A laser measuring system for measuring the warpage and edge height difference of a thin-walled cover

Abstract

The present application relates to optical geometry measurement and three-dimensional point cloud processing technical field, disclose a kind of thin-wall cover piece warping and edge height difference laser measuring system.The system obtains thin-wall cover piece three-dimensional measurement point cloud using laser scanner and completes unified reference calibration, extracts edge point cloud, analyzes local normal, curvature variation gradient, curvature principal axis and fitting residual;When cumulative normal deflection angle and curvature variation gradient exceed threshold value simultaneously, generate dynamic cutting plane, and according to the previous confirmed cross section and current edge trend, candidate cutting plane is carried out reverse correction and residual update;Smooth cross section profile is obtained by space intersection, and using upper and lower surface point pair, local correction surface, intersection angle difference and cutting residual generate point-by-point projection correction amount, obtain real cross section profile and convert warping and edge height difference.The present application can keep the consistency of continuous edge cross section posture, reduce the projection distortion caused by angle deviation, so that the measurement result corresponds to the edge shape of thin-wall cover piece.

Description

Technical Field

[0001] This invention relates to the fields of optical geometric measurement and three-dimensional point cloud processing technology, specifically to a laser measurement system for the warpage and edge height difference of a thin-walled cover. Background Technology

[0002] Thin-walled covers such as automotive doors, engine hoods, and aircraft skins typically feature free-form surfaces, flanges, rounded corners, and edge steps. Their edge morphology, warpage, and height differences between the upper and lower surfaces affect the fit and assembly. With the application of laser scanning technology in industrial inspection, extracting edge sections from 3D point clouds and measuring warpage and edge height differences has become an important method for evaluating the geometric quality of thin-walled covers.

[0003] Existing technologies typically utilize laser scanners to acquire point clouds of the thin-walled cover surface, determine the cutting plane based on global coordinate direction, preset cutting direction, or current local trend, and then perform spatial intersection, projection, or contour fitting on the point cloud near the cutting plane to obtain the edge cross section; some solutions also adjust the cutting plane based on local normal or curvature changes, and output the cross section dimensions through outlier removal and coordinate conversion.

[0004] However, when the edge of a thin-walled cover extends along a spatial curved surface, the local orientation and curvature change continuously, and the point cloud normal is affected by scanning noise, sparse regions, and abrupt changes in the actual structure. If the cutting posture is determined independently for the current section, directional jumps can easily occur between adjacent sections; even with dynamic cutting, there may still be a residual directional difference between the candidate cutting plane and the current edge trend, causing the point pairs on the upper and lower surfaces to be stretched or compressed, resulting in the highest and lowest points deviating from their true positions, which in turn affects the determination of the warping amount and the edge height difference.

[0005] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a laser measurement system for the warpage and edge height difference of thin-walled cover parts, in order to solve the problems in the prior art where the local orientation and curvature of the edge of the thin-walled cover part changes continuously, the point cloud normal is disturbed by noise, the cutting posture of adjacent sections is prone to change direction jump, and there may still be residual direction difference between the candidate cutting plane and the current edge trend after dynamic cutting, causing the cross-sectional profile to be stretched or compressed, and the highest and lowest points to deviate from the true position, making it difficult to accurately determine the true cross-sectional profile, warpage amount and edge height difference.

[0007] The technical solution of this invention is as follows: a laser measurement system for warpage and edge height difference of a thin-walled cover, comprising: a laser point cloud acquisition and benchmark calibration module: used to acquire a three-dimensional measurement point cloud of the thin-walled cover using a laser scanner, and calibrate it as a benchmark point cloud according to the positioning benchmark features; an edge point cloud extraction module: used to extract the local attitude normal from the neighborhood covariance distribution of the benchmark point cloud to obtain the edge point cloud; a local attitude analysis module: used to fit the micro-surface of the edge point cloud to obtain the normal direction, curvature change gradient, curvature principal axis and fitting residual, and correct the normal direction according to the continuity of adjacent normals; and a dynamic cutting plane generation module: used to obtain a smoothed normal by weighted smoothing of the normal angle, determine the bending point when the cumulative normal deflection angle and curvature change gradient simultaneously exceed the threshold, and determine the bending point based on the smoothed normal and curvature principal axis. The module generates dynamic cutting planes; the section continuous correction and contour acquisition module is used to generate candidate cutting planes based on the previously confirmed section. When the direction difference between the candidate surface normal vector and the trend line vector of the current edge contour exceeds a threshold, the candidate surface normal vector is rotated in the reverse direction. The section cutting parameters are updated based on the residual between the rotated normal vector and the previous section normal vector, so that it intersects with the edge point cloud space to obtain a smooth section contour; the distortion correction and measurement output module is used to establish point pairs on the upper and lower surfaces of the smooth section contour along the local normal. The point pairs with longitudinal differences exceeding the threshold are fitted as local correction surfaces. The intersection angle difference and cutting residual are determined by the spatial intersection relationship between the point pairs and the correction surfaces. Based on this, the point-by-point projection correction amount is generated to obtain the real section contour. The warping amount and edge height difference are converted according to a unified measurement benchmark.

[0008] Optionally, the laser point cloud acquisition and reference calibration module is further configured to: extract the overlapping area point cloud between adjacent scanning zones; remove outliers based on the normal vector of the overlapping area point cloud to obtain a denoised overlapping area point cloud; acquire the positioning reference feature coordinates in the denoised overlapping area point cloud; determine the positioning reference fitting difference and the overlapping point registration difference between adjacent scanning zones respectively; when both the positioning reference fitting difference and the overlapping point registration difference exceed the corresponding threshold, construct a joint correction matrix based on the correspondence between the theoretical and measured positions of the positioning reference feature coordinates and the overlapping point pairs in the denoised overlapping area point cloud; and perform affine transformation and coordinate transformation on the point poses of adjacent scanning zones to generate the reference point cloud.

[0009] Optionally, the edge point cloud extraction module is further configured to: construct a local neighborhood and calculate the covariance matrix based on the spatial coordinates of each reference point in the reference point cloud; extract the minimum eigenvalue and its corresponding eigenvector to form a set of local attitude normal vectors; determine the initial edge point based on the projection distance of each reference point along the corresponding local attitude normal vector, and determine the edge confidence level by combining the proportion of the minimum eigenvalue, the density of neighborhood points, and the deviation of the reference plane; retain the high confidence edge points that meet the edge confidence level conditions, verify the connectivity of adjacent high confidence edge points along the contour direction of the thin-walled cover body, remove isolated points, and generate the edge point cloud.

[0010] Optionally, the local pose resolution module is further configured to: determine neighborhood samples based on the point cloud density, curvature value, and preset scale of the edge points; obtain a local micro-surface using least-squares fitting; and extract the initial normal direction, curvature change gradient, and fitting residual from the local micro-surface; determine normal continuity consistency based on the angle between the initial normal direction and the normal direction of the adjacent region; when the normal continuity consistency does not meet the consistency condition, determine the degree of normal fitting deviation based on the fitting residual, and determine the pose compensation amount in combination with the normal continuity consistency; and correct the initial normal direction in the direction opposite to the normal deviation direction.

[0011] Optionally, the dynamic cutting plane generation module is further configured to: calculate the normal angle between the compensated normal direction and the normal direction of the adjacent region, and remove normal angles whose fitting residuals exceed the residual threshold; allocate smoothing weights according to the normal consistency and point cloud density corresponding to the remaining normal angles to form a continuous smooth normal deflection trend and a corresponding smooth normal; determine the cumulative normal deflection angle according to the smooth normal deflection trend, determine the bending point when the cumulative normal deflection angle and the curvature change gradient both exceed the corresponding threshold, and determine the direction and position of the dynamic cutting plane using the smooth normal and the principal axis of curvature of the bending point.

[0012] Optionally, the cross-section continuous correction and contour acquisition module is further configured to: generate a current candidate cutting plane based on the normal vector and center point coordinates of the previously confirmed cross-section, and extract the three-dimensional contour coordinates of the current edge at the corresponding position on the candidate cutting plane; determine the edge trend line vector based on the three-dimensional contour coordinates, and calculate the spatial direction difference between the edge trend line vector and the candidate surface normal vector; when the spatial direction difference exceeds an angle threshold, perform a reverse rotation on the candidate surface normal vector, and update the cross-section cutting parameters based on the residual between the rotated normal vector and the normal vector of the previously confirmed cross-section.

[0013] Optionally, the cross-section continuous correction and contour acquisition module is further configured to: form a candidate point set based on the spatial intersection result of the cross-section cutting parameters and the edge point cloud; calculate the normal distance, neighborhood point density, and continuity along the edge direction of each candidate point relative to the local main contour; identify outlier candidate points based on the normal distance, the neighborhood point density, and the continuity along the edge direction; and distinguish between real noise points and real abrupt change points at the edge of the thin-walled cover by combining the current cross-section residual and whether there are corresponding points in adjacent confirmed cross-sections; remove the real noise points from the candidate point set and retain the real abrupt change points to obtain the smooth cross-section contour.

[0014] Optionally, the distortion correction and measurement output module is further configured to: acquire the upper surface point set and the lower surface point set of the smooth cross-sectional profile respectively; determine the local normal direction based on the spatial coordinates of the two; and establish corresponding point pairs of the upper and lower surfaces along the local normal direction; when the longitudinal coordinate difference of the corresponding point pairs of the upper and lower surfaces exceeds the shape threshold, determine the projection distortion point pair; perform plane fitting on the projection distortion point pair to obtain a local correction surface; determine the intersection angle difference and cutting residual based on the spatial intersection relationship between the projection distortion point pair and the local correction surface; and generate a point-by-point projection correction amount accordingly to obtain the true cross-sectional profile.

[0015] Optionally, the distortion correction and measurement output module is further used to: determine the highest point of the upper surface, the lowest point of the lower surface, and their spatial positions from the real cross-sectional profile; convert the highest point and the lowest point to the thin-walled cover measurement coordinate system according to the unified measurement benchmark, and determine the warping amount and edge height difference of the corresponding cross-section; arrange adjacent real cross-sectional profiles according to the edge extension direction of the thin-walled cover to form a cross-sectional measurement sequence that reflects the continuous shape of the edge steps.

[0016] Optionally, the distortion correction and measurement output module is further configured to: determine the cross-section verification result based on the normal continuity, projection difference, and height difference changes between adjacent real cross-section profiles; when the cross-section verification result does not meet the verification conditions, call the cross-section cutting parameters saved for the corresponding cross-section, re-execute direction correction, spatial intersection, and point-by-point projection correction to obtain the verified cross-section profile; re-extract the highest and lowest points from the verified cross-section profile, update the warping amount and edge height difference at the corresponding positions, and use the updated results as the final measurement results at the corresponding positions.

[0017] The beneficial effects of this invention are as follows: This invention unifies the 3D measurement point cloud of thin-walled cover parts to the same measurement benchmark through laser point cloud acquisition and benchmark calibration. It obtains continuous edge point clouds, normal direction, curvature gradient, principal axis of curvature, and fitting residuals through edge point cloud extraction and local attitude analysis, providing a local geometric basis for determining the cross-sectional attitude. By dynamically generating cutting planes, it identifies bending points when the cumulative normal deflection angle and curvature gradient simultaneously exceed thresholds. It then uses the previously confirmed cross-section and the current edge trend to perform reverse correction and residual update on candidate cutting planes, suppressing attitude jumps between adjacent cross-sections. Through distortion correction and measurement output, it performs point-by-point projection correction based on upper and lower surface point pairs, local correction surfaces, intersection angle differences, and cutting residuals, reducing cross-sectional stretching or compression caused by residual direction differences and ensuring that the true cross-sectional contour, warpage, and edge height difference correspond to the actual edge shape of the thin-walled cover part. Attached Figure Description

[0018] Figure 1This is a schematic diagram of the overall system structure and data flow provided in one embodiment of the present invention; Figure 2 This is a schematic diagram of laser scanning strip splicing, positioning reference, and unified measurement coordinate calibration provided in one embodiment of the present invention; Figure 3 This is a schematic diagram of edge point cloud extraction and local pose analysis provided in one embodiment of the present invention; Figure 4 This is a schematic diagram of dynamic cutting plane jointly triggered by cumulative normal deflection angle and curvature change gradient according to an embodiment of the present invention. Figure 5 This is a schematic diagram illustrating continuous correction of candidate cutting planes, spatial intersection, and cross-sectional profile acquisition according to an embodiment of the present invention. Figure 6 This is a schematic diagram of point-by-point projection distortion correction, measurement output, and cross-section verification rollback provided in one embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, and not all embodiments. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0020] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0021] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0022] As mentioned earlier, when the edge of a thin-walled cover extends along a spatial curved surface, its local orientation and curvature continuously change with bending, flanging, and rounded corner transitions. The normal of the point cloud formed by laser scanning is also affected by scanning noise, sparse areas of the point cloud, and abrupt changes in the actual structure. If the cutting plane is determined independently based on a fixed direction or the current local trend, attitude jumps are likely to occur between adjacent sections. Even if dynamic cutting is completed, there may still be directional residuals between the candidate cutting plane and the current edge trend, causing the point pairs on the upper and lower surfaces to be stretched or compressed within the section, resulting in the highest point, lowest point, and edge height difference deviating from their true positions.

[0023] To address this, the present invention provides a laser measurement system for the warpage and edge height difference of thin-walled cover parts. A unified measurement benchmark is established through a laser point cloud acquisition and benchmark calibration module. Continuous edge point clouds and their local attitudes are obtained through an edge point cloud extraction module and a local attitude analysis module. A dynamic cutting plane generation module forms a dynamic cutting plane corresponding to the bending position. A continuous cross-section correction and contour acquisition module completes reverse correction, spatial intersection, and preservation of real abrupt changes of candidate cutting planes under the constraint of a previously confirmed cross-section. Finally, a distortion correction and measurement output module completes local correction surface fitting, point-by-point projection correction, conversion of warpage and edge height difference, and cross-section sequence verification.

[0024] The following is combined Figures 1 to 6 This invention is described in detail.

[0025] Example 1: like Figure 1 As shown, this embodiment of the invention provides a laser measurement system for warpage and edge height difference of thin-walled cover parts. The system takes the three-dimensional measurement point cloud of the thin-walled cover part acquired by a laser scanner as input, and sequentially forms a reference point cloud, edge point cloud, local posture record, dynamic cutting plane, smooth cross-sectional profile, real cross-sectional profile, warpage amount and edge height difference within the same measurement task.

[0026] The present invention provides a laser measurement system for warpage and edge height difference of thin-walled cover parts, comprising the following modules: laser point cloud acquisition and benchmark calibration module, edge point cloud extraction module, local attitude analysis module, dynamic cutting plane generation module, cross-section continuous correction and contour acquisition module, and distortion correction and measurement output module.

[0027] Laser point cloud acquisition and benchmark calibration module: used to acquire three-dimensional measurement point clouds of thin-walled cover parts with a laser scanner and calibrate them as benchmark point clouds according to positioning benchmark features; Specifically, the laser point cloud acquisition and benchmark calibration module receives the point spatial coordinates, scan zone identifiers, and valid status output by the laser scanner, and reads the theoretical and measured positions of the positioning benchmark features. The laser point cloud acquisition and benchmark calibration module transforms the point clouds of each scan zone in the same measurement task to a unified measurement coordinate system, forming a benchmark point cloud with benchmark transformation status and scan zone correspondence, and sends the benchmark point cloud to the edge point cloud extraction module.

[0028] Edge point cloud extraction module: used to extract local attitude normals from the neighborhood covariance distribution of the reference point cloud to obtain the edge point cloud; Specifically, the edge point cloud extraction module receives the reference point cloud, takes the local neighborhood of each reference point as the processing object, determines the local attitude normal based on the spatial distribution of the neighborhood points, and combines the projection relationship of the points along the local attitude normal, the local point distribution, and the direction of the entity contour to filter continuous edge points. The resulting edge point cloud corresponds to the position of the entity edge of the thin-walled cover and is sent to the local attitude analysis module.

[0029] Local pose analysis module: used to fit the edge point cloud micro-surface, obtain the normal direction, curvature change gradient, curvature principal axis and fitting residual, and correct the normal direction according to the continuity of adjacent normals; Specifically, the local attitude analysis module receives the edge point cloud, fits a local micro-surface in the neighborhood of each edge point, and obtains the normal direction, curvature gradient, principal axis of curvature, and fitting residual from the fitting result for the current edge position. The local attitude analysis module also verifies the current normal direction based on the continuity of normals between adjacent edge positions. When the continuity of normals does not meet the condition, it generates an attitude compensation amount based on the fitting residual and the direction of normal deviation, obtaining a corrected local attitude record, and sends the local attitude record to the dynamic cutting plane generation module.

[0030] Dynamic cutting plane generation module: used to obtain smooth normals by weighted smoothing of normal angles, determine the bending point when the cumulative normal deflection angle and curvature change gradient both exceed the threshold, and generate dynamic cutting planes based on smooth normals and curvature principal axes. Specifically, the dynamic cutting plane generation module receives the local attitude record, first discards the normal angles that cannot reliably represent the local micro-curved surface, and then smooths the remaining normal angles based on normal consistency and point cloud density to obtain a smooth normal that continuously changes along the extension direction of the thin-walled cover edge. The dynamic cutting plane generation module uses the cumulative normal deflection angle and curvature change gradient to jointly determine whether the current position belongs to a bending position. When both conditions are met, the plane position is fixed by the spatial position of the bending point, and the plane direction is determined by the smooth normal and the principal axis of curvature to obtain the dynamic cutting plane.

[0031] The section continuous correction and contour acquisition module is used to generate candidate cutting planes based on the previously confirmed section. When the direction difference between the candidate surface normal vector and the trend line vector of the current edge contour exceeds the threshold, the candidate surface normal vector is rotated in the reverse direction. The section cutting parameters are updated based on the residual between the rotated normal vector and the previous section normal vector, so that it intersects with the edge point cloud space to obtain a smooth section contour. Specifically, the continuous section correction and contour acquisition module receives the dynamic cutting plane, the current edge 3D contour coordinates, and the normal vector and center point coordinates of the previously confirmed section. For the first section of the measurement sequence, the dynamic cutting plane or the local initial plane serves as the initial state of the candidate cutting plane. For subsequent sections, the continuous section correction and contour acquisition module generates candidate cutting planes based on the previously confirmed section and verifies the candidate surface normal vectors based on the current edge trend line vector. When the spatial direction difference exceeds a threshold, the candidate surface normal vector is rotated in the opposite direction along the direction that reduces the direction difference, and the section cutting parameters are updated based on the residual between the rotated normal vector and the normal vector of the previously confirmed section. The updated section cutting parameters are spatially intersected with the edge point cloud, and a smooth section contour is formed after distinguishing between real noise and real abrupt change points.

[0032] Distortion correction and measurement output module: used to establish point pairs on the upper and lower surfaces of a smooth cross-section profile along the local normal, fit point pairs with longitudinal differences exceeding the threshold as local correction surfaces, determine the intersection angle difference and cutting residual based on the spatial intersection relationship between the point pairs and the correction surfaces, generate point-by-point projection correction amount to obtain the true cross-section profile, and convert the warping amount and edge height difference according to a unified measurement benchmark.

[0033] Specifically, the distortion correction and measurement output module receives the smooth cross-sectional profile, cross-sectional cutting parameters, and a unified measurement benchmark. It establishes a correspondence between upper and lower surface points along the local normal of the smooth cross-sectional profile. For point pairs with longitudinal differences exceeding a threshold, the distortion correction and measurement output module fits a local correction surface and determines the intersection angle difference and cutting residual based on the spatial intersection relationship between the distorted point pairs and the local correction surface. Then, it generates point-by-point projection correction values ​​along the local normal of each point to obtain the true cross-sectional profile. The distortion correction and measurement output module determines the highest and lowest points from the true cross-sectional profile, calculates the warping and edge height difference according to the unified measurement benchmark, and arranges adjacent true cross-sectional profiles into a cross-sectional measurement sequence.

[0034] Among them, the confirmed section refers to the section that has completed candidate cutting plane correction, spatial intersection, point-by-point projection correction and section verification, and can be used as the continuity reference for the next section; the true section profile refers to the profile after the projection distortion points in the smooth section profile have been geometrically corrected point by point based on the local correction surface; the final measurement result refers to the warping amount and edge height difference record that the unified measurement reference is valid and has passed the section verification.

[0035] Therefore, the reference point cloud output by the laser point cloud acquisition and reference calibration module serves as the input to the edge point cloud extraction module, the edge point cloud output by the edge point cloud extraction module serves as the input to the local attitude analysis module, the local attitude record output by the local attitude analysis module serves as the input to the dynamic cutting plane generation module, the dynamic cutting plane and the previously confirmed cross-section state serve as the input to the continuous cross-section correction and contour acquisition module, and the smooth cross-section contour serves as the input to the distortion correction and measurement output module. When the cross-section verification does not meet the conditions, the distortion correction and measurement output module sends the cross-section cutting parameters and backtracking status saved for the corresponding cross-section to the continuous cross-section correction and contour acquisition module, so that the corresponding cross-section re-performs direction correction, spatial intersection, and point-by-point projection correction.

[0036] Based on the above system, the present invention enables the current cross section to maintain its continuity with the previously confirmed cross section while following the local edge trend changes of the thin-walled cover. It also converts the angular distortion that still exists after dynamic cutting into point-by-point geometric correction based on the local correction surface. Only the real cross section profile that has been corrected and verified enters the final measurement sequence or serves as the reference for the next cross section, thereby forming a closed-loop system processing from solid laser point cloud input, cross section attitude determination, profile correction, physical quantity output to verification and rollback.

[0037] Example 2: To provide a clearer and more complete explanation of the above embodiments, the present invention also provides Embodiment Two.

[0038] In this second embodiment, each module continuously performs measurement tasks along the same thin-walled cover edge, and the data and status generated by the previous module are directly used as input for the next module. The number of scanning bands, the overlap width of adjacent scanning bands, the point spacing, and the number of positioning reference features are determined by the width of the measured area, the effective field of view of the laser scanner, the spatial resolution, and the arrangement of positioning references (for example, using 4 scanning bands to cover the same edge area, the overlap width of adjacent scanning bands is 20mm, the point spacing is 0.2mm, and 3 positioning reference points are used).

[0039] like Figure 2 As shown, firstly, the laser point cloud acquisition and benchmark calibration module establishes the correspondence between point clouds in each scanning zone and performs unified measurement benchmark calibration. The laser point cloud acquisition and benchmark calibration module is further used for: Extract the point cloud of the overlapping area between adjacent scan bands, and remove outliers based on the normal vector of the overlapping point cloud to obtain the denoised overlapping point cloud; Obtain the positioning reference feature coordinates in the point cloud of the denoised overlapping area, and determine the positioning reference fitting difference and the registration difference of the overlapping points of adjacent scan bands respectively; When both the fitting difference of the positioning reference and the registration difference of the overlapping points exceed the corresponding thresholds, a joint correction matrix is ​​constructed based on the correspondence between the theoretical and measured positions of the feature coordinates of the positioning reference and the overlapping point pairs in the denoised overlapping area point cloud. Affine transformation and coordinate transformation are then performed on the poses of the points in adjacent scanning zones to generate the reference point cloud.

[0040] Specifically, the laser point cloud acquisition and benchmark calibration module extracts the point cloud of the overlapping area between adjacent scan bands based on the scan band identifier and spatial proximity, and establishes a local neighborhood within the overlapping area. The normal vector of the overlapping area point cloud can be obtained through covariance matrix eigenvalue decomposition, and points that deviate from the main normal of the overlapping area and lack neighborhood support are identified as outliers. The positioning benchmark feature can be a fixture positioning hole, a benchmark mark, or other benchmark features corresponding to the fixing of the cover being measured.

[0041] The positioning datum fitting difference threshold and the overlapping point registration difference threshold are jointly determined by the repeated measurement results of the datum, the allowable clamping error, and the repeatability of the scanning band (for example, the positioning datum fitting difference threshold is 0.10 mm, and the overlapping point registration difference threshold is 0.15 mm). In this round of measurement, the laser point cloud acquisition and datum calibration module calculated the positioning datum fitting difference to be 0.12 mm and the overlapping point registration difference to be 0.18 mm, both of which exceed the corresponding thresholds. Therefore, a joint correction matrix is ​​constructed based on the correspondence of positioning datum features and the overlapping point pairs, and affine transformation and coordinate transformation are performed on the point poses of adjacent scanning bands.

[0042] After joint correction, the laser point cloud acquisition and benchmark calibration modules recalculate the two types of errors. The positioning benchmark fitting difference is reduced to 0.06mm and the overlapping point registration difference is reduced to 0.09mm, both meeting the corresponding benchmark validity conditions. Therefore, a benchmark point cloud with unified measurement coordinates, scan band markings, and benchmark transformation version is generated and sent to the edge point cloud extraction module. If the positioning benchmark features are missing, the effective points in the overlapping area are insufficient, the joint correction matrix is ​​irreversible, or any error after correction still exceeds the corresponding threshold, then the point cloud in this round will not be included in the subsequent physical quantity conversion.

[0043] For example, after filtering the overlapping areas of the second and third scan bands among the four scan bands using the normal direction, valid corresponding points are retained. The laser point cloud acquisition and benchmark calibration module can use iterative nearest point registration to determine the corresponding point residuals, and write the positioning benchmark feature constraints and overlapping point constraints into the joint correction matrix. Iterative nearest point registration is only used to obtain the overlapping point registration difference and does not change the processing relationship of the joint correction after both the positioning benchmark fitting difference and the overlapping point registration difference exceed the limit.

[0044] After completing the unified measurement benchmark calibration, the edge point cloud extraction module extracts edge points from the benchmark point cloud that continuously correspond to the contour of the thin-walled cover body. For example... Figure 3As shown, the edge point cloud extraction module is further used for: Based on the spatial coordinates of each reference point in the reference point cloud, a local neighborhood is constructed and the covariance matrix is ​​calculated. The minimum eigenvalue and its corresponding eigenvector are extracted to form a set of local attitude normal vectors. The initial edge points are determined by the projection distance of each reference point along the corresponding local attitude normal vector, and the edge confidence is determined by combining the proportion of minimum eigenvalue, the density of neighboring points and the deviation of the reference plane. High-confidence edge points that meet the edge confidence criteria are retained. The connectivity of adjacent high-confidence edge points is verified along the contour direction of the thin-walled cover body. Isolated points are eliminated to generate the edge point cloud.

[0045] Specifically, the local neighborhood scale is determined by the laser point spacing, the local point cloud density, and the minimum edge curvature scale to be identified (e.g., a local neighborhood scale of 2.0 mm, which is 10 times the example point spacing of 0.2 mm). The edge point cloud extraction module performs eigenvalue decomposition on the local neighborhood covariance matrix, uses the eigenvector corresponding to the minimum eigenvalue as the local pose normal vector, and calculates the projection distance of each reference point along the corresponding local pose normal vector. The projection distance condition is determined by the distance distribution between the edge region of the reference part and the main body surface (e.g., a projection distance threshold of 0.30 mm).

[0046] Edge confidence is determined by the proportion of minimum eigenvalues, neighborhood point density, and reference plane deviation. The edge confidence input conditions are determined by the normalized statistical results corresponding to the confirmed edge samples and device resolution (e.g., an edge confidence threshold of 0.70, dimensionless). Connectivity conditions are determined by point spacing, local curvature, and the continuous scale of the entity contour (e.g., the maximum spatial spacing between adjacent high-confidence edge points is 0.60 mm, and the directional angle does not exceed 12°). The edge point cloud extraction module retains only high-confidence edge points that simultaneously satisfy both edge confidence and connectivity conditions, and removes isolated points, short chains with insufficient length, and degenerate neighborhood points.

[0047] For example, within an edge region of the reference point cloud, a candidate point has a projection distance of 0.42 mm along the local attitude normal vector, an edge confidence of 0.82, and spatial distances to the preceding and following high-confidence edge points of 0.38 mm and 0.41 mm, respectively, with directional angles of less than 12°. Therefore, this candidate point is retained and added to the continuous edge chain. Another candidate point, although having an edge confidence of 0.76, has a spatial distance of 0.95 mm to the nearest high-confidence edge point and is therefore discarded as an isolated point because it does not meet the connectivity condition.

[0048] After the edge point cloud is formed, the local pose analysis module fits a local micro-surface within the neighborhood of each effective edge point and corrects the normal direction. The local pose analysis module is further used for: Based on the point cloud density, curvature value and preset scale of the edge points, the neighborhood samples are determined, the local micro-surface is obtained by least squares fitting, and the initial normal direction, curvature change gradient and fitting residual are extracted from the local micro-surface. The continuity of the normal direction is determined based on the angle between the initial normal direction and the normal direction of the adjacent region. When the continuity of the normal direction does not meet the consistency condition, the degree of deviation of the normal fitting is determined based on the fitting residual, and the attitude compensation amount is determined in combination with the continuity of the normal direction. The initial normal direction is then corrected in the direction opposite to the deviation of the normal direction.

[0049] Specifically, the local pose analysis module uses the current edge point as the local coordinate origin, selects neighborhood samples based on the aforementioned point cloud density, curvature value, and local neighborhood scale, and fits a quadratic microsurface using the least squares method. The initial normal direction is determined by the first-order partial derivative of the local microsurface at the current edge point, the curvature change gradient and the principal axis of curvature are determined by the second-order term of the local microsurface along the edge arc length direction, and the fitting residual is determined by statistically analyzing the distances from the neighborhood samples to the local microsurface.

[0050] The normal continuity consistency condition and attitude compensation upper limit are determined by the adjacent normal fluctuations in the smoothed edge region of the reference component, the point cloud noise upper limit, and the allowable cross-sectional attitude change (for example, the normal continuity consistency condition is that the angle between adjacent normals does not exceed 10°, the fitting residual threshold is 0.08mm, and the single-point attitude compensation upper limit is 2.0°). When the angle between the initial normal direction and the normal direction of the adjacent region exceeds 10°, the local attitude analysis module determines the fitting deviation of the current normal based on the fitting residual and applies an attitude compensation amount of no more than 2.0° in the direction opposite to the normal deviation direction; when the fitting residual exceeds 0.08mm, the current normal is marked as a low-confidence normal and does not participate in subsequent normal angle smoothing.

[0051] For example, if the angle between the initial normal direction of an edge point and the normal direction of the adjacent region is 11.5°, and the corresponding fitting residual is 0.06mm, the local attitude analysis module determines a reverse attitude compensation amount of 1.5° based on the fitting residual and the continuity of the normal direction, so that the angle between the compensated normal direction and the normal direction of the adjacent region is reduced to 10.0°, and sends the compensated normal direction, curvature change gradient, curvature principal axis and fitting residual to the dynamic cutting plane generation module.

[0052] Subsequently, the dynamic cutting plane generation module smooths the compensated normal direction and determines the trigger position, direction, and location of the dynamic cutting plane using the cumulative normal deflection angle and curvature change gradient. For example... Figure 4 As shown, the dynamic cutting plane generation module is further used for: Calculate the normal angle between the compensated normal direction and the normal direction of the adjacent region, and remove normal angles whose fitting residual exceeds the residual threshold; Smoothing weights are assigned based on the normal consistency and point cloud density corresponding to the remaining normal angles, forming a continuous smooth normal deflection trend and the corresponding smooth normal. The cumulative normal deflection angle is determined based on the smooth normal deflection trend. When the cumulative normal deflection angle and the curvature change gradient both exceed the corresponding threshold, the bending point is determined. The direction and position of the dynamic cutting plane are determined by the smooth normal and the principal axis of curvature of the bending point.

[0053] Specifically, the dynamic cutting plane generation module includes normal angles with fitting residuals not exceeding the aforementioned 0.08 mm into the effective sample set. The normal consistency in the smoothing weights is derived from the cosine of the angle between the compensated normal direction and the normal direction of adjacent regions, and the normalized point cloud density is derived from the normalization result of the current neighboring point count relative to the effective neighboring point count range of the same edge chain. The smoothing weights are determined by dividing the product of normal consistency and normalized point cloud density by the sum of the fitting residual and the numerical stability. The smoothing weight index and numerical stability are determined by the weight normalization rule and floating-point calculation precision (for example, both smoothing weight indices are 1.0, dimensionless, and the numerical stability is 0.01 mm). The smoothed normal is determined according to the following formula: In equation (1), The set of valid samples whose fitting residuals do not exceed the fitting residual threshold within the current window; For the first The unit normal vector of each valid sample, dimensionless; The smoothing weights are based on normality, normalized point cloud density, and fitting residuals, and are expressed in mm. -1 ; The smooth normal is the output of the current window and is dimensionless. The smoothing weight is determined by the weight normalization rule. The higher the normal consistency and point cloud density, and the smaller the fitting residual, the greater the corresponding smoothing weight. The numerator and its vector mold of equation (1) have the same dimension. After normalization, the output smooth normal is still a dimensionless unit vector.

[0054] For example, the unit normal vectors of the three valid samples are (0.000, 0.100, 0.995), (0.020, 0.110, 0.994), and (0.040, 0.120, 0.992), with corresponding normal consistency values ​​of 0.99, 0.98, and 0.97, normalized point cloud densities of 0.95, 1.00, and 0.90, and fitting residuals of 0.04 mm, 0.05 mm, and 0.06 mm, respectively. The smoothing weights obtained according to the aforementioned weight formation rule are 18.81, 16.33, and 12.47, respectively, in mm. -1 Substituting the smoothing weights and unit normals into equation (1), the smoothed normal of the current window is approximately (0.017, 0.109, 0.994). This smoothed normal is then used for calculating the cumulative normal deflection angle and for dynamic cutting plane orientation.

[0055] The cumulative window length, cumulative deflection angle threshold, and curvature change gradient threshold are determined by the edge sampling interval, normal edge normal fluctuation, statistics of straight segments and gentle curves, and the minimum bending scale to be identified (e.g., a cumulative window length of 5 edge points, a cumulative deflection angle threshold of 5.0°, and a curvature change gradient threshold of 0.030 mm). -2 The dynamic cutting plane generation module determines the bending trigger state according to the following formula: In equation (2), For the first The bending trigger state at each edge position is dimensionless. For conditional indicator functions; The cumulative value of the angle between adjacent smooth normals within the current cumulative window, in degrees, is output by equation (1); This is the absolute value of the ratio of the cumulative curvature change within the current window to the difference in the corresponding edge arc length, in mm. -2 ; This is the cumulative normal deflection angle threshold, in degrees. This is the gradient threshold for curvature change, in mm. -2 When equation (2) outputs the trigger state, the dynamic cutting plane generation module determines the plane position using the spatial coordinates of the bending point and the plane direction using the geometric relationship between the smooth normal and the principal axis of curvature. For the first section of the measurement sequence, when there is no previous confirmed section, the position of the local initial plane is determined by the coordinates of the center edge point of the first effective attitude window, and the direction of the local initial plane is determined by the smooth normal and the principal axis of curvature of that window; for subsequent sections, when equation (2) is not triggered or there are insufficient effective samples, the cutting plane state of the previous confirmed section is maintained, and no new bending update is generated.

[0056] For example, in the aforementioned five edge point windows, the cumulative normal deflection angle is 6.8°, and the curvature gradient is 0.045mm. -2 The two exceeded 5.0° and 0.030mm respectively. -2 Therefore, the bending trigger state of equation (2) is set to 1, and the current position is determined as the bending point. If the current smooth normal is (0.017, 0.109, 0.994) and the principal axis of curvature is (1.000, 0.000, -0.017), then the dynamic cutting plane generation module obtains the dynamic cutting plane normal as approximately (-0.002, 0.994, -0.109) based on the geometric relationship between the two, and determines the position of the dynamic cutting plane using the spatial coordinates of the bending point.

[0057] After obtaining the dynamic cutting plane, the cross-section continuous correction and contour acquisition module continuously corrects the candidate cutting plane based on the previously confirmed cross-section and the current edge trend. For example... Figure 5 As shown, the cross-section continuous correction and contour acquisition module is further used for: The current candidate cutting plane is generated based on the normal vector and center point coordinates of the previously confirmed cross section, and the three-dimensional contour coordinates of the current edge at the corresponding position on the candidate cutting plane are extracted. The edge trend line vector is determined based on the three-dimensional contour coordinates, and the spatial direction difference between the edge trend line vector and the candidate surface normal vector is calculated. When the spatial orientation difference exceeds the angle threshold, the candidate surface normal vector is rotated in reverse, and the section cutting parameters are updated based on the residual between the rotated normal vector and the normal vector of the previously confirmed section.

[0058] Specifically, the first cross-section of the measurement sequence is initialized by the current dynamic cutting plane; the candidate cutting planes for subsequent cross-sections are formed by extending the normal vector and center point coordinates of the previously confirmed cross-section towards the current sampling position. The center point coordinates of the previously confirmed cross-section are derived from the geometric center of the point set of that cross-section (e.g., the center point coordinates of the previously confirmed cross-section are (50.00mm, 18.00mm, 2.00mm)). The current edge trend line vector can be determined by the local principal direction of the three-dimensional contour coordinates of the candidate position, the local curve tangent, or the difference between adjacent edge points.

[0059] The angle threshold, maximum single-time correction amount, and continuity fusion coefficient are determined by statistical analysis of the direction difference between adjacent confirmed cross-sections, scanning noise, cross-section spacing, and allowable cross-section pose changes (e.g., angle threshold of 3.0°, maximum single-time correction amount of 1.5°, continuity fusion coefficient of 0.30, dimensionless). When the direction difference between the candidate surface normal vector and the current edge trend line vector exceeds the angle threshold, the cross-section continuous correction and contour acquisition module determines the reverse rotation direction based on the deviation direction and updates the cross-section normal according to the following formula: In equation (3), This is the unit normal vector of the current candidate surface, which is dimensionless. It is the unit normal vector of the previously confirmed cross section, and is dimensionless. This is the reverse rotation angle, in degrees, which is determined by the portion of the current spatial orientation difference that exceeds the angle threshold and is limited by the maximum single correction amount. For rotation operators; The rotation axis is a unit vector, dimensionless; it is obtained by normalizing the cross product direction of the current candidate face unit normal vector and the current edge trend line unit vector. The sign for deviation direction is dimensionless; The continuity fusion coefficient is dimensionless. This is the updated unit normal vector of the cross section, which is dimensionless.

[0060] For example, the spatial direction difference between the candidate surface normal vector and the current edge trend line vector is 4.2°, exceeding the angle threshold of 3.0°. Therefore, the reverse rotation angle is 1.2°, which does not exceed the maximum single correction amount of 1.5°. Substituting the reverse-rotated normal vector and the previously confirmed cross-section normal vector into equation (3) with a continuity fusion coefficient of 0.30, the updated cross-section normal vector is approximately (-0.002, 0.996, -0.093). The cross-section continuous correction and contour acquisition module saves this normal vector, the current cross-section center point, and the parameter version together as cross-section cutting parameters.

[0061] After the cross-section cutting parameters are determined, the continuous cross-section correction and contour acquisition module intersects the corresponding spatial plane with the edge point cloud, and distinguishes between real noise points and real abrupt change points by combining the information of the current cross-section and adjacent cross-sections. The continuous cross-section correction and contour acquisition module is further used for: A set of candidate points is formed based on the spatial intersection results of the cross-sectional cutting parameters and the edge point cloud. The normal distance, neighborhood point density and continuity along the edge direction of each candidate point relative to the local main contour are calculated. Outlier candidate points are identified based on the normal distance, the neighborhood point density, and the continuity along the edge direction. In combination with the current cross-sectional residual and whether there are corresponding points in the adjacent confirmed cross-sections, real noise points are distinguished from real abrupt change points at the edge of the thin-walled cover. The smooth cross-sectional profile is obtained by removing the real noise points from the candidate point set and retaining the real abrupt change points.

[0062] Specifically, the intersection tolerance is determined by the laser point spacing, scanning accuracy, and cross-sectional point continuity requirements (e.g., the intersection tolerance is 0.25 mm). The cross-sectional continuous correction and contour acquisition module selects edge points whose distance to the updated cutting plane does not exceed the intersection tolerance to form a candidate point set, and calculates the normal distance, neighborhood point density, and continuity along the edge direction of each candidate point relative to the local main contour. The current cross-sectional residual is obtained by excluding the current outlier candidate point from the current cross-sectional candidate point set, fitting the local main contour with the remaining candidate points, and calculating the absolute value of the distance from the outlier candidate point along the normal of the local main contour to the local main contour. The upper limit of the current cross-sectional residual judgment and the search range of corresponding points of adjacent cross-sections are determined by the intersection tolerance, point spacing, and adjacent cross-section spacing (e.g., the upper limit of the current cross-sectional residual judgment is 0.30 mm, and the search radius of corresponding points of adjacent cross-sections is 0.40 mm).

[0063] Normal distance, neighborhood point density, and continuity along the edge direction are initially used to generate outlier candidate points. For outlier candidate points, the cross-section continuous correction and contour acquisition module further checks whether there are corresponding points in the current cross-section residual determined in the aforementioned manner and in adjacent confirmed cross-sections: candidate points whose current cross-section residual exceeds the judgment upper limit and for which there are no corresponding points in adjacent confirmed cross-sections are determined as real noise points; candidate points for which there are corresponding points in adjacent confirmed cross-sections, or for which a continuous structure is formed along the edge direction of the thin-walled cover, are determined as real abrupt change points. Local smoothing is performed only within the range that does not cross the real abrupt change point label.

[0064] For example, in the candidate point set formed by this round of spatial intersection, there are three outlier candidate points. Two of these candidate points have corresponding points within a 0.40mm search radius at both the previous confirmed cross-section and the next effective cross-section, and are continuously distributed along the edge direction. Therefore, they are retained as true abrupt change points at the edge of the thin-walled cover. The other candidate point has a current cross-section residual of 0.35mm, and no corresponding point exists in the adjacent confirmed cross-section. Therefore, it is discarded as a true noise point. The smooth cross-sectional profile obtained after processing retains the true structural changes of edge steps and rounded corner transitions.

[0065] After the smooth cross-sectional profile is formed, the distortion correction and measurement output module establishes corresponding point pairs on the upper and lower surfaces along the local normal, and performs point-by-point projection correction on the projected distortion points. For example... Figure 6 As shown, the distortion correction and measurement output module is further used for: Obtain the upper surface point set and lower surface point set of the smooth cross-sectional profile respectively, determine the local normal direction based on the spatial coordinates of the two, and establish corresponding point pairs of the upper and lower surfaces along the local normal direction; When the longitudinal coordinate difference between corresponding point pairs on the upper and lower surfaces exceeds the morphological threshold, a projection distortion point pair is determined. Plane fitting is performed on the projection distortion point pairs to obtain a local correction surface. The intersection angle difference and cutting residual are determined according to the spatial intersection relationship between the projection distortion point pairs and the local correction surface, and a point-by-point projection correction amount is generated accordingly to obtain the true cross-sectional profile.

[0066] Specifically, the distortion correction and measurement output module divides the contour points into upper surface point sets and lower surface point sets based on the local surface orientation of each point in the smooth cross-sectional contour, and searches for the nearest valid corresponding point in the lower surface point set along the local normal of each upper surface point. The morphological threshold is jointly determined by the height difference tolerance of the thin-walled cover design, the longitudinal difference fluctuation of the normal cross-section, and the repeatability of laser measurement (for example, the morphological threshold is 0.20 mm). The morphological threshold represents the upper limit of the allowable longitudinal difference between corresponding point pairs on the normal upper and lower surfaces. Point pairs with longitudinal coordinate differences exceeding this upper limit enter the projection distortion correction process; this morphological threshold is not used as the threshold for determining whether the thin-walled cover product is qualified.

[0067] For corresponding point pairs on the upper and lower surfaces with a longitudinal coordinate difference exceeding 0.20 mm, the distortion correction and measurement output module fits a local correction surface using the projected distortion point pair and its neighboring points. The numerical stability condition and maximum correction amount are determined by the line-surface intersection angle tolerance, floating-point calculation accuracy, point spacing, and allowable local geometric correction range (for example, the lower limit of the absolute value of the dot product of the local correction surface normal and the point's local normal is 0.10, dimensionless, and the maximum correction amount is 0.50 mm). The point-by-point projection correction amount for each projected distortion point is determined according to the following formula: In equation (4), The unit normal vector of the local correction surface is dimensionless. This is a constant term for the local correction surface, in mm; For the first The spatial coordinates of the projection distortion points are in mm. Let be the local normal unit vector at that point, which is dimensionless; These are the corrected point coordinates, in mm. The intersection angle difference referred to in this manual refers to the unit normal vector of the local correction surface. unit vector of local normal at point The angle between them; Let be the cosine of the intersection angle difference. In equation (4), the numerator corresponds to the cutting residual of the projection distortion point relative to the local correction surface, and the denominator corresponds to the intersection relationship between the local correction surface normal and the point local normal. Together, they determine the point-by-point projection correction amount along the local normal direction.

[0068] For example, the longitudinal coordinate difference between corresponding points on the upper and lower surfaces is 0.32 mm, which exceeds the morphological threshold of 0.20 mm, and is therefore identified as a projection distortion point pair. After fitting the local correction surface, the cutting residual of the distortion point is 0.24 mm, the dot product of the local correction surface normal and the local normal of the point is 0.80, and the corresponding intersection angle difference is 36.87°. Substituting the values ​​into equation (4), the point-by-point projection correction distance is -0.30 mm, which does not exceed the maximum correction amount of 0.50 mm; if the coordinates of the distortion point are (52.10 mm, 18.30 mm, 3.24 mm), and the local normal is along the uniform measurement longitudinal axis, then the corrected point coordinates are (52.10 mm, 18.30 mm, 2.94 mm).

[0069] When the absolute value of the dot product of the local correction surface normal and the local point normal is not greater than 0.10, or the absolute value of the point-by-point projection correction distance exceeds 0.50mm, the distortion correction and measurement output module does not perform unstable correction, but instead calls the section cutting parameters saved for the corresponding section and returns to the section continuous correction and contour acquisition module to re-execute direction correction and spatial intersection.

[0070] After the true cross-sectional profile is formed, the distortion correction and measurement output module determines the extreme points, converts the measured physical quantities, and forms a cross-sectional measurement sequence in a unified measurement coordinate system. The distortion correction and measurement output module is further used for: Determine the highest point of the upper surface, the lowest point of the lower surface, and their spatial positions from the actual cross-sectional profile. Based on the unified measurement benchmark, the highest point and the lowest point are transformed to the thin-walled cover measurement coordinate system to determine the warping amount and edge height difference of the corresponding cross section; The adjacent real cross-sectional profiles are arranged according to the edge extension direction of the thin-walled cover to form a cross-sectional measurement sequence that reflects the continuous morphology of the edge steps.

[0071] Specifically, the highest and lowest points are determined along the longitudinal axis of a unified measurement coordinate system, while simultaneously preserving their three-dimensional spatial positions. Edge height difference. The warpage is determined by the coordinate difference between the highest point of the upper surface and the lowest point of the lower surface along a uniform longitudinal axis, in mm. The unit is mm, determined by the maximum normal deviation of the actual cross-sectional profile from the unified datum surface or the confirmed design reference surface, which is determined by the positioning datum features. The unified datum surface is derived from the same datum transformation version output by the laser point cloud acquisition and datum calibration modules.

[0072] For example, the longitudinal coordinate of the highest point on the upper surface of the aforementioned actual cross-sectional profile is 3.40 mm, and the longitudinal coordinate of the lowest point on the lower surface is 1.22 mm. Therefore, the height difference at the edge of the cross-section is... The warpage is 2.18 mm; the maximum normal deviation of the actual cross-sectional profile relative to the unified datum plane is 0.42 mm, therefore the warpage of this cross-section is... The value is 0.42 mm. The distortion correction and measurement output module writes the position, highest point, lowest point, edge height difference, warping amount, reference transformation version, and verification status of the cross-section into the same cross-section record.

[0073] Adjacent real cross-sectional profiles are arranged according to the edge extension direction and corresponding arc length position of the thin-walled cover, forming a cross-sectional measurement sequence reflecting the continuous morphology of the edge steps. The cross-sectional measurement sequence can also be used to generate a three-dimensional morphological record of the edge steps, but does not replace the warping amount and edge height difference with abstract defect categories or scores.

[0074] After forming the cross-sectional measurement sequence, the distortion correction and measurement output module verifies the contours of adjacent real cross-sections, and recalculates the parameters saved for the corresponding cross-section if the verification fails. The distortion correction and measurement output module is further used for: The results of the section verification are determined based on the changes in normal continuity, projection difference, and height difference between adjacent real section profiles. If the cross-section verification result does not meet the verification conditions, the cross-section cutting parameters saved for the corresponding cross-section are called, and the direction correction, spatial intersection and point-by-point projection correction are re-executed to obtain the verification cross-section profile. The highest and lowest points are re-extracted from the cross-sectional profile of the verification section, the warping amount and edge height difference at the corresponding positions are updated, and the updated results are used as the final measurement results at the corresponding positions.

[0075] Specifically, the limits for normal continuity, average projection difference, and height difference variation are determined jointly by repeated measurements of the continuous cross-section of the reference piece, normal edge morphology variations, and design tolerances (e.g., the limit for normal continuity is 2.0°, the limit for average projection difference is 0.10 mm, and the limit for height difference variation is 0.30 mm). The average projection difference is the average absolute value of the point-by-point projection correction distance along the local normal of each projection distortion point in the current real cross-section that has completed stable correction. The three verification quantities are first compared with their corresponding limits, and a normalized verification quantity is formed while retaining the original data of the three quantities. The verification weights are determined by the order of influence of each verification quantity on the continuity of the cross-section (e.g., the weight for normal continuity is 0.40, the weight for average projection difference is 0.30, and the weight for height difference variation is 0.30; all three weights are dimensionless). The normalized verification quantity is determined according to the following formula: In equation (5), For the first The normalized complex value of a real cross section, dimensionless; This represents the change in normal between the current actual cross section and the previously confirmed cross section, in degrees. This is the normal continuity limit, in degrees (°). This is the average absolute value of the point-by-point projection correction distance of each stable projection correction point in the current real cross section, in mm; This is the average projection difference limit, in mm; and These represent the difference in edge height between the current actual cross-section and the previously confirmed cross-section, in mm. These are the limits for changes in height difference, in mm. , and These are the verification weights for normal continuity, average projection difference, and height difference variation, respectively. They are dimensionless, and the sum of the values ​​of the three verification weights is 1.0. , and All do not exceed the corresponding limits, and When the value is not greater than 1, the current section meets the comprehensive review conditions.

[0076] For example, the normal variation between the current actual cross section and the previously confirmed cross section is 1.2°, the average projection difference is 0.06mm, the current cross section edge height difference is 2.18mm, and the previous confirmed cross section edge height difference is 2.08mm. Substituting the above values ​​and the verification weights of 0.40, 0.30, and 0.30 into equation (5), we obtain... The value is approximately 0.52. The changes in normal direction, average projection difference, and height difference do not exceed their respective limits. The value is no greater than 1, therefore the current section is marked as confirmed and enters the final measurement sequence.

[0077] When any of the three review items does not meet the corresponding review conditions, or When the value is greater than 1, the distortion correction and measurement output module calls the dynamic cutting plane, reverse rotation parameters, continuity fusion state, and intersection tolerance saved for the current cross-section, re-executes direction correction, spatial intersection, and point-by-point projection correction, and re-extracts the highest and lowest points from the verified cross-section profile. The recalculated warpage and edge height difference update the corresponding cross-section record, without replacing the measurement result of the current cross-section with the final measurement value of the adjacent cross-section. If the verification conditions are still not met after recalculation, the current cross-section is marked as a cross-section to be manually verified and is not used as the continuity reference for the next cross-section.

[0078] During system operation, if scan data is missing, positioning reference features cannot correspond, or the error after reference calibration still exceeds the reference validity condition, the system will prevent the corresponding point cloud from entering edge point cloud extraction and physical quantity conversion; if there are insufficient effective samples in the local neighborhood, the neighborhood can be expanded once within the aforementioned neighborhood scale range. If it is still insufficient after expansion, the corresponding position will be marked as an unmeasurable gap; if the cumulative normal deflection angle and curvature change gradient do not exceed the corresponding threshold at the same time, the cutting plane state of the previously confirmed section or the local initial plane will be maintained, and no new bending trigger state will be generated; if the spatial direction difference does not exceed the angle threshold, reverse rotation will not be performed; if the longitudinal difference between the upper and lower surface point pairs does not exceed the shape threshold, the original smooth section point position will be retained without generating unnecessary point-by-point projection correction.

[0079] When the current task data for a certain module is not fully formed, the system saves the state of the most recent complete scan band or the most recent confirmed section. It does not generate a reference point cloud based on the incomplete scan band, nor does it use the incompletely verified section as the reference for the next section. After obtaining complete data again, the system continues to run from the corresponding complete state, and retains the correspondence between the original point cloud, the reference transformation version, the section cutting parameters, the contour point labels, and the verification results.

[0080] Accordingly, this invention uses a laser point cloud acquisition and benchmark calibration module to determine the pose correction of the scanned area points by combining the fitting difference of the positioning benchmark and the registration difference of the overlapping points, so that the point clouds, cross-sectional coordinates, and final measured physical quantities of each scanned area are at the same measurement benchmark; an edge point cloud extraction module combines neighborhood covariance, projection distance, edge confidence, and entity contour connectivity to form a continuous edge point cloud, and a local pose analysis module obtains the normal direction, curvature change gradient, curvature principal axis, and fitting residual; a dynamic cutting plane generation module, after removing residual excessive normal angles, forms a smooth normal based on normal consistency and point cloud density, and uses the cumulative normal deflection angle and curvature change gradient as dual thresholds to determine the bending point and the direction and position of the dynamic cutting plane; a cross-section continuous correction and contour acquisition module generates candidate cutting planes from the previously confirmed cross-section, using the current edge The trend line vector triggers reverse rotation and residual update, and after spatial intersection, it combines the current cross-sectional residual and the corresponding points of adjacent confirmed cross-sections to distinguish between real noise points and real abrupt change points. Through distortion correction and measurement output modules, it establishes upper and lower surface point pairs along the local normal, and forms intersection angle difference, cutting residual and point-by-point projection correction amount by local correction surface and spatial intersection relationship, thereby obtaining the real cross-sectional profile, and determining the warping amount and edge height difference under a unified measurement benchmark. It further verifies the normal continuity, projection difference and height difference changes of adjacent real cross-sectional profiles, and if the verification does not meet the conditions, it calls the cross-sectional cutting parameters saved by the corresponding cross-section to re-execute the direction correction, spatial intersection and point-by-point projection correction, thus forming a complete measurement closed loop from laser point cloud input, dynamic cross-section generation, continuous correction, local distortion correction, physical quantity output to verification backtracking.

[0081] The above description is merely a preferred embodiment of the technical solution of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A laser measurement system for warpage and edge height difference of a thin-walled cover, characterized in that, include: Laser point cloud acquisition and benchmark calibration module: used to acquire three-dimensional measurement point clouds of thin-walled cover parts with a laser scanner and calibrate them as benchmark point clouds according to positioning benchmark features; Edge point cloud extraction module: used to extract local attitude normals from the neighborhood covariance distribution of the reference point cloud to obtain the edge point cloud; Local pose analysis module: used to fit the edge point cloud micro-surface, obtain the normal direction, curvature change gradient, curvature principal axis and fitting residual, and correct the normal direction according to the continuity of adjacent normals; Dynamic cutting plane generation module: used to obtain smooth normals by weighted smoothing of normal angles, determine the bending point when the cumulative normal deflection angle and curvature change gradient both exceed the threshold, and generate dynamic cutting planes based on smooth normals and curvature principal axes. The section continuous correction and contour acquisition module is used to generate candidate cutting planes based on the previously confirmed section. When the direction difference between the candidate surface normal vector and the trend line vector of the current edge contour exceeds the threshold, the candidate surface normal vector is rotated in the reverse direction. The section cutting parameters are updated based on the residual between the rotated normal vector and the previous section normal vector, so that it intersects with the edge point cloud space to obtain a smooth section contour. Distortion correction and measurement output module: used to establish point pairs on the upper and lower surfaces of a smooth cross-section profile along the local normal, fit point pairs with longitudinal differences exceeding the threshold as local correction surfaces, determine the intersection angle difference and cutting residual based on the spatial intersection relationship between the point pairs and the correction surfaces, generate point-by-point projection correction amount to obtain the true cross-section profile, and convert the warping amount and edge height difference according to a unified measurement benchmark.

2. The system according to claim 1, characterized in that, The laser point cloud acquisition and benchmark calibration module is further used for: Extract the point cloud of the overlapping area between adjacent scan bands, and remove outliers based on the normal vector of the overlapping point cloud to obtain the denoised overlapping point cloud; Obtain the positioning reference feature coordinates in the point cloud of the denoised overlapping area, and determine the positioning reference fitting difference and the registration difference of the overlapping points of adjacent scan bands respectively; When both the fitting difference of the positioning reference and the registration difference of the overlapping points exceed the corresponding thresholds, a joint correction matrix is ​​constructed based on the correspondence between the theoretical and measured positions of the feature coordinates of the positioning reference and the overlapping point pairs in the denoised overlapping area point cloud. Affine transformation and coordinate transformation are then performed on the poses of the points in adjacent scanning zones to generate the reference point cloud.

3. The system according to claim 1, characterized in that, The edge point cloud extraction module is further used for: Based on the spatial coordinates of each reference point in the reference point cloud, a local neighborhood is constructed and the covariance matrix is ​​calculated. The minimum eigenvalue and its corresponding eigenvector are extracted to form a set of local attitude normal vectors. The initial edge points are determined by the projection distance of each reference point along the corresponding local attitude normal vector, and the edge confidence is determined by combining the proportion of minimum eigenvalue, the density of neighboring points and the deviation of the reference plane. High-confidence edge points that meet the edge confidence criteria are retained. The connectivity of adjacent high-confidence edge points is verified along the contour direction of the thin-walled cover body. Isolated points are eliminated to generate the edge point cloud.

4. The system according to claim 1, characterized in that, The local attitude resolution module is further used for: Based on the point cloud density, curvature value and preset scale of the edge points, the neighborhood samples are determined, the local micro-surface is obtained by least squares fitting, and the initial normal direction, curvature change gradient and fitting residual are extracted from the local micro-surface. The continuity of the normal direction is determined based on the angle between the initial normal direction and the normal direction of the adjacent region. When the continuity of the normal direction does not meet the consistency condition, the degree of deviation of the normal fitting is determined based on the fitting residual, and the attitude compensation amount is determined in combination with the continuity of the normal direction. The initial normal direction is then corrected in the direction opposite to the deviation of the normal direction.

5. The system according to claim 4, characterized in that, The dynamic cutting plane generation module is further used for: Calculate the normal angle between the compensated normal direction and the normal direction of the adjacent region, and remove normal angles whose fitting residual exceeds the residual threshold; Smoothing weights are assigned based on the normal consistency and point cloud density corresponding to the remaining normal angles, forming a continuous smooth normal deflection trend and the corresponding smooth normal. The cumulative normal deflection angle is determined based on the smooth normal deflection trend. When the cumulative normal deflection angle and the curvature change gradient both exceed the corresponding threshold, the bending point is determined. The direction and position of the dynamic cutting plane are determined by the smooth normal and the principal axis of curvature of the bending point.

6. The system according to claim 5, characterized in that, The cross-section continuous correction and contour acquisition module is further used for: The current candidate cutting plane is generated based on the normal vector and center point coordinates of the previously confirmed cross section, and the three-dimensional contour coordinates of the current edge at the corresponding position on the candidate cutting plane are extracted. The edge trend line vector is determined based on the three-dimensional contour coordinates, and the spatial direction difference between the edge trend line vector and the candidate surface normal vector is calculated. When the spatial orientation difference exceeds the angle threshold, the candidate surface normal vector is rotated in reverse, and the section cutting parameters are updated based on the residual between the rotated normal vector and the normal vector of the previously confirmed section.

7. The system according to claim 1, characterized in that, The cross-section continuous correction and contour acquisition module is further used for: A set of candidate points is formed based on the spatial intersection results of the cross-sectional cutting parameters and the edge point cloud. The normal distance, neighborhood point density and continuity along the edge direction of each candidate point relative to the local main contour are calculated. Outlier candidate points are identified based on the normal distance, the neighborhood point density, and the continuity along the edge direction. In combination with the current cross-sectional residual and whether there are corresponding points in the adjacent confirmed cross-sections, real noise points are distinguished from real abrupt change points at the edge of the thin-walled cover. The smooth cross-sectional profile is obtained by removing the real noise points from the candidate point set and retaining the real abrupt change points.

8. The system according to claim 7, characterized in that, The distortion correction and measurement output module is further used for: Obtain the upper surface point set and lower surface point set of the smooth cross-sectional profile respectively, determine the local normal direction based on the spatial coordinates of the two, and establish corresponding point pairs of the upper and lower surfaces along the local normal direction; When the longitudinal coordinate difference between corresponding point pairs on the upper and lower surfaces exceeds the morphological threshold, a projection distortion point pair is determined. Plane fitting is performed on the projection distortion point pairs to obtain a local correction surface. The intersection angle difference and cutting residual are determined according to the spatial intersection relationship between the projection distortion point pairs and the local correction surface, and a point-by-point projection correction amount is generated accordingly to obtain the true cross-sectional profile.

9. The system according to claim 8, characterized in that, The distortion correction and measurement output module is further used for: Determine the highest point of the upper surface, the lowest point of the lower surface, and their spatial positions from the actual cross-sectional profile. Based on the unified measurement benchmark, the highest point and the lowest point are transformed to the thin-walled cover measurement coordinate system to determine the warping amount and edge height difference of the corresponding cross section; The adjacent real cross-sectional profiles are arranged according to the edge extension direction of the thin-walled cover to form a cross-sectional measurement sequence that reflects the continuous morphology of the edge steps.

10. The system according to claim 9, characterized in that, The distortion correction and measurement output module is further used for: The results of the section verification are determined based on the changes in normal continuity, projection difference, and height difference between adjacent real section profiles; If the cross-section verification result does not meet the verification conditions, the cross-section cutting parameters saved for the corresponding cross-section are called, and the direction correction, spatial intersection and point-by-point projection correction are re-executed to obtain the verification cross-section profile. The highest and lowest points are re-extracted from the cross-sectional profile of the verification section, the warping amount and edge height difference at the corresponding positions are updated, and the updated results are used as the final measurement results at the corresponding positions.

Images