channels, cooling channels, or complex connecting channels. With the development of precision manufacturing and assembly inspection, the center distance between two curved holes is not only used to evaluate the hole position relationship, but also affects the assembly consistency and channel connectivity of related structures. Existing measurement methods typically use plug gauges, calipers, single-sided contact probes, or hole image inspection to measure the position of directly accessible holes.

[0003] Because the axis of the curved hole bends along the depth direction and cannot be directly penetrated inside the hole, the center of the hole opening cannot fully represent the actual spatial axis inside the hole. Furthermore, the local inner wall points obtained by a single-sided probe may have inconsistent measurement quality due to obstruction, different incident angles, and varying measurement depths, making it difficult to stably determine the effective measurement section and its center. Therefore, measurements of the hole opening position or a single depth section are insufficient to reflect the overall geometric orientation of the curved hole.

[0004] When the orientation, depth, or surface location of two curved holes are inconsistent, existing methods typically require multiple adjustments to the workpiece posture or changes in the measurement reference. Reference transfer between different measurement positions, workpiece tilting, clamping translation, and manual readings all contribute to the center distance result. Furthermore, existing processes often lack covariance propagation for each error source and a rescanning mechanism for the dominant error, making it difficult to determine the reliability of the test results and to perform directional supplementary measurements in low-quality depth ranges.

[0005] Therefore, there is a need for a dual-laser reference method and system for detecting the center distance of curved holes that can overcome the shortcomings of the existing technologies. Summary of the Invention

[0006] One objective of this invention is to propose a dual-laser reference detection method and system for the center distance of curved holes. Addressing the problems of existing technologies that struggle to obtain the actual spatial axis of curved holes, require multiple measurement reference conversions for curved holes with varying orientations, and are susceptible to workpiece orientation and clamping offsets, this invention proposes a method using a fixture reference coordinate system and dual-laser measurement data to perform orientation compensation on the hole contour and layered inner wall points. Robust weighted fitting is performed on the cross-sections of each depth layer according to the light stripe quality and heteroscedasticity. The spatial axis is fitted under curvature constraints, and the measurement, fitting, and orientation errors are propagated to obtain the center distance uncertainty. Finally, based on the dominant error adaptive rescanning and incremental fusion technique, this invention achieves the technical effects of reducing the impact of multiple reference conversions, characterizing the true geometric orientation of curved holes, and providing reliable detection results.

[0007] This invention provides a method for detecting the center distance of curved holes using dual laser references, comprising: S1. Establish a fixture reference coordinate system using the fixture reference surface and non-collinear reference marks. Use laser measurement units corresponding to the two curved holes to collect the hole contour points, layered inner wall points, light stripe quality data, reference mark points, and tilt angles. S2. Based on the calibration parameters, reference marks and tilt angle of the two laser measurement units, determine the measurement coordinate transformation relationship, workpiece tilt rotation matrix, clamping translation offset and attitude compensation covariance, perform attitude compensation on the orifice contour points and layered inner wall points, and determine the layered depth reference from the compensated orifice contour points to obtain the curved hole point set. S3. Determine the measurement point weights of the layered inner wall points based on the light stripe quality data, and obtain the cross-sectional center sequence of the two curved holes by robustly weighting the cross-sectional center and center covariance according to the depth layer. S4. Based on the cross-sectional center sequence and center covariance, fit a spatial curve under curvature continuity constraint and curvature upper limit constraint to obtain the spatial axis of the two curved holes, the effective measurement cross section and the axis covariance. S5. Project the two spatial axes onto the same reference plane in the fixture reference coordinate system, calculate the center distance, and propagate the axis covariance and attitude compensation covariance to obtain the center distance uncertainty. When the center distance uncertainty exceeds the preset threshold, adjust the measurement configuration of the corresponding curved hole according to the uncertainty decomposition result to perform a rescan, and incrementally fuse the rescan points according to steps S2 to S4, and output the center distance, center distance uncertainty, effective measurement section and detection status at the stop.

[0008] Optionally, S1 includes: The two curved holes are respectively designated as the first curved hole and the second curved hole; At least three non-collinear reference marks, an inclinometer, a first laser measurement unit, and a second laser measurement unit are provided on the reference surface of the fixture, such that the first laser measurement unit corresponds to the first curved hole and the second laser measurement unit corresponds to the second curved hole; Using a calibration device with known geometric dimensions and known feature point coordinates, calibration images and laser beams of two laser measurement units are acquired respectively to obtain camera intrinsic parameters, distortion parameters, and laser plane parameters; The scanning position is changed along the depth direction of each curved hole, and the hole contour point or inner wall point, sampling depth, incident angle, light stripe width, signal-to-noise ratio, depth attenuation coefficient, number of effective points on the left and right walls and acquisition time of each sampling position are associated and stored to form the measurement dataset of the corresponding curved hole.

[0009] Optionally, S2 includes: Based on the known feature point coordinates of the calibration component and the corresponding feature point coordinates collected by the two laser measurement units, the external parameters of the two laser measurement units relative to the fixture reference coordinate system are calculated respectively. The attitude transformation of the reference marker is obtained by calculating the measured coordinates of the reference marker and its nominal coordinates in the reference coordinate system of the fixture. The attitude transformation of the tilt angle is obtained by calculating the tilt angle. The attitude transformation of the marker and the attitude transformation of the tilt angle are weighted and fused to obtain the workpiece tilt rotation matrix. The clamping translation offset is determined based on the coordinate residual of the reference marker point after correction by the workpiece tilt and rotation matrix. The orifice contour points and layered inner wall points are transformed sequentially using the external parameters, workpiece tilt rotation matrix, and clamping translation offset. The attitude compensation covariance is determined based on the covariance of each transformation parameter, and the covariance of each transformation parameter and the attitude compensation covariance are associated with and stored with the transformed points.

[0010] Optionally, S3 includes: The calibration light strip width, noise threshold, sensor noise variance, and interval normalization boundary are determined based on the repeated sampling results of the calibration component at each scanning depth. The initial local cross-sectional model residual variance is determined based on the prefit residual of the inner wall point of the corresponding depth layer. The cosine value of the incident angle, the ratio of the light stripe width to the calibrated light stripe width, the ratio of the signal-to-noise ratio to the noise threshold, the depth attenuation coefficient, and the ratio of the first to the last value of the effective points on the left and right walls arranged in ascending order are all normalized by intervals to obtain the incident quality factor, width quality factor, signal-to-noise quality factor, depth quality factor, and double-wall coverage quality factor. The product of each quality factor is divided by the sum of the sensor noise variance and the local cross-sectional model residual variance to obtain the measurement point weight of each layered inner wall point. Using the measured point weights as the basic weights, the weights of the inner wall points where the residual of the cross-sectional model exceeds the residual scale parameter are reduced iteratively using the robust loss function. A circular or elliptical cross-section is fitted for each depth layer to obtain the center of the cross-section. The central covariance is determined based on the fitted Jacobian matrix, the iterated measurement point weights, and the variance of the cross-sectional model residuals.

[0011] Optionally, S4 includes: Using the inverse matrix of the central covariance as the fitting weight of the cross-sectional center sequence, a data fitting term, a robust loss term, and a second-order difference penalty term for adjacent control points are constructed for the cubic B-spline curve. Under the condition that the curvature at each sampling position of the curve does not exceed the upper limit of curvature determined by the design curvature and manufacturing tolerance of the bore to be measured, the control points of the cubic B-spline curve are obtained to obtain the spatial axis. The continuous depth interval that satisfies the following conditions is not less than the preset number of effective points on both the left and right walls, the double-wall coverage quality factor is not less than the preset coverage threshold, and the characteristic values ​​of the center covariance are not greater than the preset covariance threshold. The axis covariance is determined based on the central covariance, the fitting weights, and the Jacobian matrix of the cubic B-spline curve with respect to the control points.

[0012] Optionally, S5 includes: The center distance is selected as the distance between the intersection points of two axes on the predetermined reference section according to the detection task configuration, or the shortest distance between two projected spatial axes is selected as the center distance. Based on the center distance calculation, the covariance propagation is performed on the external parameters of the two spatial axes, the two laser measurement units, the workpiece tilt rotation matrix, and the clamping translation offset matrix for the two spatial axes, the cross section fitting error, the axis fitting error, the external parameter error, the tilt error, and the translation error to obtain the center distance uncertainty and the uncertainty components of each error source, and each uncertainty component is associated with the laser measurement unit identifier, the curved hole identifier, and the depth range. The preset threshold is determined based on the detection tolerance bandwidth and the preset confidence level, and the coverage factor corresponding to the preset confidence level is combined with the center distance uncertainty to generate a confidence interval. A pass / fail judgment is generated only when the center distance uncertainty is not greater than the preset threshold, and when the confidence interval is completely within the detection tolerance zone; a fail / fail judgment is generated when the confidence interval is separated from the detection tolerance zone; and a re-inspection status is generated when the confidence interval intersects with the detection tolerance zone but is not completely within the detection tolerance zone. When the center distance uncertainty is greater than the preset threshold, a state pending re-inspection is generated. Furthermore, it also includes an anomaly recovery process, which receives measurement anomaly records and uncertainty components of each error source, determines the target laser measurement unit, target curved hole and depth range to be scanned based on the laser measurement unit identifier, curved hole identifier and depth range associated with the measurement anomaly record or uncertainty component, and generates a candidate measurement configuration that includes at least one parameter among laser incident angle, scanning depth, scanning step distance and workpiece micro-rotation angle. According to the coordinate compensation rule in step S2 and the measurement point weight rule in step S3, the rescanned points are incrementally merged with the original scanned inner wall points, and steps S3 to S5 are executed again. Furthermore, the measurement anomaly records include single-wall coverage anomaly records, missing measurement interval anomaly records, and curvature constraint infeasibility anomaly records; When a depth layer only obtains the inner wall point on one side of the curved hole, stop outputting the cross-sectional center of that depth layer, and determine the depth range to be scanned and the direction of the missing wall based on the adjacent depth layers that satisfy double wall coverage; When the depth interval between adjacent effective measurement sections exceeds the missing measurement interval threshold determined by the scanning step, the depth interval is determined as the depth interval to be scanned. When the curvature of the candidate curve before constraint exceeds the upper limit of curvature determined by the design curvature and manufacturing tolerance of the borehole to be measured, and a feasible fitting result that satisfies the upper limit of curvature constraint cannot be obtained, the depth interval to be supplemented is extended to both sides along the depth interval, and the depth interval is excluded from the effective measurement section. Associate each measurement anomaly record with the corresponding laser measurement unit identifier, orifice identifier, and depth range to be scanned; Furthermore, the uncertainty components of each error source associated with the target laser measurement unit, the target curved hole, and the depth range to be scanned are arranged in descending order of value, and the first component is determined as the dominant error component. When the measurement point error or cross-section fitting error is the dominant error component, for the candidate laser incident angle or scanning depth, the predicted value of the light stripe quality data of each candidate measurement point in the depth interval to be rescanned is determined according to the calibration parameters of the two laser measurement units and the candidate measurement configuration. The predicted weight of each candidate measurement point is determined and summed according to the measurement point weight rule in step S3. The candidate laser incident angle or scanning depth is selected by the increase of the sum of predicted weights relative to the sum of measurement point weights in the same depth interval before rescanning. When the axis fitting error is the dominant error component, reduce the scanning step size and increase the sampling depth layer within the depth range to be scanned. When external parameter error, tilting error or translation error are the dominant error components, select a workpiece micro-rotation angle within the fixture's allowable rotation angle and the workpiece's anti-collision boundary that increases the number of visible reference markers and makes the tilt angle change measurable. Record the rescan measurement configuration, number of rescans, center distance uncertainty after rescan, and reason for stopping; When the center distance uncertainty is not greater than the preset threshold, the rescanning is stopped and a rescanning convergence state is generated. The rescanning will only stop when the center distance uncertainty is greater than the preset threshold, the number of rescans reaches the preset upper limit, the candidate laser measurement configuration exceeds the movement range of the laser measurement unit, or the micro-rotation angle of the candidate workpiece exceeds the allowable rotation angle of the fixture or the anti-collision boundary of the workpiece, and an uncertainty failure state and a re-inspection state will be generated.

[0013] On the other hand, the present invention also provides a dual-laser reference detection system for the center distance of curved holes, comprising: The data acquisition module is used to establish the fixture reference coordinate system and acquire the orifice contour points, layered inner wall points, light stripe quality data, reference marker points, and inclination angles of the two curved holes. The coordinate compensation module is used to determine the measurement coordinate transformation relationship, workpiece tilt rotation matrix, clamping translation offset, and attitude compensation covariance, and generate the attitude-compensated curved hole point set and layer depth reference. The cross-section fitting module is used to determine the measurement point weights and obtain the cross-section center sequence and center covariance. The axis fitting module is used to obtain the spatial axis, effective measurement cross-section, and axis covariance under curvature continuity constraints and curvature upper limit constraints. The center distance detection module is used to project the two spatial axes onto the same reference plane, calculate the center distance and center distance uncertainty, adjust the measurement configuration of the corresponding curved holes according to the uncertainty decomposition results, control the incremental fusion of rescan data, and output the center distance, center distance uncertainty, effective measurement cross-section, and detection status.

[0014] The beneficial effects of this invention are: 1. By using two laser measurement units to collect data of two curved holes under a common fixture reference coordinate system, and by combining the reference mark points and tilt angles to compensate for workpiece tilt and clamping translation, curved holes with different orientations, depths or surfaces can be projected onto the axis and calculated for center distance under the same reference, reducing the impact of multiple reference conversions on the measurement results.

[0015] 2. By combining the incident angle, light stripe width, signal-to-noise ratio, depth attenuation, and double-wall coverage with the sensor noise variance and the local cross-sectional model residual variance into point-by-point weights, and then fitting the cross-sectional center and spatial axis sequentially under robust loss and curvature constraints, the inner wall points with different measurement qualities contribute to the fitting results in a way that is commensurate with their reliability. At the same time, the effective measurement cross-section and the corresponding covariance can be determined.

[0016] 3. The center distance uncertainty is obtained by propagating the measurement points, cross-section fitting, axis fitting, external parameters, tilt and translation errors. Based on the dominant error components, the laser incident angle, scanning depth, scanning step distance or workpiece micro-rotation angle is selected to perform a rescan, so that the supplementary measurement action can target low confidence sources. The rescan data uses the same coordinate compensation and weighting rule incremental fusion, thereby forming an output with confidence interval, effective measurement cross-section and clear detection status. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the dual-laser reference detection method for the center distance of curved holes according to the present invention.

[0018] Figure 2 This is a flowchart of the adaptive rescan based on dominant error in S5 of the present invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-2 A method for detecting the center distance of curved holes using dual laser references, comprising: S1. Establish a fixture reference coordinate system using the fixture reference surface and non-collinear reference marks. Use laser measurement units corresponding to the two curved holes to collect the hole contour points, layered inner wall points, light stripe quality data, reference mark points, and tilt angles. S2. Based on the calibration parameters, reference marks and tilt angle of the two laser measurement units, determine the measurement coordinate transformation relationship, workpiece tilt rotation matrix, clamping translation offset and attitude compensation covariance, perform attitude compensation on the orifice contour points and layered inner wall points, and determine the layered depth reference from the compensated orifice contour points to obtain the curved hole point set. S3. Determine the measurement point weights of the layered inner wall points based on the light stripe quality data, and obtain the cross-sectional center sequence of the two curved holes by robustly weighting the cross-sectional center and center covariance according to the depth layer. S4. Based on the cross-sectional center sequence and center covariance, fit a spatial curve under curvature continuity constraint and curvature upper limit constraint to obtain the spatial axis of the two curved holes, the effective measurement cross section and the axis covariance. S5. Project the two spatial axes onto the same reference plane in the fixture reference coordinate system, calculate the center distance, and propagate the axis covariance and attitude compensation covariance to obtain the center distance uncertainty. When the center distance uncertainty exceeds the preset threshold, adjust the measurement configuration of the corresponding curved hole according to the uncertainty decomposition result to perform a rescan, and incrementally fuse the rescan points according to steps S2 to S4, and output the center distance, center distance uncertainty, effective measurement section and detection status at the stop.

[0021] In this specific embodiment, S1 includes: In this specific embodiment, a precision housing with two curved cooling holes is used as the workpiece to be measured. The two curved holes are sequentially registered as the first curved hole and the second curved hole. Four non-collinear ceramic fixture coordinate marks are fixed on the fixture reference surface, and three non-collinear workpiece posture marks are fixed on the workpiece reference carrier that moves synchronously with the workpiece. A binocular reference mark measuring device consisting of two global shutter cameras is fixed to the outer frame of the fixture. The common field of view of the two cameras covers the four fixture coordinate marks and the three workpiece posture marks. An absolute angle encoder is configured on the micro-rotation mechanism of the fixture around the reference surface normal. The binocular reference mark measuring device, the absolute angle encoder, the inclinometer, the first laser measuring unit, the second laser measuring unit, and the fixture controller are connected to the same hardware trigger bus. The scanning field of view of the first laser measuring unit covers the first curved hole, and the scanning field of view of the second laser measuring unit covers the second curved hole. The controller assigns workpiece identifier, curved hole identifier, measuring unit identifier, and scanning batch number to the workpiece, the curved hole, the laser measuring unit, and each scan, respectively, so that subsequent points, calibration parameters, and abnormal records maintain the same object spectrum. The fixture controller establishes a transverse unit axis with the center of the first datum mark as the origin, pointing from the first datum mark to the second datum mark, and uses the third datum mark to determine the datum plane normal. and Establish a reference coordinate system for the fixture, and then calculate the longitudinal unit axis using the right-hand rule. This represents the horizontal unit axis of the fixture's reference coordinate system and is dimensionless. This indicates that the normal to the fixture reference plane is a unit axis and is dimensionless. , and These represent the three-dimensional coordinates of the centers of the first three reference marks in the nominal fixture coordinate system, in millimeters. Denotes the Euclidean norm. The fourth reference flag is used to verify the coordinate system reproduction residual. In this specific implementation, the upper limit of the reproduction residual is set to 0.010 mm. When the upper limit is exceeded, the current batch of acquisition is stopped and a reference reconstruction record is generated. Before deployment, the camera intrinsic parameters, distortion parameters, and camera extrinsic parameters relative to the fixture reference coordinate system of the binocular reference mark measuring device are obtained using the 3D coordinates of feature points and a dual-target calibration plate for metrological traceability. The reconstructed coordinates of the workpiece reference carrier mark are output to the fixture reference coordinate system. Calibration is performed using at least twenty postures, and verification is conducted using five postures not involved in the solution. The dual-target calibration version is activated when the root mean square error of the reprojection is no greater than 0.20 pixels and the 3D control point residual is no greater than 0.015 mm. When each hardware trigger count is reached, the two global shutter cameras simultaneously acquire the reference mark image. The controller determines the mark number through the mark encoding, initializes with grayscale centroids, and determines the sub-pixel center of each mark using elliptical edge least squares fitting. Left and right image points are matched one-to-one according to the epipolar constraints of the dual-target calibration. The mark center is then reconstructed at the midpoint of the shortest common perpendicular of the two dedistorted imaging rays on the fixture base. The measured three-dimensional coordinates in the quasi-coordinate system; the left and right reprojection errors of a single marker are not greater than 0.30 pixels, the angle between two imaging rays is not less than 5 degrees, and at least three non-collinear markers are obtained at the same time, the marker observation is set as valid; otherwise, an invalid record of the baseline marker observation is generated; the image point center covariance is obtained from the two-dimensional sample covariance of thirty frames of static repeated positioning samples of the same marker, and the bi-target calibration parameter covariance is obtained from the aforementioned multi-attitude calibration residual. The controller propagates the two through the first-order Jacobian of triangulation reconstruction and retains the cross covariance between markers generated by the shared calibration parameters. The units of the obtained marker coordinate covariance and cross covariance are square millimeters; the measured three-dimensional coordinates, covariance, bi-target calibration version, hardware trigger count and microsecond timestamp constitute the baseline marker observation record, and are transmitted to S2 after being associated with the laser image, tilt amount and absolute angle encoder reading according to the same trigger count. When the maximum time difference of any associated record exceeds 2 milliseconds, the moment is marked as invalid attitude input; Both laser measurement units employ insertable side-view laser triangulation heads. The measurement head consists of a fiber optic laser projection channel, a side-viewing steering prism, a telecentric retrospective objective lens, and a surface array camera, and is mounted on a scanning mechanism with axial feed and rotational degrees of freedom around the borehole. The maximum outer diameter of the probe is limited to 60% of the minimum designed borehole diameter of the curved hole to be measured. The calibration working distance covers a range of 0.5 mm to 5 mm from the borehole wall to the outer surface of the probe. The camera acquires grayscale images at 2048 x 1536 pixels at 60 frames per second. The laser uses synchronous pulse exposure. The inclinometer samples the pitch and roll angles 100 times per second. The hardware trigger counter provides microsecond-level acquisition times. The controller groups the image, tilt angle, and reference marker point under the same trigger count into the same sampling record. If the difference between the image time and the most recent tilt angle time exceeds 2 milliseconds, the record is marked as invalid and not sent to S2. Pre-deployment calibration is performed using a calibration component with a dot array and 3D coordinates of feature points that are traceable by metrology. The controller acquires no less than 20 calibration images and laser stripes at six scanning depths and eight different attitudes. The camera intrinsic parameters and radial and tangential distortion parameters are obtained through Zhang Zhengyou calibration. The laser plane parameters are fitted by the 3D points of the light stripe on the known plane, and the calibration results are verified by reprojection with one attitude left. In this specific implementation, the root mean square error of pixel reprojection is required to be no greater than 0.20 pixels and the residual distance of laser plane is no greater than 0.015 mm. After the conditions are met, the camera intrinsic parameter matrix, distortion coefficient, laser plane unit normal, plane intercept, parameter covariance, calibration component number and calibration version number are written into the calibration parameter record of the corresponding measurement unit. The unit of position parameter covariance is square millimeters and the unit of angle parameter covariance is square radians. When acquiring the orifice contour, the controller first turns off the line laser and acquires the ambient light image, then turns on the line laser and acquires the light stripe image. The grayscale value after the difference between the two images is greater than the calibration noise threshold as the candidate pixel of the light stripe. The sub-pixel center is determined by three grayscale centroids along the light stripe normal. The orifice contour points are extracted by Canny edge and closed contour constraints on the orifice edge. The three-dimensional inner wall points in the measurement unit coordinate system are generated by the intersection rule of the camera imaging ray and the calibration laser plane for the inner wall light stripe. Candidate points whose intersection is located behind the camera, exceed the calibration depth range, or whose adjacent sub-pixel center jumps by more than 3 pixels are written into the rejection field and are not included in the valid point array. Before changing the scanning position along the depth direction of each curved hole, the controller reads the curved hole design axis, design inner wall, and probe envelope model. At design axial positions spaced 0.25 mm apart, a segment-by-segment intersection test is performed on the line laser projection ray, the reflection points of the left and right walls to the entrance pupil of the retroreflecting objective, and the probe envelope. Only positions where neither the projection ray nor the retroreflecting cone penetrates the non-target hole wall, the minimum gap between the probe and the design inner wall is not less than 0.10 mm, and at least twelve predictable valid points are formed on each of the left and right walls are written into the optical reachability table. The optical reachability table records the visible wall, gap, number of predicted valid points, calibration version, and test time according to the curved hole identification, design axial position, insertion depth, and incident angle index. If the target interval is within the current incident angle (positive or negative)... If there are no double-wall accessible stations within the 4-degree and fixture-allowed micro-angle range, an optically inaccessible record is formed and transmitted to S5. Repeated rescanning configurations of the same type are prohibited for this interval. For optically accessible stations, sampling is performed incrementally from the aperture inward according to the nominal aperture depth. In this specific embodiment, the initial scan step distance is set to 1.0 mm. Five frames are continuously collected at each position and sorted according to the collection time. The median of the sub-pixel light stripe center of each pixel column is taken. The outlier boundary is three times the absolute deviation of the median, and the flickering points are removed. The retained points are marked as left wall or right wall according to their normal position relative to the aperture contour. If there are fewer than twelve valid points on either side of the same depth layer, the original record is still saved, but the double-wall valid mark is set to false, so that S3 and S5 can form a single-wall coverage anomaly record. The light stripe quality data is stored in association with each inner wall point. The incident angle is obtained by the angle between the calibrated laser ray direction and the estimated normal of the local inner wall, in degrees. The light stripe width is the full width at half maximum of the grayscale profile of the light stripe normal, in pixels. The signal-to-noise ratio is the ratio of the light stripe peak value to the standard deviation of the background noise and is dimensionless. The depth attenuation coefficient is the ratio of the current light stripe peak value to the calibrated peak value at the orifice of the same measurement unit and is limited to zero to one. The number of effective points on the left and right walls is the number of points that have passed the screening on both sides of the corresponding depth layer. The acquisition time is the microsecond timestamp converted by hardware trigger counting. After deployment and calibration, and each time the laser measurement unit is replaced, a reference curved hole part consistent with the material, surface roughness grade, and curved hole type of the workpiece under test is used to establish a candidate light stripe quality response record. The controller establishes an index according to the measurement unit identifier, calibration version, curved hole type, material and surface roughness grade, fixed 0.25 mm depth box, 1 degree incident angle box, and 0.25 mm working distance box. Each index unit collects no less than thirty frames, recording the sample mean, standard uncertainty, sample number, and effective time of the light stripe width ratio, signal-to-noise threshold ratio, depth attenuation, and number of effective points on the left and right walls. The local wall normal at the candidate position is determined by the first-order surface derivative of the inner wall of the curved hole design bound to the workpiece version at the candidate depth and circumferential position. The candidate working distance and occlusion state are determined by the optical reachability table and the fixture digital model. The material, surface roughness grade, and curved hole type must be precisely matched with the response record. For candidates that hit the response domain, Trilinear interpolation is performed on eight valid index cells surrounding the candidate point across three continuous dimensions: depth, incident angle, and working distance. The interpolation order is fixed as depth first, then incident angle, and finally working distance. If any of the eight cells is missing, the candidate exceeds the minimum to maximum value range of any axis, the local wall normal is undeterminable, or the occlusion state is unreachable, the candidate is marked as outside the quality response domain. The standard uncertainty of the interpolated sample is weighted equally on the variance of each index cell and the non-negative square root is taken. Conservative values ​​are used for quality prediction. The light stripe width ratio is taken as the endpoint with the greater deviation from the two endpoints of the predicted mean plus or minus twice the standard uncertainty. The signal-to-noise threshold ratio and depth attenuation are taken as the predicted mean minus twice the standard uncertainty and not lower than zero. The number of valid points on the left and right walls is taken as the predicted mean minus twice the standard uncertainty and rounded down and not lower than zero. The response record version, the number of the eight index cells, and the interpolation weights are written into the candidate prediction record. After each scanning position is completed, the controller matches the laser image, binocular reference mark image, fixture coordinate mark 3D point, workpiece posture mark 3D point, tilt angle and absolute angle encoder reading according to the measurement unit identifier, hole identifier, sampling depth and acquisition time. It filters layer records according to field integrity, sorts them according to sampling depth and acquisition time and writes them into the cache. When duplicate records of the same composite key appear, the record with the larger total number of valid points is selected. When the total number of valid points is the same, the record with the smaller absolute value of timing difference is selected. The controller compares adjacent sampling depths according to the scanning position sequence. If the difference between adjacent depths is not equal to the set step distance and the difference exceeds 0.01 mm, a sampling position jump flag is written. If the target hole depth range has not been covered, the scanning mechanism is controlled to update to the next sampling position to continue sampling. The initial scanning of the curved hole ends when the end of the target hole depth is reached or the motion boundary is triggered. The controller establishes a measurement dataset for each curved hole. The layer record of the dataset includes the scanning batch number, curved hole identifier, measurement unit identifier, sampling depth, scanning step distance, workpiece micro-rotation angle, absolute angle encoder reading, inclination, hole contour point array, left wall point array, right wall point array, light stripe quality field of each point, measured coordinates of fixture coordinate markers, measured coordinates of workpiece posture markers and their joint covariance, dual-target calibration version, candidate light stripe quality response record version, hardware trigger count, acquisition time and valid flag. The original laser image and dual-target reference marker image are stored in the read-only image directory according to the scanning batch number. The layer record, fixture coordinate marker observation record, workpiece posture marker observation record, candidate light stripe quality response record and calibration parameter record are written to the measurement database. After the integrity check is passed, the measurement datasets of the two curved holes and the version number used are passed to S2. Otherwise, the current batch is stopped and the missing field abnormal record is output.

[0022] In this specific embodiment, S2 includes: The coordinate compensation module reads two measurement datasets, corresponding calibration parameter records, and fixture reference coordinate system from S1. First, it establishes a one-to-one correspondence between the measured coordinates of the feature points of the calibration part in the coordinate system of each measurement unit and the nominal coordinates of the fixture reference coordinate system. Then, it uses rigid body least squares with Huber loss to obtain the extrinsic parameters of each measurement unit to the fixture reference coordinate system. The extrinsic parameter record includes the rotation matrix, translation vector, six-dimensional parameter covariance, feature point number used, and calibration version number. The unit of translation parameter covariance is square millimeters, and the unit of rotation parameter covariance is square radians. If there are fewer than six valid corresponding points or the root mean square residual of rigid body registration exceeds 0.020 millimeters, the corresponding measurement unit is marked as having invalid extrinsic parameters, and the coordinate compensation of the curved hole is stopped. For each acquisition moment, a rotational increment vector containing three rotational degrees of freedom is calculated based on the effective measured three-dimensional coordinates, joint covariance, and nominal coordinates fixed to the workpiece reference carrier of at least three non-collinear workpiece attitude markers. Simultaneously, the pitch and roll angles of the inclinometer are combined with the yaw angle of the absolute angle encoder relative to the zero position of this batch to form an auxiliary attitude increment vector containing three rotational degrees of freedom. The absolute angle encoder reading is only considered usable for yaw observation if the hardware trigger counts are consistent, the daily zero-position error is no greater than twice its resolution, and the difference between forward and reverse readings in the same direction is no greater than 0.01 degrees. If the pitch or roll angle of the inclinometer is missing, or if yaw observation is not possible... If an observation is unavailable or the time difference between any of the three (pitch and roll) exceeds 2 milliseconds, the complete auxiliary attitude observation is deemed unavailable and three-degree-of-freedom rotation cannot be generated by zeroing pitch and roll angles. The coordinate compensation module first performs symmetry and Cholesky positive definiteness tests on the three-dimensional covariance of the marker attitude observation and the complete auxiliary attitude observation, and calculates the L2 condition number. Observations that are positive definite and have a condition number not greater than 100 million are determined as available observations. Covariances that fail the test are retested only once after adding a deterministic regularization term to their diagonal, which is the maximum of one hundred millionth of the average diagonal element of the matrix and the square of 10 to the power of negative 12 radians. When both types of observations are available, the information matrix weighting rule applies. When only one of the two complete three-degree-of-freedom observations is available, the observation and its covariance are directly used. When both observations are unavailable after regularization, coordinate compensation at the acquisition time is stopped, an invalid attitude input record is generated and transmitted to S5, and the record is prohibited from entering the center distance qualified or unqualified judgment. This represents the fused rotation increment vector, in radians. This represents the three-degree-of-freedom rotational increment vector obtained from the workpiece's attitude markers, in radians. This represents a three-degree-of-freedom auxiliary attitude increment vector composed of the pitch and roll angles from the inclinometer and the yaw angle from the absolute angle encoder, expressed in radians. The three-dimensional covariance matrix representing the rotation increment of the workpiece attitude marker, in square radians, is obtained by combining the covariance of the measured coordinates of the reference marker through rigid body registration and Jacobian propagation. The three-dimensional covariance matrix representing the auxiliary attitude increment is in radian squares. Its pitch and roll components are obtained from repeated tests at the inclinometer zero position, and its yaw component is obtained from repeated positioning tests in the same direction and in the same direction of the absolute angle encoder and is calibrated daily. The fusion covariance is the inverse matrix of the sum of the available complete observation information matrices, and the source labels and their respective information contributions of the reference mark observation, inclin observation and encoder observation are retained in the record. The coordinate compensation module converts the fused rotation increment vector into a workpiece tilt rotation matrix through exponential mapping. In this clamping translation calculation, both nominal and measured coordinates specifically refer to the workpiece posture markers that move synchronously with the workpiece. Fixture coordinate markers are only used to establish the fixture reference coordinate system and check the binocular reconstruction residuals; they are not included in the workpiece posture or clamping translation solution. Subsequently, the rotation-corrected coordinate residuals are calculated for all visible workpiece posture markers, and then... Determine the clamping translation offset. This indicates the clamping translation offset, in millimeters. This represents the number of workpiece attitude markers that are currently visible and have passed the residual screening; it is an integer not less than three. Indicates the workpiece attitude flag index. Indicates the first The nominal coordinates of each workpiece's posture marker, in millimeters. Indicates the first The measured coordinates of the workpiece's posture markers, in millimeters. The workpiece tilt rotation matrix is ​​dimensionless and the residual exceeds 0.030 mm. The mark is eliminated in one iteration. If there are less than three marks after elimination, the reference mark deficiency is recorded as an anomaly and the process is transferred to the anomaly recovery process of S5. For each orifice contour point and layered inner wall point obtained by the first laser measurement unit or the second laser measurement unit, according to Perform coordinate transformation. This represents the three-dimensional coordinates of the point in the fixture's reference coordinate system after attitude compensation, in millimeters. Indicates the point before compensation at the 1st The three-dimensional coordinates of each laser measurement unit coordinate system, in millimeters. This indicates the laser measurement unit index, and it can be either one or two. Indicates the first The external parameter rotation matrix of the laser measurement unit is dimensionless. Indicates the first The translation vector of the extrinsic parameters of each laser measurement unit is in millimeters. and The aforementioned workpiece tilt rotation matrix and clamping translation offset are used respectively; the coordinate compensation module aims to uniquely map the input point to the fixture reference coordinate system at each sampling time. The input point identifier, the version of the external parameters used, the attitude version, the translation offset, the coordinate transformation action and the output point are written into the transformation record. When the orthogonality error of the rotation matrix exceeds ten to the power of negative six, the input parameters are missing or the output coordinates are not finite, the transformation of that point is stopped and written into the coordinate transformation exception record. For each compensated point, establish a parameter vector and its covariance. Using coordinate transformation, calculate the first-order Jacobian matrix of the point coordinates, extrinsic parameters, fused rotation increment, and clamping translation offset, according to... The propagation yields attitude compensation covariance. This represents the 3D marginal attitude compensation covariance matrix of the compensated points, in square millimeters. The first-order Jacobian matrix represents the coordinate transformation relative to the parameter vector, where the units of each column are determined by the corresponding parameter. This represents the parametric covariance matrix, composed of the covariances of the measurement points, extrinsic parameters, fused rotation increments, and clamping translation offsets. This indicates matrix transpose, and matrix multiplication is output in square millimeters after parameter unit pairing. The coordinate compensation module simultaneously establishes a global random variable registry. The measurement point error uses the point identifier as the unique key, the extrinsic parameter error uses the measurement unit identifier and extrinsic parameter version as the unique key, the tilt error uses the hardware trigger count and attitude fusion version as the unique key, and the translation error uses the hardware trigger count and reference flag solution version as the unique key. Each point records its first-order Jacobian with respect to each unique variable without copying shared variables. The aforementioned three-dimensional marginal attitude compensation covariance is only used for single-point quality gating. S3 to S5 continue to use unique variables, joint covariance, and chained Jacobian to calculate cross-point and cross-section cross covariance. When two measurement units use the same set of reference flag observations or fixture parameters, the corresponding non-zero cross block is retained in the joint covariance. After that, S3 only adds cross-section fitting source variables, and S4 only adds axis fitting source variables, forming six types of fixed source keys: measurement point, cross-section fitting, axis fitting, extrinsic parameter, tilt, and translation. Robust plane fitting is performed on the compensated orifice contour points. The normal of the fitted plane that matches the design entry direction of the curved hole is determined as the positive direction of the hole depth. The directed distance from each compensated inner wall point to the orifice plane is determined as the coarse layer depth. An initial retrieval index is established with depth boxes every 0.25 mm from the orifice and with fixed boundaries. The median depth within the box is used as the coarse layer depth reference. The orifice plane and depth boxes are only used for S3 retrieval of the local neighborhood and are not used as the fixed section basis for the deep section. S3 reconstructs the local normal plane for each layer based on the design axis and the iteration axis. The depth box boundary and layer number are frozen during the initial scan. The rescan step distance is only written to the sampling step distance field without changing the depth box boundary or renumbering. The original sampling depth, the compensated directed depth, and the corresponding depth layer number are written to the curved hole point set at the same time. When the orifice plane fitting residual exceeds 0.025 mm, the generation of a new layer depth reference is stopped and an orifice reference anomaly record is formed. The point record of the curved hole point set includes the compensated three-dimensional coordinates, the attitude compensation covariance in square millimeters, the original light stripe quality data, left and right wall markings, depth layer number, curved hole identification, measurement unit identification, scanning batch number and calibration version number. The coordinate compensation module establishes an index according to the curved hole identification and depth layer number, so that S3 can read the points and covariance in the same layer without changing the point coordinates. The coordinate compensation module filters external parameter records according to the calibration version number and validity time, matches the attitude observation according to the acquisition time, selects the corresponding external parameter according to the measurement unit identifier in the point record, and writes the point into the unmatched record when any match is empty. When the proportion of unmatched points in the same layer exceeds five percent, the update of the depth layer is stopped and the abnormal layer number is passed to S5. The coordinate compensation module sorts the depth reference candidates from smallest to largest according to the residual of the orifice plane. It selects the candidate with the largest number of effective orifice contour points and a residual of no more than 0.025 mm. If the number of candidate points is the same, the one with the earliest acquisition time is selected. The selected orifice plane number and depth box boundary are written into the curved orifice point set head record so that the same layering rule can be reused in the rescanning batch. After the external parameters or attitude fusion parameters are updated, the coordinate compensation module calculates the coordinates of the verification point according to the attitude of the reserved calibration part, compares the residuals before and after the update according to the distance from the verification point to the nominal position, and activates the new parameter version when the root mean square residual after the update does not exceed the verification threshold of 0.020 mm. When the residual exceeds the verification threshold, the previous valid version is retained and the current parameter is marked as verification failure. For the curvature constraint infeasibility anomaly record input by S5, the coordinate compensation module filters the rescanning points according to the curved hole identifier and the depth range to be rescanned, and retains the exclusion flag of the depth range in the curved hole point set until S4 refits to obtain a feasible axis that meets the upper limit of curvature before S4 updates the flag. Before that, the depth range is excluded from the effective measurement section. When the rescanning batch enters this step, the coordinate compensation module still uses the same external parameter version, fusion attitude rules and coordinate transformation relationship as the initial scan. First, it calculates the new attitude compensation covariance in square millimeters for the rescanning points. Then, it appends the original curved hole point set with the curved hole identifier, depth layer number and acquisition time as composite keys. For repeated points with a spatial distance of less than 0.010 mm and the same acquisition time, they are sorted in descending order of signal-to-noise ratio value and the first record is retained. When the rescanning point exceeds the depth range to be rescanned specified in S5, it is written as an out-of-bounds record and does not participate in incremental fusion. After completing the incremental fusion of the rescan points, the coordinate compensation module updates the version of the curved hole point set according to the depth range to be rescanned, selects the affected layer according to the depth layer number and passes it to S3, and then triggers the re-execution of steps S3 to S5. If the affected layer is empty, the original axis and detection status are maintained and the absence of valid rescan points is taken as the reason for stopping this round.

[0023] In this specific embodiment, S3 includes: The cross-section fitting module reads the set of curved hole points generated by S2, extracts the left wall points, right wall points, and light stripe quality fields of each point according to the curved hole identifier and depth layer number, and uses the calibration component of S1 to repeatedly sample data to establish a quality calibration record. The record includes the calibration light stripe width, grayscale noise threshold, background noise standard deviation, threshold signal-to-noise ratio obtained by dividing the grayscale noise threshold by the background noise standard deviation, sensor noise variance, incident angle cosine boundary, width ratio boundary, signal-to-noise threshold ratio boundary, depth attenuation boundary, double wall point number boundary, version number, and validity time for each depth layer. The threshold signal-to-noise ratio and the running signal-to-noise ratio are both dimensionless. The running signal-to-noise ratio is divided by the threshold signal-to-noise ratio of the same depth box to obtain the signal-to-noise threshold ratio, which is also dimensionless. When the threshold signal-to-noise ratio is zero or the background noise standard deviation is zero, the corresponding quality calibration record is marked as invalid. The quality calibration record is generated by at least thirty repeated samplings at each scanning depth. Each sampling generates five primitives: the incident angle cosine, the ratio of the light stripe width to the calibrated light stripe width at the same depth, the signal-to-noise threshold ratio, the depth attenuation coefficient of the current light stripe peak value relative to the orifice calibrated peak value, and the number of effective points on the left and right walls. Each primitive is sorted in ascending order. The 5th percentile and 95th percentile values ​​are only used to remove abnormal samples at both ends and are not directly used as normalization endpoints. The first and last values ​​of the retained samples are sorted again in ascending order to determine the first and last endpoints, respectively. During operation, the incident angle cosine, signal-to-noise threshold ratio, depth attenuation coefficient, and number of effective points on the left and right walls are linearly mapped from the first endpoint to the last endpoint and truncated to zero to one. For width quality, the first and last values ​​of the width ratio in ascending order are retained, and the maximum absolute deviation between the two relative to one is used as the end deviation. The absolute deviation of the running width ratio relative to one is reversed from zero to this end deviation and truncated to zero to one, so that the widths located on both sides of one with the same absolute deviation are... When the same quality factor is obtained from the degree ratio, the width quality is marked as invalid if the calibration width is zero or missing. When the first and last values ​​of any original quantity are equal, or the maximum absolute deviation of the first and last values ​​of the width ratio relative to one is zero, the section fitting module first searches for records with unequal endpoints and valid versions in adjacent fixed depth boxes of the same curved hole type according to the depth distance from near to far, and adopts the first record. When the distances in two adjacent directions are the same, the record with the smaller layer number is adopted. When there are no valid endpoints in all adjacent boxes, the corresponding quality factor of the running value equal to the common endpoint is determined as one. When the running value is not equal to the common endpoint, the quality factor is marked as invalid, the point is excluded, and a zero-span quality calibration abnormal record is formed. When the cosine of the incident angle is larger, the width ratio is closer to one, the signal-to-noise threshold ratio is larger, the depth attenuation coefficient is larger, and the left and right walls are more fully covered, the corresponding quality factor is closer to one. Records that do not hit the curved hole type and the fixed 0.25 mm depth box use linear interpolation of adjacent depth boxes. Points that exceed the calibration depth range are marked as quality invalid. For the The first depth layer For each inner wall point, multiply the incident quality factor, width quality factor, signal-to-noise ratio quality factor, depth quality factor, and double-wall coverage quality factor, and then follow the formula... Calculate the basic weights of the measuring points. This represents the basic weight of the measuring points, expressed in square millimeters. Indicates the depth layer index. Indicates the index of the measurement points within the layer. This represents the incident mass factor, which is dimensionless. This represents the width quality factor, which is dimensionless. This represents the signal-to-noise quality factor and is dimensionless. This represents the depth quality factor and is dimensionless. This indicates a dimensionless quality factor for double-walled coverage. This represents the sensor noise variance obtained from repeated sampling of the calibration sample, expressed in square millimeters. Indicates the first The variance of the residuals of the initial local cross-section model of the layer, in square millimeters; The cross-section fitting module uses the curve after rigid body registration of the curved hole design axis through the S2 orifice point and the entry direction as the initial coarse axis. At each frozen depth layer, the tangent of the coarse axis is obtained at the corresponding design axial station, and a local normal plane passing through that station and perpendicular to the tangent is established. Points are read from the current layer and two adjacent depth boxes, and the unique nearest projection of each point on the coarse axis is calculated. Only points whose absolute difference between the projected arc length and the arc length at the current station is no greater than 0.125 mm are retained and projected onto the local normal plane. The initial local cross-section model is obtained by performing an unweighted algebraic circle fitting on the retained points. If the design file of the curved hole to be measured marks the depth interval as an elliptical cross-section, then direct least squares elliptical fitting with positive major and minor axis constraints is used instead. The cross-section model residual is taken from the measured point to the fitted circle. The orthogonal distance of the circumference or ellipse is used. The residual scale parameter is taken as 1.4826 times the median of the absolute residual and a lower limit of 0.005 mm is set in the same unit. After the center of each layer forms the updated coarse axis, the local tangent, normal plane, point projection and section center are recalculated until the displacement between two adjacent rounds of all centers does not exceed 0.002 mm and the corresponding tangential angle does not exceed 0.1 degrees. A maximum of five rounds are executed. Each round requires that the nearest projection from the point to the coarse axis is unique, the projection arc length increases strictly with the frozen layer number, the tangential angle between adjacent 0.25 mm stations does not exceed 5 degrees and the product of the local curvature and the 0.25 mm box thickness does not exceed 0.087. If any condition is not met or the five rounds do not converge, the output of the center of the affected interval is stopped and the axis structure is not available and is passed to S5. Iterative reweighted least squares is performed using the base weights of the measurement points as initial weights, and robustness coefficients are updated using Huber loss. Forming iterative weights, Indicates the first Layer The iterative weights for each point are expressed in units of one square millimeter. This indicates the basic weight of the aforementioned measurement points. Indicates the first Layer residual scale parameters, in millimeters. Indicates the first Layer The orthogonal residuals of the cross-sectional model at each point are expressed in millimeters. The value represents the absolute value. The robustness coefficient is set to one when the residual is zero. The cross-sectional parameters and residual scale are updated simultaneously in each round. The round ends when the change in the center coordinate is less than 0.001 mm or when the round reaches twenty. The circular cross-section records the center's three-dimensional coordinates, radius, and fitting residuals. The elliptical cross-section records the center's three-dimensional coordinates, major axis, minor axis, in-plane rotation angle, and fitting residuals. The cross-section center is transformed back to the fixture reference coordinate system by the local normal plane base converged in this layer's iteration, and simultaneously records the local axial arc length, local tangent, normal plane number, and corresponding frozen depth layer number. This center is the intersection of the update axis and the local normal plane. If the local normal plane iteration has not converged, it is not allowed to revert to using the borehole plane as the deep cross-section base. If the number of effective points on either side is less than twelve, the output of the cross-section center of this layer is stopped and a single-wall overlay anomaly record is generated. The record includes the missing wall direction, the effective layer number of the adjacent double-wall, the curved hole identifier, the measurement unit identifier, and the suggested depth interval to be scanned. Based on the first-order fitting Jacobian matrix of the parameters to the cross-sectional model and the final iterative weights, according to Determine the central covariance of the source of the cross-section fit. Indicates the first The layer is a three-dimensional central covariance component generated by cross-sectional fitting, with units of square millimeters. Indicates the depth layer index. Indicates the first The layer section model has a dimensionless first-order Jacobian matrix relative to the central parameter. This represents a diagonal matrix consisting of the final iteration weights, in units of millimeters per square millimeter. This indicates matrix transpose; the cross-section fitting module performs chain multiplication on the first-order Jacobian of the coordinates of each point in the same layer and the Jacobian of the point coordinates saved in S2 using the central solution. It saves the sensitivity of each center to the measurement point, extrinsic parameters, tilt, and translation unique bottom-level variables by variable key, and establishes independent variable keys for the cross-section fitting residuals of this layer; when two centers reference the same extrinsic parameter version, attitude trigger count, translation solution version, or common benchmark variable, a non-zero central cross covariance is generated by its sensitivity and the covariance of that unique variable, and the covariances of the two marginal centers are not considered as independent blocks; the center record saves the sensitivity of the six sources, the total marginal central covariance, and the variable keys shared with other centers. If the L2 condition number of the inverse normal equation exceeds one million or is not positive, it is not inverted and the layer is marked as unavailable for cross-section fitting. The cross-section fitting module filters the points to be fitted according to the hole identification and depth layer number, sorts them from largest to smallest according to the basic weight to form iterative input, first deletes invalid quality points and points with attitude compensation covariance eigenvalues ​​exceeding 0.090 square millimeters, then checks the double wall coverage according to the left and right wall markings, and outputs single wall coverage abnormal records and does not fit the cross-section center when there are fewer than twelve points on either side after filtering. For the two candidate cross-section models, circular and elliptical, calculate the Akaike information criterion according to the final weighted sum of squared residuals, and select the model in ascending order of information criterion values. When the difference between the two models is less than 2, select according to the cross-section type of the curved hole to be tested. If the design document does not specify, select the circular model with fewer parameters. Write the model type and selection basis into the cross-section center record. When the rescan point is reached, the cross-section fitting module only recalculates the affected depth layers within the depth range to be scanned specified in S5. The same quality calibration version and the aforementioned weighting rules are used for the origin and the rescan point. The scan batch number of each point is retained in the weight record and robust fitting is performed together. The center results of the two batches are not directly averaged. The recalculated layer and the unrecalculated layer are merged into a new cross-section center sequence according to the depth number. If the sum of the basic weights after the rescan does not exceed the sum of the weights of the same layer before the rescan, the new sequence is retained but the insufficient gain flag is written into the rescan anomaly record. When the axis fitting error is the dominant error component, the scanning step distance is reduced and the sampling position is increased within the depth range to be scanned. The cross-section fitting module still merges and adds new points according to the 0.25 mm depth box boundary and layer number frozen during the initial scan of S2. The rescan step distance is only used as a sampling field for point records and does not generate new layer numbers. When a new point falls into an existing depth box, it is jointly fitted with the origin of that box. When it falls into a depth box without points, the cross-section center is added according to the fixed box number, and the layer is not merged into the adjacent depth box. After each incremental fitting, the cross-section fitting module compares the total center covariance trace of the new center candidates with the previous effective center record in the same layer according to the depth layer number. First, all new candidates with trace values ​​greater than the old record are eliminated. The eliminated records are still saved as rescan evidence but cannot enter S4. When there are no effective centers in the old layer, only new candidates that have passed the positive definiteness, ill-conditioned ratio and double-wall coverage gating in this step are retained. When there are no retained candidates in all affected layers, a cross-section update failure flag is written and passed to S5. The cross-section fitting module only performs sorting on the new center candidates retained after the aforementioned covariance trace gating. First, it sorts them in descending order by the sum of basic weights, then in ascending order by the total center covariance trace. If the first two are the same, it selects the first record as the S4 input in ascending order by the acquisition time of the scanning batch, and saves the unselected records in the candidate history. If the old center in the same layer remains valid and no new candidate passes the gating, the old center is directly used and does not participate in the sorting. For each curved hole, write the cross-section center sequence in ascending order of layer depth. The sequence elements include center coordinates, center covariance, cross-section type, cross-section size, final residual variance, number of effective points on the left and right walls, double-wall coverage quality factor, sum of basic weights, batch source and valid flag. Continuous effective elements and single-wall coverage anomaly records are passed to S4 for use in spatial axis fitting and effective measurement section determination. After the batch is rescanned and this step is completed, repeat steps S3 to S5.

[0024] In this specific embodiment, S4 includes: The axis fitting module reads the cross-sectional center sequence of the two curved holes sorted by depth, normalizes the layer depth of the effective cross-sectional center to curve parameters of zero to one, uses cubic B-splines to represent the spatial axis of the curved holes, sets the internal nodes according to the effective layer depth quantile points, and sets the minimum depth interval between adjacent nodes to twice the initial scan step distance to avoid generating duplicate control points in the same local point cluster. spatial axis according to calculate, Represents the normalization parameter The three-dimensional coordinates of the axis at the location, in millimeters. This represents the curve parameters obtained by linear normalization of layer depth, with values ​​ranging from zero to one. Indicates the control point index. This represents the total number of control points, determined by the number of effective layers and the node interval. Indicates the first It has three cubic B-spline basis functions and is dimensionless. Indicates the first The three-dimensional coordinates of each control point are in millimeters. The initial values ​​of the control points are obtained by weighted average of the covariances of the centers of adjacent sections, in square millimeters. Using the inverse matrix of the central covariance as the data weights for each cross-section center, the control points are solved using data terms with Huber loss and second-order difference penalty terms for adjacent control points. The objective function is... Huber loss is directly defined as In this specific implementation method, ; This represents the axis fitting objective to be minimized, and its dimensions are dimensionless after being weighted by covariance and balanced by penalty coefficients. This represents the total number of centers of the current effective cross-section, and is an integer not less than six. Indicates the center index of the effective section and takes one to the nearest whole number. , This indicates that the aforementioned Huber robust loss function has a dimensionless output. This represents the dimensionless Mahalanobis distance of the input Huber loss. This represents the Huber threshold, calibrated with 95% efficiency under standard normal error, and is dimensionless. Indicates the first The center coordinates of each cross section are given in millimeters. This indicates that its normalized depth parameter is dimensionless. The formula for the axis of the cubic B-spline in the previous section is explicitly adopted. The value at that location, This represents the inverse of its total central covariance matrix, in units of millimeters per square millimeter. This represents the second-order difference penalty coefficient, expressed in units of per square millimeter. and Using the aforementioned control points and total number, the penalty coefficient is determined by the value that minimizes the verification residual in the five-fold, one-left-out fitting of the known crank axis of the calibration part and is written into the axis fitting parameter record; For each control point update, calculate the curvature at the midpoint of each original sampling depth and adjacent depth. , This refers to the curve parameters obtained by linear normalization of the layer depth, with values ​​ranging from zero to one. Indicates parameters The curvature of the spatial curve at a given point, expressed in millimeters. This represents the first-order guiding force of the spatial axis on the curve parameters, expressed in millimeters. This represents the second-order guiding force of the spatial axis on the curve parameters, expressed in millimeters. Represents the cross product of vectors. The upper limit of curvature is calculated by adding manufacturing tolerance to the designed curvature of the hole to be measured. In this specific embodiment, the upper limit of curvature is stored in the record indexed by the hole model and depth range. Control point solutions are only accepted when the curvature of all sampling positions does not exceed the corresponding upper limit. The solver uses sequential quadratic programming and sets an upper limit of fifty iterations. If the candidate curve before constraint exceeds the curvature upper limit in a certain depth interval and the constraint solution has no feasible solution within the upper limit of iterations, then the current scanning step is extended to both sides of the interval to form an abnormal record of curvature constraint infeasibility. The extended interval is excluded from the current effective measurement section, and the hole identifier, measurement unit identifier, maximum over-limit curvature, upper limit of curvature, and depth interval to be scanned are passed to S5. The axis fitting module checks layer by layer three conditions: the number of effective points on both the left and right walls is not less than twelve, the double-wall coverage quality factor is not less than 0.60, and the three characteristic values ​​of the center covariance are not greater than 0.040 square millimeters. Layers that meet the conditions are processed in ascending order of the actual layer depth in the center record. For each pair of adjacent centers, the sampling step distance of the scanning batch that formed the center is read. When the center is jointly fitted by multiple batches, the maximum step distance of each batch participating in the selected center is taken. The maximum step distance corresponding to two adjacent centers is determined as the local effective scanning step distance of the adjacent pair. When the actual layer depth interval does not exceed 1.5 times the local effective scanning step distance, they are merged into the same candidate effective interval. If it exceeds, an abnormal record of missing interval is generated. If it does not cross the interval, it is merged and the interval is used as the depth interval to be scanned. The fixed 0.25 mm depth box number is only used as a retrieval and traceability index and is not used as a condition for center continuity. Therefore, centers formed by the 1.0 mm initial scan and whose actual interval does not exceed 1.5 mm are treated as continuous centers even if the fixed layer number differs by four. The mixed initial scan and rescan data are determined as a unique continuous result according to the aforementioned local effective scanning step distance of adjacent pairs. For each retained continuous candidate interval, its axial length is required to be no less than 60% of the effective length of the curved hole to be measured and to contain at least six cross-section centers. Intervals that meet the conditions are determined as effective measurement cross-sections. When multiple intervals meet the conditions at the same time, they are arranged in descending order of axial length. When the lengths are the same, the interval with the smaller mean value of the center covariance trace is selected. When no effective measurement cross-section is obtained, the candidate axis is still saved but its status is set to not be used for center distance qualification. The axis fitting module uses the coordinates of control points and the endpoints of the effective interval as the solution state, the increment of control points as the iterative action, and the objective function decreases and all curvatures do not exceed the corresponding upper limit as the acceptance condition. In each round, the control points are updated according to the line search results. When the objective function does not decrease, the step size is halved. When the objective function does not decrease for five consecutive rounds, the current round is terminated and a record of non-convergence of the fitting is output. First-order linearization is performed on the control points at the constraint solution. The axis fitting module reads the sensitivity of each center of S3 to the globally unique underlying variable and the shared variable key. First, the joint central covariance containing non-zero cross-center blocks is formed by the two sets of central sensitivities and their covariances of the same variable. Then, the Jacobian of the B-spline basis function is used to propagate to the control points and axis sampling points once. For the related displacements formed by the same extrinsic parameter version, attitude trigger count, translation solution version or common benchmark variable, independent blocks should not be repeatedly established according to the number of sections. When two curved holes share fixtures or marker observations, cross-axis cross blocks are retained. The axis fitting residuals establish axis fitting source variables separately. The control point records save the chain Jacobian, three-dimensional axis marginal covariance and joint cross covariance index for the six unique underlying variables. S5 propagates directly from these unique variables without diagonally splicing the marginal center or marginal axis covariance of the blocks. The spatial axis, effective measurement section, axis covariance and abnormal records of the two curved holes are jointly transmitted to S5.

[0025] In this specific embodiment, S5 includes: The center distance detection module reads the two spatial axes, effective measurement cross section, axis covariance, and abnormal records output by S4. At the same time, it reads the center distance mode, reference cross section, nominal center distance, upper and lower tolerance limits, confidence level, coverage factor, allowed number of rescans, laser measurement unit movement range, fixture allowable rotation angle, and workpiece anti-collision boundary from the detection task configuration. The upper limit of the allowed number of rescans is a non-negative integer frozen at the beginning of the batch and is not greater than ten. The confidence level and coverage factor are also frozen at the beginning of the batch and participate in the configuration consistency verification. When the fixture reference plane is used as a common reference plane, according to Establish the projection matrix. The matrix represents the projection of three-dimensional coordinates onto the fixture reference plane and is dimensionless. Describes a third-order identity matrix that is dimensionless. This indicates that the fixture reference plane established by S1 is normal to the unit axis and is dimensionless. This indicates matrix transpose, where the same projection matrix is ​​used for the coordinates and tangent vectors of both axes. After projection, the axis records retain the original curve parameters, hole identifiers, and axis covariance index. When the reference section mode is selected in the detection task configuration, the center distance detection module calculates the signed distance from the axis point to the reference section in all B-spline node intervals of each effective measurement section of each axis. For intervals with opposite signs of signed distances, the bisection method is used to find the root. For tangent intervals with the same sign but where the absolute value of the signed distance within the interval reaches a local minimum and is not greater than 0.005 mm, the Newton method with boundaries is used to find the root. Both root-finding methods stop when the curve parameter changes by no more than 10 to the power of negative 6. If any axis has no root, an invalid reference section record is generated. If multiple roots exist, they are selected in ascending order of the absolute value of the difference between the physical depth corresponding to the root and the target reference depth configured in the detection task. If the difference is the same to 0.001 mm, the one with the smallest curve parameter value is selected. In the case of multiple roots when the detection task configuration lacks a target reference depth, the same invalid reference section and re-inspection status are generated. Only when each of the two axes obtains a unique valid root and the roots are all located within the corresponding effective measurement section, the intersection point is projected onto a common reference plane and then... Calculate the center distance. This indicates the center-to-center distance in the reference section mode, in millimeters. The coordinates of the intersection point of the spatial axis of the first curved hole and the reference section are shown in millimeters. The coordinates of the intersection point of the spatial axis of the second curved hole and the reference section are shown in millimeters. This represents the dimensionless projection matrix of the aforementioned common reference plane. Represents the Euclidean norm; When the detection task configuration selects the shortest distance mode, within the curve parameter range of the effective measurement section of each of the two projection axes, an initial parameter pair is generated with an axis sampling interval of 0.25 mm. Then, the Gauss-Newton method with boundaries is used to minimize the square distance between the two projection points. The convergence condition is that the distance change between two adjacent rounds is less than 0.0005 mm and the parameters do not exceed the limits. When multiple local minima coexist, the one with the smallest distance is selected. When the distance is the same to 0.001 mm, the parameter pair with the average depth of the two points that is closer to the midpoint of the overlapping interval of the two effective measurement sections is selected. If the convergence is not achieved within 30 rounds, an abnormal record of the shortest distance solution is generated. For the selected center distance calculation mode, a globally unique underlying parameter vector is established using the fixed six source keys from S2 to S4, according to... Propagation center distance variance This represents the standard uncertainty of the center distance, in millimeters. This represents a first-order Jacobian row vector representing the center distance relative to the underlying parameter vector, where the units of its components match the corresponding parameters. This represents a joint parameter covariance matrix consisting of a single variable originating from six sources: measurement point error, cross-section fitting error, axis fitting error, extrinsic parameter error, tilt error, and translation error, and containing cross-blocks of shared variables. The matrix transpose is represented, and the center distance variance is expressed in square millimeters. The standard uncertainty of the center distance is obtained by taking its positive square root. Each extrinsic parameter version, attitude trigger count, translation solution version, and common reference flag variable appears only once in the bottom vector. Each source is propagated sequentially through the S2 point coordinate Jacobian, S3 center solution Jacobian, S4 axis Jacobian, and the center distance Jacobian of this step. When two measurement units or two axes refer to the same variable, the corresponding cross terms are retained. The source label, measurement unit identifier, borehole identifier, depth range, and scanning batch number remain unchanged in the link. The center distance detection module maintains the nominal values ​​of the other five types of underlying source variables and sequentially only perturbs the measurement point, cross-section fitting, axis fitting, extrinsic parameters, tilt, and translation sources to obtain six non-negative first-order uncertainty variance components. Each component records the associated error source type, unique variable key, laser measurement unit identifier, curved hole identifier, depth interval, and scanning batch number. When the relative difference between the six components and the complete first-order propagation variance including cross-blocks exceeds one percent, the current decomposition is marked as inconsistent with covariance accounting and a pending re-inspection status is generated. The first-order propagation also undergoes an independent nonlinear validity test. The module uses the workpiece identifier and scanning batch number hash as fixed seeds to generate 4096 stratified sampling samples from the multivariate normal distribution of the global joint parameter covariance. Each sample is re-performed with full-interval root finding of the reference cross-section or shortest distance optimization with boundaries and the center distance is recorded. When the number of successful samples is less than 95% of the total, a nonlinear propagation invalid record is generated. When the number of successful samples reaches the threshold, the sample variance and the 2.5% and 97.5% quantiles are calculated. When the relative difference between the error and the first-order variance exceeds 5%, the difference between any quantile endpoint and the corresponding endpoint of the first-order confidence interval exceeds 10% of the standard uncertainty of the center distance, the shortest parameter is less than 10 to the power of negative 6 from the effective interval boundary, the difference between the two local minima is not greater than 0.001 mm, or the nominal center distance is not greater than three times the standard uncertainty of the center distance, the first-order linearization is marked as invalid. The median of the successful samples is used as the center distance, the non-negative square root of the variance of the successful samples is used as the nonlinear standard uncertainty, and the two quantile endpoints are used as the confidence interval. At the same time, the result source flag is written as nonlinear sample. After nonlinear substitution occurs, the current center distance standard uncertainty field in the result record is occupied by the nonlinear standard uncertainty. The standard uncertainty obtained by first-order propagation is only retained in the audit field and is prohibited from entering threshold comparison, dominant error triggering, rescan convergence, or final detection output. When substitution is not triggered, the first-order center distance, first-order standard uncertainty, and first-order confidence interval are retained, and the result source flag is written as first-order propagation. The two paths are mutually exclusive. The detection task configuration performs judgments based on the read confidence level and coverage factor. In this specific implementation, the inspection procedure version configures the confidence level to 95% and the coverage factor to 2. At the start of the batch, the controller verifies that both fields must equal 95% and 2 respectively. If either configuration is incorrect, a configuration infeasibility and re-inspection status is generated, and entry into the center distance judgment is prohibited. When linearization is effective, it follows... Generate the center distance confidence interval. This represents the confidence interval for the center distance, in millimeters. This represents the center distance calculated in this step, in millimeters. This indicates the coverage factor verified by the above batch configuration and is dimensionless. This represents the standard uncertainty of the center distance for the first-order propagation path, in millimeters. When a nonlinear validity test triggers substitution, the center distance confidence interval is changed to the endpoints of the 2.5% and 97.5% quantiles corresponding to the validated 95% confidence level, and is no longer superimposed on the linear interval. The current standard uncertainty of the center distance is simultaneously changed to the non-negative square root of the variance of the successful samples. The preset threshold for the center distance uncertainty is the detection tolerance bandwidth divided by six. Linear paths are compared with the first-order standard uncertainty, and nonlinear substitution paths are compared with the nonlinear standard uncertainty. Both use millimeters as the unit and must not be cross-referenced. The status determination is performed only when there is no inconsistency in covariance accounting or invalid nonlinear propagation, the standard uncertainty of the center distance corresponding to the current result source mark is not greater than the preset threshold, and both curved holes have valid measurement sections. When linearization is valid, the first-order center distance, first-order standard uncertainty, and first-order confidence interval are used. When linearization is invalid but the nonlinear sample is valid, only the sample median, nonlinear standard uncertainty, and nonlinear quantile interval are used. When the lower and upper limits of the intervals used are both within the detection tolerance zone, a pass / fail judgment is generated. When the upper limit of the interval is less than the lower tolerance limit or the lower limit of the interval is greater than the upper tolerance limit, a fail / unqualified judgment is generated. Other intersecting cases generate a pending inspection status. When the standard uncertainty of the center distance of the current path is greater than the preset threshold, the axis status is unavailable, there is inconsistency in covariance accounting, or nonlinear propagation is invalid, a pending inspection status is directly generated. When the nonlinear result source mark is valid, the first-order standard uncertainty of the already judged failure should not be read. The anomaly recovery process uniformly merges records from the following categories: S1 (field missing, invalid timing, sampling position jump, motion boundary, and optically inaccessible); S2 (invalid extrinsic parameters, invalid attitude input, orifice reference anomaly, point matching exceeding limits, and no valid rescan point records); S3 (zero-span quality calibration anomaly, invalid quality, single-wall coverage, unusable axis construction, unusable cross-section fitting, insufficient gain, and failed cross-section update records); S4 (missing measurement interval, infeasible curvature constraint, non-converged fitting, and no valid measurement cross-section records); and S5 (invalid reference cross-section, shortest distance solution anomaly, uncertainty exceeding limits, inconsistent covariance accounting, invalid nonlinear propagation, and infeasible configuration records). An anomaly index is established based on measurement unit identifier, orifice identifier, and intersection depth interval. Records with valid calibration and valid attitude are also included. Anomalies in reachable stations and executable scan intervals with optical double walls are marked as recoverable anomalies and enter the candidate configuration process. Anomalies in key fields missing, external parameters invalid, observations in both attitudes unavailable, optical inaccessibility of the target interval, unavailable axis structure, invalid reference section, or no effective measurement section on either of the two curved holes are marked as key input invalid. When key input is invalid, output of qualified or unqualified judgment is immediately prohibited, and data invalid state and final state pending re-inspection are generated deterministically. For single-wall coverage anomalies, the scan interval to be scanned is the closed interval between adjacent effective layers of double walls and the direction of the missing wall is preserved. For missing measurement interval anomalies, the scan interval to be scanned is the interval between adjacent effective layers that exceeds 1.5 times the current scan step. For curvature constraint infeasibility anomalies, the scan interval to be scanned is the interval that S4 has expanded to both sides. The six types of uncertainty variance components of the target curved hole and the depth range to be scanned are arranged in descending order of value. The first component is determined as the dominant error component. When the values ​​are the same and reach one-thousandth of the center distance variance, the first component is determined in the order of measurement point error, cross section fitting error, axis fitting error, external parameter error, tilt error and translation error, so as to ensure that the same input yields the same rescan configuration. When the measurement point error or cross-section fitting error is the dominant error component, candidate combinations of incident angle and scanning depth are generated based on the motion range of the target laser measurement unit, the optical reachability table, and the depth range to be scanned. The candidate angles are first expanded with the current angle as the center at ±2 degrees and ±4 degrees, and then the angles that have not yet been evaluated and are within the allowable range are traversed according to the angular resolution of the measurement unit. The candidate depths are first adopted using the current sampling position, and then the unevaluated reachable positions are traversed from the midpoint of the interval to both ends according to the current scanning step distance. The candidate light stripe quality response records in S1 that match the current measurement unit, calibration version, bore type, material, surface roughness level, and depth range are called. The response domain test and trilinear interpolation are performed according to the local normal of the inner wall of the candidate design, incident angle, working distance, and obstruction status to obtain conservative predicted values ​​of light stripe width ratio, signal-to-noise threshold ratio, depth attenuation, and number of effective points of the double wall with predictive standard uncertainty. The cosine of the incident angle is directly obtained from the candidate geometry. Only candidates with all eight index units and hitting the effective response domain are subject to the quality regression of S3. The prediction weight sum is calculated using the uniformity and measurement point weighting rules. If any response quantity is invalid or outside the response domain, the candidate is removed from the current source set and the outside dimension is recorded. It cannot be replaced by zero value or adjacent category. Candidates whose prediction weight sum increases by at least 10% relative to the actual weight sum of the same interval before rescanning are entered into the priority layer. If the priority layer is not empty, the candidate with the largest weight increment is selected. If the priority layer is empty, the process should not stop. Instead, the candidate with the largest increment is selected from those within the constraints of the entire motion range, optical reachability, collision avoidance, and mass response applicable domain, and whose prediction weight increment is greater than zero. If the increments are the same, the candidate with the smaller standard uncertainty of the prediction weight sum is selected. If they are still the same, the candidate with the smaller absolute value of the change relative to the current incident angle is selected. Finally, the candidate with the scan depth closer to the midpoint of the interval to be rescanned is selected. If the current error source has no positive gain candidate in all allowed discrete configurations or all candidates are outside the mass response domain, the source is only marked as exhausted and the reason is recorded. The candidate configuration selector continues to check the next error source. The process should not stop simply because the 10% threshold has not been reached. When the axis fitting error is the dominant error component, the scanning step size within the depth interval to be supplemented is halved from the current value but not less than 0.25 mm, and a sampling depth layer is added at each end of the interval. When the extrinsic parameter error is the dominant error component, workpiece micro-rotation angle candidates are generated with a step size of 1 degree within the allowable rotation angle of the fixture. Candidates that cross the anti-collision boundary are eliminated using the fixture digital model. For each retained candidate, the current extrinsic parameter version and its six-dimensional covariance are used. The predicted first-order Jacobian of the center distance relative to the extrinsic parameter is recalculated according to the candidate posture, the expected effective measurement section, and the S5 center distance calculation mode. The candidate extrinsic parameter error variance component is then obtained through propagation. Only when the candidate extrinsic parameter error variance component has a reduced determinism relative to the current configuration and at least three workpiece posture markers (excluding fixture coordinate markers) pass the quality gate are included in the extrinsic parameter candidate set. The first item is selected by descending order of the reduction in the variance component of the external parameter error, descending order of the number of workpiece posture markers that have passed the quality gate, and ascending order of the absolute value of the micro-rotation angle. If all candidates within the boundary cannot reduce the variance component of the external parameter error, the external parameter source set is emptied and the reason for the lack of positive gain in the external parameter sensitivity is recorded. The candidate configuration selector is then used to check the next error source. The external parameter rescan configuration should not be selected solely based on the number of fixture coordinate markers or markers that have not passed the quality gate. When tilt error or translation error is the dominant error component, workpiece micro-rotation angle candidates are generated in 1-degree steps within the allowable rotation angle of the fixture, and candidates that cross the anti-collision boundary are eliminated. Then, they are arranged in descending order of the number of visible workpiece posture markers and descending order of the absolute value of the predicted tilt angle change. The first micro-rotation angle that simultaneously makes at least three visible workpiece posture markers and the tilt angle change exceeds five times the resolution of the inclinometer is selected. The candidate configuration selector first establishes candidate sets for six types of errors: measurement point, cross-section fitting, axis fitting, extrinsic parameters, tilt, and translation. It then filters these based on motion range, optical reachability, fixture allowable rotation angle, anti-collision boundary, predicted positive gain, and observation threshold. For measurement point or cross-section fitting errors, candidates are first stratified by whether they reach 10% gain, then sorted in descending order by predicted weight increment. For axis fitting errors, candidates are sorted in ascending order by scan step distance. For tilt and translation errors, candidates are sorted in descending order by the number of visible workpiece attitude markers. Extrinsic parameter errors are not subject to this general sorting and always retain a dedicated sorting: descending order of extrinsic parameter error variance component, descending order of the number of visible workpiece attitude markers, and ascending order of the absolute value of micro-rotation angle. After sorting each set, the inspection starts from the error source with the largest uncertainty variance. When the current source set is empty, the next source is checked sequentially according to the established six-type variance sorting. Each source is checked only once in this round, and the reason for the empty set is recorded. There is no limit to one round. If any executable candidate exists within the boundary of any source, the first item of the first source is selected. If the first check of all six sources is empty and the number of rescans is less than the upper limit of the allowed number of rescans, the interval to be scanned is first extended to both sides by one current scan step distance within the actual motion range and the anti-collision boundary of the fixture, and the angle, depth, step distance and micro-rotation configuration that have not yet been evaluated are regenerated. When the current step distance has reached the lower limit of 0.25 mm, the axis fitting source is changed to add a fixed depth box that has not yet been sampled at the end of the interval. After the extension, it is still checked in the order of the above six categories. Only when the extension touches the actual motion range, optically accessible boundary, fixture allowed rotation angle or anti-collision boundary and all allowed discrete configurations have been evaluated and the six categories are still empty, is the configuration infeasible, uncertainty not met and the final state to be re-inspected deterministically written. The output of qualified or unqualified judgment is prohibited. The process shall not stop because the candidate has not reached 10% gain when the physical boundary has not been exhausted and the number of rescans has not reached the upper limit. When multiple abnormal records correspond to overlapping depth intervals to be scanned, the abnormal recovery process matches intervals according to the hole identifier, updates the range to be scanned according to the interval union, selects the dominant error according to the uncertainty variance component values, matches a candidate measurement configuration according to the dominant error type, and if both holes need to be scanned, the execution order is determined according to the center distance variance contribution from largest to smallest. The candidate measurement configuration record includes the target laser measurement unit, the target curved hole, the depth range to be scanned, the laser incident angle, the scanning depth, the scanning step distance, the workpiece micro-rotation angle, the dominant error type, the prediction weight increment, and the configuration sequence number. Before the configuration is issued, the motion range of the laser measurement unit, the allowable rotation angle of the fixture, and the anti-collision boundary of the workpiece are re-verified. After the verification is passed, the re-scanning data is collected by S1. Then, the re-scanning points are added to the curved hole point set according to the coordinate compensation rule of S2. The affected cross section is recalculated according to the measurement point weight and robust fitting rule of S3. The axis and effective measurement cross section are incrementally updated according to the curvature constraint rule of S4. After the incremental fusion of the rescanned data is completed, the center distance detection module triggers the re-execution of steps S3 to S5, fits the spatial axis according to the updated cross-sectional center sequence, selects the center distance calculation mode according to the detection task configuration, updates the center distance standard uncertainty according to the new covariance record, and associates the results of this round with the results of the previous round according to the number of rescans. After each rescan, the standard uncertainty of the center distance and the six types of uncertainty components of the current path are recalculated based on the result source marker, and the covariance decomposition consistency check and independent nonlinear validity test are re-executed. For nonlinear alternative paths, the non-negative square root of the variance of successful samples is used as the standard uncertainty of the center distance for comparisons between the current and previous rounds and for the stopping threshold. First-order paths use the first-order propagation value. The rescan measurement configuration, number of rescans, result source marker, current standard uncertainty before and after the rescan, changes in the effective measurement cross-section, and the reason for stopping are written into the rescan history. The current path is only considered valid if there are no key input invalidities, infeasible configurations, inconsistent covariance accounting, invalid nonlinear propagation, or other unclosed key anomalies. The process stops and generates a rescan convergence state only when the standard uncertainty of the center distance is no greater than the preset threshold; the process stops and generates an uncertainty failure state and a re-inspection state only when the standard uncertainty of the center distance of the current path is still greater than the preset threshold and the number of rescans reaches the upper limit of the allowed number of rescans, or when the candidate generation has actually reached the motion range, optical reach, fixture allowable rotation angle or anti-collision boundary and exhausted all allowed discrete configurations; otherwise, the process continues to the next round of rescanning according to the new dominant error component; any critical anomaly from S1 to S5 that reaches this step is executed before the confidence interval is determined; if the critical anomaly that has not been recovered is determined to be invalid data or infeasible configuration and awaits re-inspection, the previous qualified or unqualified state must not be used; At the end of the inspection, the center distance detection module outputs the inspection result record, which includes at least the workpiece identifier, two curved hole identifiers, center distance mode, center distance, current center distance standard uncertainty, result source flag, first-order standard uncertainty audit value, nonlinear standard uncertainty audit value, confidence interval, effective measurement section, six types of uncertainty components, inspection status, rescan convergence status, number of rescans, reason for stopping, calibration version, and inspection procedure version. When the result source flag is first-order propagation, the current center distance standard uncertainty is taken as the first-order value. When the result source flag is nonlinear sample, the non-negative square root of the variance of the successful sample is taken, and it is prohibited to replace it with the first-order audit value. The inspection result record and the traceable axis record are written into the inspection database as the final inspection output of the center distance of the curved holes of the workpiece.

[0026] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0027] This invention uses the fixture reference coordinate system as the common coordinate basis for dual-laser measurement data. It sequentially connects the attitude compensation of the orifice contour and layered inner wall points, heteroscedasticity weighted section fitting, curvature constraint axis fitting, and center distance uncertainty propagation, so that the data obtained by different measurement units can enter the center distance calculation along a unified data and covariance chain. This forms a continuous processing relationship for the problems of curved hole axis bending, internal non-penetration, and difficulty in determining the effective measurement section.

[0028] This invention introduces point-by-point confidence fusion before cross-section fitting, robust loss, second-order difference curvature continuity constraint, and curvature upper limit constraint during axis fitting, and adds uncertainty decomposition and adaptive rescan structure after center distance output. This structure selects the corresponding measurement adjustment action based on the dominant error and fuses rescan points according to the same coordinate compensation and weighting rules, ensuring that the detection results have traceable center distance, effective measurement cross-section, and detection status when the uncertainty threshold is met or the stopping condition is reached.

Claims

1. A method for detecting the center distance of curved holes using dual laser references, characterized in that, include: S1. Establish a fixture reference coordinate system using the fixture reference surface and non-collinear reference marks. Use laser measurement units corresponding to the two curved holes to collect the hole contour points, layered inner wall points, light stripe quality data, reference mark points, and tilt angles. S2. Based on the calibration parameters, reference points and tilt angles of the two laser measurement units, determine the measurement coordinate transformation relationship, workpiece tilt rotation matrix, clamping translation offset and attitude compensation covariance, implement attitude compensation for the orifice contour points and layered inner wall points, and determine the layered depth reference from the compensated orifice contour points to obtain the curved hole point set. S3. Determine the measurement point weights of the inner wall points of the layer based on the light stripe quality data, and obtain the cross-sectional center sequence of the two curved holes by robustly weighting the cross-sectional center and the center covariance according to the depth layer. S4. Based on the cross-sectional center sequence and center covariance, fit the space curve under curvature continuity constraint and curvature upper limit constraint to obtain the spatial axis, effective measurement cross section and axis covariance of the two curved holes. S5. Project the two spatial axes onto the same reference plane in the fixture reference coordinate system, calculate the center distance, and propagate the axis covariance and attitude compensation covariance to obtain the center distance uncertainty. When the center distance uncertainty exceeds the preset threshold, adjust the measurement configuration of the corresponding curved hole according to the uncertainty decomposition result to perform a rescan, and incrementally fuse the rescan points according to steps S2 to S4, and output the center distance, center distance uncertainty, effective measurement section and detection status at the stop.

2. The method for detecting the center distance of curved holes using dual laser references according to claim 1, characterized in that, S1 includes: The two curved holes are respectively designated as the first curved hole and the second curved hole; At least three non-collinear reference marks, an inclinometer, a first laser measurement unit, and a second laser measurement unit are set on the reference surface of the fixture, so that the first laser measurement unit corresponds to the first curved hole and the second laser measurement unit corresponds to the second curved hole. Using a calibration component with known geometric dimensions and known feature point coordinates, calibration images and laser beams of the two laser measurement units are acquired respectively to obtain camera intrinsic parameters, distortion parameters, and laser plane parameters. The scanning position is changed along the hole depth direction of each curved hole, and the hole opening contour point or inner wall point, sampling depth, incident angle, beam width, signal-to-noise ratio, depth attenuation coefficient, number of effective points on the left and right walls, and acquisition time of each sampling position are associated and stored to form a measurement dataset for the corresponding curved hole.

3. The method for detecting the center distance of curved holes using dual laser references according to claim 2, characterized in that, S2 includes: Based on the known feature point coordinates of the calibration component and the corresponding feature point coordinates acquired by the two laser measurement units, the extrinsic parameters of the two laser measurement units relative to the fixture reference coordinate system are calculated respectively. The attitude transformation of the reference marker point is calculated based on its measured coordinates and nominal coordinates in the fixture reference coordinate system. The tilt angle attitude transformation is calculated based on the tilt angle, and the marker point attitude transformation and the tilt angle attitude transformation are weighted and fused to obtain the workpiece tilt rotation matrix. The clamping translation offset is determined based on the coordinate residual of the reference marker point after correction by the workpiece tilt rotation matrix. The orifice contour point and the layered inner wall point are transformed sequentially using the extrinsic parameters, the workpiece tilt rotation matrix, and the clamping translation offset. The attitude compensation covariance is determined based on the covariance of each transformation parameter, and the covariance of each transformation parameter and the attitude compensation covariance are associated and stored with the transformed points.

4. The method for detecting the center distance of curved holes using dual laser references according to claim 2, characterized in that, S3 includes: determining the calibration light stripe width, noise threshold, sensor noise variance, and interval normalization boundary based on the repeated sampling results of the calibration component at each scanning depth; determining the initial local cross-sectional model residual variance based on the prefitting residual of the inner wall points at the corresponding depth layer; and performing interval normalization on the cosine value of the incident angle, the ratio of the light stripe width to the calibration light stripe width, the ratio of the signal-to-noise ratio to the noise threshold, the depth attenuation coefficient, and the ratio of the first to the last value of the number of effective points on the left and right walls arranged in ascending order, to obtain the incident quality factor and the width quality factor. The following parameters are considered: signal-to-noise quality factor, depth quality factor, and double-wall coverage quality factor; the product of each quality factor is divided by the sum of the sensor noise variance and the local cross-sectional model residual variance to obtain the measurement point weight of each layer's inner wall point; using the measurement point weight as the basic weight, the weight of the inner wall point whose cross-sectional model residual exceeds the residual scale parameter is iteratively reduced using a robust loss function; a circular or elliptical cross-section is fitted for each depth layer to obtain the center of the cross-section; the center covariance is determined based on the fitted Jacobian matrix, the iterated measurement point weights, and the cross-sectional model residual variance.

5. The method for detecting the center distance of curved holes using dual laser references according to claim 4, characterized in that, S4 includes: using the inverse matrix of the central covariance as the fitting weights for the cross-sectional center sequence, constructing a data fitting term, a robust loss term, and a second-order difference penalty term for adjacent control points for a cubic B-spline curve; obtaining the spatial axis by calculating the control points of the cubic B-spline curve under the condition that the curvature at each sampling position of the curve does not exceed the upper limit of curvature determined by the design curvature and manufacturing tolerance of the borehole to be measured; determining the effective measurement cross-section as the continuous depth interval that satisfies that the number of effective points on both the left and right walls is not less than the preset number of effective points, the double-wall coverage quality factor is not less than the preset coverage threshold, and the eigenvalues ​​of the central covariance are not greater than the preset covariance threshold; and determining the axis covariance based on the central covariance, the fitting weights, and the Jacobian matrix of the cubic B-spline curve with respect to the control points.

6. The method for detecting the center distance of curved holes using dual laser references according to claim 1, characterized in that, S5 includes: selecting the distance between the intersection points of two axes on a predetermined reference section as the center distance according to the detection task configuration, or selecting the shortest distance between two projected spatial axes as the center distance; calculating the Jacobian matrix of the two spatial axes, the extrinsic parameters of the two laser measurement units, the workpiece tilt rotation matrix, and the clamping translation offset based on the center distance; performing covariance propagation on the measurement point error, section fitting error, axis fitting error, extrinsic parameter error, tilt error, and translation error to obtain the center distance uncertainty and the uncertainty components of each error source; and associating each uncertainty component with the laser measurement unit. The system identifies the element, the hole, and the depth range; it determines the preset threshold based on the detection tolerance bandwidth and a preset confidence level, and combines the coverage factor corresponding to the preset confidence level with the center distance uncertainty to generate a confidence interval; it generates a pass / fail judgment only when the center distance uncertainty is not greater than the preset threshold, when the confidence interval is completely within the detection tolerance band, when the confidence interval is separated from the detection tolerance band, when the confidence interval intersects with the detection tolerance band but is not completely within the detection tolerance band, and when the center distance uncertainty is greater than the preset threshold, it generates a re-inspection status.

7. The method for detecting the center distance of curved holes using dual laser references according to claim 6, characterized in that, Step S5 also includes an anomaly recovery process, which receives measurement anomaly records and uncertainty components of each error source, determines the target laser measurement unit, target curved hole, and depth range to be scanned based on the laser measurement unit identifier, curved hole identifier, and depth range associated with the measurement anomaly record or uncertainty component, and generates a candidate measurement configuration containing at least one parameter among laser incident angle, scanning depth, scanning step distance, and workpiece micro-rotation angle; according to the coordinate compensation rule of step S2 and the measurement point weight rule of step S3, the re-scanning points are incrementally merged with the original scanning inner wall points, and steps S3 to S5 are re-executed.

8. The method for detecting the center distance of curved holes using dual laser references according to claim 7, characterized in that, The measurement anomaly records include single-wall coverage anomaly records, missing measurement interval anomaly records, and curvature constraint infeasibility anomaly records; when only one inner wall point on one side of the curved hole is obtained in a depth layer, the output of the cross-sectional center of that depth layer is stopped, and the depth interval to be scanned and the direction of the missing wall are determined according to the adjacent depth layers that satisfy double-wall coverage; when the depth interval between adjacent effective measurement cross sections exceeds the missing measurement interval threshold determined by the scanning step, the depth interval is determined as the depth interval to be scanned. When the curvature of the candidate curve before constraint exceeds the upper limit of curvature determined by the design curvature and manufacturing tolerance of the borehole to be measured in a certain depth range, and a feasible fitting result that satisfies the upper limit of curvature constraint cannot be obtained, the depth range to be supplemented is expanded to both sides along the depth range, and the depth range is excluded from the effective measurement section; each measurement anomaly record is associated with the corresponding laser measurement unit identifier, borehole identifier and depth range to be supplemented.

9. The method for detecting the center distance of curved holes using dual laser references according to claim 8, characterized in that, The uncertainty components of each error source associated with the target laser measurement unit, the target curved hole, and the depth range to be scanned are arranged in descending order of value, and the first component is determined as the dominant error component. When the measurement point error or cross-section fitting error is the dominant error component, for the candidate laser incident angle or scanning depth, the predicted value of the light stripe quality data of each candidate measurement point in the depth range to be scanned is determined according to the calibration parameters of the two laser measurement units and the candidate measurement configuration. The predicted weight of each candidate measurement point is determined according to the measurement point weight rule in step S3 and summed. The candidate laser incident angle or scanning depth that increases the sum of the predicted weights relative to the sum of the measurement point weights in the same depth range before rescanning is selected. When the axis fitting error is the dominant error component, the scanning step distance is reduced in the depth range to be scanned. In addition, a sampling depth layer is added; when the external parameter error, tilt error, or translation error is the dominant error component, a workpiece micro-rotation angle is selected within the allowable rotation angle of the fixture and the anti-collision boundary of the workpiece to increase the number of visible reference marker points and make the tilt angle change measurable; the rescan measurement configuration, the number of rescans, the center distance uncertainty after rescan, and the reason for stopping are recorded; rescanning is stopped and a rescan convergence state is generated when the center distance uncertainty is not greater than the preset threshold; rescanning is stopped only when the center distance uncertainty is greater than the preset threshold, the number of rescans reaches the preset upper limit, the candidate laser measurement configuration exceeds the movement range of the laser measurement unit, or the candidate workpiece micro-rotation angle exceeds the allowable rotation angle of the fixture or the anti-collision boundary of the workpiece, and an uncertainty failure state and a re-inspection state are generated.

10. A dual-laser reference detection system for the center distance of curved holes, used to perform the dual-laser reference detection method for the center distance of curved holes as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to establish the fixture reference coordinate system and acquire the orifice contour points, layered inner wall points, light stripe quality data, reference marker points, and inclination angle of the two curved holes; The coordinate compensation module is used to determine the measurement coordinate transformation relationship, workpiece tilt rotation matrix, clamping translation offset and attitude compensation covariance, and generate the attitude-compensated curved hole point set and layer depth reference. The cross-section fitting module is used to determine the weights of the measurement points and obtain the cross-section center sequence and central covariance. The axis fitting module is used to obtain the spatial axis, effective measurement section, and axis covariance under curvature continuity constraints and curvature upper limit constraints. The center distance detection module is used to project two spatial axes onto the same reference plane, calculate the center distance and center distance uncertainty, adjust the measurement configuration of the corresponding curved hole based on the uncertainty decomposition results, control the incremental fusion of rescan data, and output the center distance, center distance uncertainty, effective measurement section and detection status.

Images