A System for Evaluating the Surface Flatness of Rubber Liners Based on Laser Profile Scanning
By using laser contour scanning technology, the problems of inconsistent measurement point heights and unstable line-by-line calculations in the surface flatness detection of rubber linings have been solved. This has enabled the integrity of multi-scale features and the traceability of defect maps, thereby improving the accuracy and reliability of the surface flatness assessment of rubber linings.
Patent Information
- Application Number
- CN202611588525.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-14
- Publication Date
- 2026-11-17
AI Technical Summary
Existing methods for detecting the surface smoothness of rubber linings suffer from problems such as inconsistent measurement point heights, unstable row-by-row smoothness calculations, and a disconnect between multi-scale features, defect maps, and grade assessments, resulting in insufficient consistency and traceability of assessment results.
A laser contour scanning-based method is adopted, which uses point cloud construction, recursive observation, flatness calculation, scale decomposition, feature calculation and defect map module to achieve high consistency correction of rubber lining surface, pose interpolation of progressive scanning mechanism, multi-scale feature extraction and defect type determination, and form flatness evaluation results.
It improves the high consistency and coordinate uniformity of the measured point cloud on the rubber lining surface, enhances the stability of line-by-line calculation, realizes the integrity of multi-scale features and the traceability of defect identification, and improves the integrity and engineering applicability of flatness assessment.
Smart Images

Figure CN123111147A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial surface inspection and three-dimensional measurement technology, and in particular to a system for evaluating the surface flatness of rubber linings based on laser contour scanning. Background Technology
[0002] Current methods for detecting the surface flatness of rubber linings commonly employ laser ranging, single-point profilometry, or manual inspection. During scanning, the reflection characteristics of the laser on the rubber surface change, and the echo intensity is affected by color, roughness, and coating. Directly using the height of the measurement point leads to inconsistent height data for the same plane. In reciprocating grid scanning, adjacent scan rows have opposite directions, and changes in the scanning mechanism's pose, inter-row misalignment, and mirror correspondence are not addressed. Point cloud coordinate transformation lacks row-by-row interpolation, making it difficult to stably extrapolate the row height distribution. Existing methods also lack quality marking and active set constraints for suspicious measurement points, making abrupt changes and noise easily introduced into the flatness height field.
[0003] In terms of feature measurement and evaluation, existing solutions often combine flatness, waviness, and roughness in their calculations, without using the inflection point of the autocorrelation function to define the scale, resulting in insufficient separation of shape, waviness, and roughness components. The division of concave and convex areas relies on fixed thresholds, and the extraction of features such as peak and valley height, projected area, row span, column span, aspect ratio, and waviness wavelength is incomplete. It is also difficult to distinguish defect types such as lap joint steps, bulges, depressions, and wrinkles. Furthermore, there is a lack of a process for dividing evaluation units by lap joint steps and comprehensively classifying them, resulting in insufficient consistency and traceability of evaluation results. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention provides a rubber lining surface flatness evaluation system based on laser contour scanning, aiming to solve the problems of insufficient point cloud height consistency, unstable line-by-line flatness calculation, and disconnect between multi-scale features and defect maps and grade evaluation in rubber lining surface contour measurement.
[0005] To achieve the above objectives, the present invention provides the following technical solution: a rubber lining surface flatness evaluation system based on laser contour scanning, comprising the following steps: Point cloud construction module: used to perform contour scanning on the surface of rubber lining to obtain the height of profile measurement points, echo intensity and scanning mechanism pose; perform consistency correction on the height of profile measurement points based on echo intensity, and construct measurement point cloud based on the corrected height of profile measurement points and scanning mechanism pose; Recursive observation module: It is used to construct a local slope observation set based on the height difference and lateral spacing of adjacent profile measurement points in the measurement point cloud, interpolate the pose of the scanning mechanism of adjacent scan lines to obtain the pose of the scanning mechanism line by line, establish the state transition relationship based on the change of the pose of the scanning mechanism of adjacent scan lines, and form a recursive observation structure by the local slope observation set and the state transition relationship. Smoothness calculation module: used to advance along the scanning direction of the recursive observation structure line by line, calculate the line height distribution of each scanning line, and obtain the smoothness height field; Scale decomposition module: used to perform multi-scale decomposition on the flatness height field to obtain multi-scale component sets; Feature Calculation Module: Used to calculate flatness, waviness amplitude and roughness amplitude according to multi-scale component sets, divide the concave and convex areas according to the shape component field in the multi-scale component set, and extract the features of the concave and convex areas to form a flatness feature set; Defect Mapping Module: Used to cluster uneven areas according to the flatness feature set to obtain clusters, determine the defect type corresponding to the clusters, and form a defect map; Evaluation Results Module: Used to classify levels based on defect maps and flatness feature sets, and output flatness evaluation results.
[0006] Furthermore, the process involves performing a contour scan on the rubber lining surface to obtain the height of the profile measurement points, the echo intensity, and the pose of the scanning mechanism; performing consistency correction on the height of the profile measurement points based on the echo intensity; and constructing a measurement point cloud based on the corrected height of the profile measurement points and the pose of the scanning mechanism, including: The coverage width of the laser on the surface of the rubber lining is collected, and the side length of the coverage width in the inter-row direction is taken as the row spacing of the reciprocating grid scanning path. The projection size of a single pixel on the surface of the rubber liner is acquired by the area frame camera, and the component of the projection size in the scanning direction is taken as the in-row step of the reciprocating grid scanning path. The reciprocating grid scanning path is determined by the row spacing and the in-row step distance. The height of the profile measurement points is calculated by laser triangulation along the reciprocating grid scanning path. The echo intensity of each profile measurement point is collected by an area array camera. The pose of the scanning mechanism is collected by an encoder and an inertial measurement unit. For each scan line along the scanning direction, a polynomial fit is performed on the neighborhood of the profile measurement point. The fitting result is used as the local smoothing height, and the difference between the height of the profile measurement point and the local smoothing height is used as the local residual. Intensity bins are divided according to the equal frequency of the echo intensity. The median of the local residuals of the intensity bins is calculated, and a broken line is plotted between the median of the residuals and the echo intensity. Ordinal regression is applied to the broken line to construct the intensity-residual relationship curve. Subtract the value of the corresponding point in the strength-residual relationship curve from the height of the profile measuring point to obtain the corrected height. Statistically analyze the local residual distribution in the strength box, mark suspicious profile measuring points according to the local residual distribution, and add quality marks to the profile measuring points. The coordinates of the profile measurement points are constructed by using the row number, column number, corrected height, row spacing and in-row step distance of the reciprocating grid scanning path. The pose of the scanning mechanism is interpolated according to the acquisition time. The coordinates of the profile measurement points are then transformed to a unified measurement coordinate system based on the interpolated pose of the scanning mechanism to form a measurement point cloud.
[0007] Furthermore, the step of constructing a local slope observation set based on the height difference and lateral spacing of adjacent profile measurement points in the measurement point cloud, and interpolating the scanning mechanism poses of adjacent scanning lines to obtain the line-by-line scanning mechanism pose, includes: Calculate the ratio of the height difference of each pair of adjacent profile measurement points in each scan line of the measurement point cloud to the in-line step distance of the reciprocating grid scan path, and use the obtained ratio as the local slope observation value. The entire pair of adjacent profile measurement points is traversed row by row. All local slope observations constitute the local slope observation set. The adjacent profile measurement point pairs with at least one end bearing the profile measurement point quality mark are removed from the local slope observation set. Using the acquisition time of two adjacent scanning mechanism poses as interpolation nodes, the scanning mechanism poses are interpolated, and the interpolation results are used as the row-by-row scanning mechanism poses according to the row number.
[0008] Furthermore, the establishment of state transition relationships based on the pose changes of the scanning mechanism in adjacent scan lines, and the formation of a recursive observation structure by the local slope observation set and the state transition relationships, includes: Using the line connecting the poses of the scanning mechanisms in the first and last scan lines as the reference straight line, the ideal scanning mechanism pose sequence is calculated by dividing the line into equal parts according to the row number. The scanning mechanism pose of each scan line is subtracted from the ideal scanning mechanism pose sequence parameter by parameter to obtain the scanning mechanism pose residual. The scanning mechanism pose residual is then decomposed into line height offset and line tilt. Based on the difference between the row height offset of the current scan line and the difference between the row tilt of the previous scan line, a state transition relationship is established to map the row height distribution of the previous scan line to the current scan line. Based on the geometric relationship that the directions of travel of two adjacent scan lines in the reciprocating grid scanning path are opposite, the profile measurement points of the current scan line are mirrored with the profile measurement points of the previous scan line according to the column number to obtain the inter-row transfer coefficient. Align the line height distribution of the previous scan line with the profile measurement points of the current scan line point by point according to the inter-line transfer coefficient. Superimpose the difference between the line height offset and the difference between the in-line tilt according to the state transition relationship to obtain the prior height distribution of the current scan line. Then, obtain the prior slope sequence according to the ratio of the height difference between adjacent profile measurement points to the lateral spacing. The prior slope sequence and the local slope observation set of the current scan row are arranged side by side according to column number to form a recursive observation structure.
[0009] Furthermore, the step of advancing the recursive observation structure along the scanning direction line by line, calculating the row height distribution of each scan line, and obtaining the flatness height field includes: Using the local slope observation set of the first scan line as the initial state, the product of the local slope observation value and the row step distance of the reciprocating grid scan path is accumulated along the scanning direction to obtain the row height distribution of the first scan line. The upper limit of bending tolerance is determined based on the mean and standard deviation of the second difference of the full scan line. The profile measurement points that exceed the upper limit of bending tolerance are identified as abrupt profile measurement points, and the profile measurement points that do not exceed the upper limit of bending tolerance are identified as flat profile measurement points. All abrupt profile measurement points constitute an active set. The row height of each column in the current scan row is taken as an unknown. The unknown is merged according to the hard constraint that the row height between two adjacent flat profile measurement points is equal. The unknown is substituted into the recursive observation structure to perform constrained least squares solution to obtain the row height distribution of the current scan row. The slope sequence is obtained from the row height distribution of the current scan row by the ratio of the height difference between adjacent profile measurement points to the row step distance of the reciprocating grid scanning path. The slope sequence is then subtracted from the local slope observation set to obtain the observation deviation. The upper limit of the deviation is determined by the mean and standard deviation of the observation deviation of the entire scan row. Profile measurement points that exceed the upper limit of the deviation are merged into the active set at both ends. The solution and expansion are repeated until no more profile measurement points are added to the active set. From the second scan line to the last scan line, the hard constraint solution and active set expansion are repeated for each scan line. The row height distribution of each scan line is used as the input for the recursion of the next scan line. The flatness height field is obtained by compiling the row height distribution of each scanned row according to the row number and column number.
[0010] Furthermore, the flatness height field is decomposed into a multi-scale component set, including: For each scan line, the row height distribution of the same scan line is multiplied point by point with the row height distribution sequence after the same scan line is translated by several profile measurement points, and the average value of all products is obtained to obtain the autocorrelation coefficient under each translation amount. The autocorrelation function values under all translation amounts constitute the autocorrelation function of each scan line. The second-order difference is calculated point by point for the autocorrelation function. The inflection points of the two second-order differences are taken, and scale division points are generated according to the translation of the two inflection points. The smaller scale division point corresponds to the boundary between roughness and waviness, and the larger scale division point corresponds to the boundary between waviness and shape. Morphological filtering is applied to the flatness height field using structural elements whose size is larger than the larger scale segmentation point, and the filtering result is registered as a shape component field. Morphological filtering is applied to the flatness height field using structural elements whose dimensions are between two scale division points. The difference between the filtering result and the shape component field is recorded as the waviness component field, and the difference between the flatness height field and the filtering result is recorded as the roughness component field. The shape component field, roughness component field, and waviness component field are combined into a multi-scale component set.
[0011] Furthermore, the flatness, waviness amplitude, and roughness amplitude are calculated according to the multi-scale component set. The uneven region is divided according to the shape component field in the multi-scale component set, and the features of the uneven region are extracted to form a flatness feature set, including: Flatness of the shape component field is calculated using the minimum region method; The ridges and valleys of the waviness component field are extracted according to the morphological skeleton. The height difference between the profile measurement points between adjacent ridges and valleys is calculated and the median is taken to obtain the waviness amplitude. The roughness component field is scanned sequentially along the scanning direction to calculate the arithmetic mean deviation and root mean square deviation of the profile, thus forming the roughness amplitude. Based on the shape component field, profile measuring points whose height is greater than the height of adjacent profile measuring points in the row and column are selected as raised seed profile measuring points, and profile measuring points whose height is less than the height of adjacent profile measuring points in the row and column are selected as recessed seed profile measuring points. The growth height tolerance is determined based on the mean and standard deviation of the heights of adjacent profile measurement points in the rows and columns of the seed profile measurement points. Regional growth is then performed. Profile measurement points whose height difference with the protruding seed profile measurement points does not exceed the growth height tolerance and whose rows and columns are adjacent are merged into the protruding area. Profile measurement points whose height difference with the concave seed profile measurement points does not exceed the growth height tolerance and whose rows and columns are adjacent are merged into the concave area, thus forming a concave-convex area. The difference between the height of the maximum and minimum profile measuring points within the concave-convex region is recorded as the peak-valley height. The area of a single grid point is obtained by multiplying the row spacing of the reciprocating grid scanning path by the row step distance of the reciprocating grid scanning path. The projected area is obtained by multiplying the number of profile measuring points within the concave-convex region by the area of a single grid point. The peak-valley height and the projected area are recorded as the concave-convex region features. Flatness, waviness amplitude, roughness amplitude, and unevenness features are combined into a flatness feature set.
[0012] Furthermore, the step of clustering the uneven areas according to the flatness feature set to obtain clusters, determining the defect type corresponding to the clusters, and forming a defect map includes: For each uneven area in the flatness feature set, calculate the peak and valley height, projected area, row span, column span, aspect ratio, and ripple wavelength to form a morphological feature vector; The mean and standard deviation of each component of the morphological feature vector across all concave and convex regions are normalized, and the Euclidean distance between any two concave and convex regions is calculated as the morphological distance. Clustering is performed based on morphological distance to obtain a clustering tree; The clustering tree is truncated at the point where the increase in merging distance between adjacent levels is the largest, and the clusters are divided accordingly. By comparing the median of each component of the morphological feature vector of the concave and convex regions within each cluster, the order of the values of aspect ratio, peak and valley height and ripple wavelength is compared to determine the defect type of the cluster and the clusters are named as lap joint steps, bulges, depressions and wrinkles. By using the row spacing and in-row step distance of the reciprocating grid scanning path, the row number and column number of the concave and convex areas are converted into planar coordinates. The defect type, concave and convex areas, planar coordinates and morphological feature vectors are registered together to form a defect map.
[0013] Furthermore, the step of classifying the flatness level based on the defect map and flatness feature set, and outputting the flatness evaluation result, includes: The rubber lining surface is divided into assessment units according to the row number of the lap joint step in the defect map, with each assessment unit located between two adjacent lap joint steps. Based on the flatness feature set, the median of flatness, waviness amplitude, roughness amplitude, and unevenness features is extracted according to the evaluation unit; The median of each component—flatness, waviness amplitude, roughness amplitude, and unevenness feature—is calculated as the ratio of its ascending rank in all evaluation units to the total number of evaluation units, thus obtaining the score for each component. The geometric mean of the scores for each component is then calculated to synthesize the comprehensive flatness index. The rating is determined based on the median and upper quartile of the overall flatness index of all assessment units. Assessment units above the upper quartile are rated as excellent, those between the median and the upper quartile are rated as good, and those below the median are rated as acceptable. The overall flatness index, the scores of each component, and the defect types of each assessment unit are recorded together, and the flatness assessment results are output.
[0014] The present invention provides a rubber lining surface flatness evaluation system based on laser contour scanning, the advantages of which are mainly reflected in the following aspects: 1. This invention performs contour scanning on the surface of a rubber lining, determines the row spacing and in-row step distance based on the coverage width and pixel projection size, and acquires the height, echo intensity, and scanning mechanism pose of the profile measurement points using moving-line laser triangulation, an area array camera, an encoder, and an inertial measurement unit. Based on equal-frequency binning of the echo intensity, a broken line of the local residual median, and ordinal regression, an intensity-residual relationship curve is established. Height consistency correction is performed, suspicious profile measurement points are marked, and the coordinates of the profile measurement points are transformed to a unified measurement coordinate system according to the interpolated scanning mechanism pose, thus forming a measurement point cloud. This process considers the influence of echo intensity on height and the differences in measurement point quality, which helps improve the height consistency and coordinate uniformity of the measurement point cloud, providing a data foundation for subsequent row-by-row calculations.
[0015] 2. This invention utilizes recursive observation, flatness calculation, scale decomposition, feature measurement, defect mapping, and evaluation result modules. Based on the height difference and lateral spacing of adjacent profile measurement points, a local slope observation set is constructed. The pose of the scanning mechanism is interpolated line by line, establishing state transition relationships and a recursive observation structure. Along the scanning direction, the row height distribution is solved line by line using hard-constrained least squares and active set expansion to obtain the flatness height field. Shape, waviness, and roughness components are divided according to the autocorrelation second-order difference inflection point. Flatness, waviness amplitude, and roughness amplitude are calculated. Region growth is used to divide uneven areas and extract features such as peak and valley heights and projected areas. Clustering is used to determine lap joint steps, bulges, depressions, and wrinkles, forming a defect map. Evaluation units are separated according to lap joint steps, and a comprehensive flatness index and grade are calculated. This process integrates point cloud correction, line-by-line recursion, multi-scale features, defect identification, and graded evaluation, helping to improve the completeness, traceability, and engineering applicability of rubber lining surface flatness evaluation. Attached Figure Description
[0016] Figure 1 This is a functional block diagram of a rubber lining surface flatness evaluation system based on laser contour scanning according to the present invention. Figure 2 This is a flowchart of the recursive observation module of the present invention; Figure 3 This is a flowchart of the scale decomposition module of the present invention. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] Please see Figures 1-3This invention provides a system for evaluating the surface flatness of rubber linings based on laser contour scanning, comprising the following steps: Point cloud construction module: used to perform contour scanning on the surface of rubber lining to obtain the height of profile measurement points, echo intensity and scanning mechanism pose; perform consistency correction on the height of profile measurement points based on echo intensity, and construct measurement point cloud based on the corrected height of profile measurement points and scanning mechanism pose; Recursive observation module: It is used to construct a local slope observation set based on the height difference and lateral spacing of adjacent profile measurement points in the measurement point cloud, interpolate the pose of the scanning mechanism of adjacent scan lines to obtain the pose of the scanning mechanism line by line, establish the state transition relationship based on the change of the pose of the scanning mechanism of adjacent scan lines, and form a recursive observation structure by the local slope observation set and the state transition relationship. Smoothness calculation module: used to advance along the scanning direction of the recursive observation structure line by line, calculate the line height distribution of each scanning line, and obtain the smoothness height field; Scale decomposition module: used to perform multi-scale decomposition on the flatness height field to obtain multi-scale component sets; Feature Calculation Module: Used to calculate flatness, waviness amplitude and roughness amplitude according to multi-scale component sets, divide the concave and convex areas according to the shape component field in the multi-scale component set, and extract the features of the concave and convex areas to form a flatness feature set; Defect Mapping Module: Used to cluster uneven areas according to the flatness feature set to obtain clusters, determine the defect type corresponding to the clusters, and form a defect map; Evaluation Results Module: Used to classify levels based on defect maps and flatness feature sets, and output flatness evaluation results.
[0019] In this embodiment, the process of performing contour scanning on the rubber lining surface to obtain the height of the profile measurement points, echo intensity, and scanning mechanism pose; performing consistency correction on the height of the profile measurement points based on the echo intensity; and constructing a measurement point cloud based on the corrected height of the profile measurement points and the scanning mechanism pose includes: The coverage width of the laser on the surface of the rubber lining is collected, and the side length of the coverage width in the inter-row direction is taken as the row spacing of the reciprocating grid scanning path. The projection size of a single pixel on the surface of the rubber liner is acquired by the area frame camera, and the component of the projection size in the scanning direction is taken as the in-row step of the reciprocating grid scanning path. The reciprocating grid scanning path is determined by the row spacing and the in-row step distance. The height of the profile measurement points is calculated by laser triangulation along the reciprocating grid scanning path. The echo intensity of each profile measurement point is collected by an area array camera. The pose of the scanning mechanism is collected by an encoder and an inertial measurement unit. When the scanning mechanism moves along a scanning line, it triggers a profile acquisition once for each row step distance of the scanning path, and completes the entire line point by point. After reaching the end of the line, it moves one row distance along the line spacing to enter the next scanning line, and reverses the direction of travel. This process is repeated row by row until the entire rubber lining is covered. The row step distance determines the density of the profile measurement points in the row, and the row spacing determines the distance and overlap between adjacent rows.
[0020] Following the defined reciprocating grid scanning path, the motion mechanism drives the line laser triangulation unit to translate along the scanning direction. Each translation of the row spacing of the reciprocating grid scanning path triggers a profile acquisition. During each acquisition, the height sequence of the profile measurement points is calculated by the laser line triangulation. Simultaneously, the encoders mounted on each axis of the motion mechanism output the axial displacement, the inertial measurement unit mounted on the motion mechanism outputs the rotation angles of the three rotating axes, and the area scan camera outputs the echo intensity of each profile measurement point on the current profile. The axial displacement output by the encoder and the rotation angles of the three rotating axes output by the inertial measurement unit together constitute the pose of the scanning mechanism.
[0021] For each scan line along the scanning direction, a polynomial fit is performed on the neighborhood of the profile measurement point. The fitting result is used as the local smoothing height, and the difference between the height of the profile measurement point and the local smoothing height is used as the local residual. For each profile, the height sequence of the profile measurement points is used. Several profile measurement points are taken on both sides of each profile measurement point along the profile to form a neighborhood. Starting from a smaller length, the neighborhood is expanded successively. Each time, a quadratic polynomial least squares fitting is performed on the profile measurement points in the neighborhood, and the sum of squared fitting residuals is recorded. A curve of the sum of squared fitting residuals changing with the neighborhood length is plotted, and the neighborhood length corresponding to the inflection point where the second difference on the curve turns from negative to positive is taken.
[0022] Perform a quadratic polynomial least squares fitting based on the selected neighborhood length (determine the quadratic polynomial coefficients under the premise of minimizing the sum of squares of the height deviations between the fitted value in the neighborhood and the height of each profile measuring point in the neighborhood), take the value of the fitting result at the profile measuring point as the local smoothing height of the profile measuring point, and record the difference between the height of the profile measuring point and the local smoothing height as the local residual of the profile measuring point.
[0023] Intensity bins are divided according to the equal frequency of the echo intensity. The median of the local residuals of the intensity bins is calculated, and a broken line is plotted between the median of the residuals and the echo intensity. Ordinal regression is applied to the broken line to construct the intensity-residual relationship curve. The echoes are sorted from smallest to largest and grouped in an equal frequency manner. Each group has the same number of profile measurement points, and the group is registered as one intensity box. The number of boxes is obtained by gradually increasing the number of boxes starting from a smaller number. Each time, the median of the local residuals in each intensity box is calculated, and the absolute value of the difference between the medians of two adjacent intensity boxes is recorded. A sequence of the absolute value changing with the number of boxes is plotted, and the number of boxes corresponding to the minimum point where the sequence changes from decreasing to increasing is taken.
[0024] The method of applying ordinal regression is to find the piecewise constant sequence that minimizes the sum of squares of the differences between the median residual of each intensity bin and the regression value, provided that the median residual does not reverse with the intensity level. The resulting sequence is then registered as the intensity-residual relationship curve.
[0025] Subtract the value of the corresponding point in the strength-residual relationship curve from the height of the profile measuring point to obtain the corrected height. Statistically analyze the local residual distribution in the strength box, mark suspicious profile measuring points according to the local residual distribution, and add quality marks to the profile measuring points. The coordinates of the profile measurement points are constructed by using the row number, column number, corrected height, row spacing and in-row step distance of the reciprocating grid scanning path. The pose of the scanning mechanism is interpolated according to the acquisition time. The coordinates of the profile measurement points are then transformed to a unified measurement coordinate system based on the interpolated pose of the scanning mechanism to form a measurement point cloud.
[0026] The planar position of the profile measuring point in the sensor's local coordinate system is calculated by combining the row and column numbers of the profile measuring point with the row spacing and in-row step distance of the reciprocating grid scanning path. The normal position of the profile measuring point in the sensor's local coordinate system is determined by the height of the profile measuring point after correction, thus forming the coordinates of the profile measuring point in the sensor's local coordinate system. The quality mark of the profile measuring point is transformed along with the profile measuring point, and the original row and column numbers are retained after the transformation.
[0027] In this embodiment, the step of constructing a local slope observation set based on the height difference and lateral spacing of adjacent profile measurement points in the measurement point cloud, and interpolating the scanning mechanism poses of adjacent scanning lines to obtain the line-by-line scanning mechanism pose includes: Calculate the ratio of the height difference of each pair of adjacent profile measurement points in each scan line of the measurement point cloud to the in-line step distance of the reciprocating grid scan path, and use the obtained ratio as the local slope observation value. The entire pair of adjacent profile measurement points is traversed row by row. All local slope observations constitute the local slope observation set. The adjacent profile measurement point pairs with at least one end bearing the profile measurement point quality mark are removed from the local slope observation set. The process is repeated row by row and pair by pair. If any end of a pair of profile measurement points has a quality mark, the slope obtained from that pair is not reliable enough and is removed from the observation set to avoid the correction of outliers from contaminating subsequent slope constraints.
[0028] Using the acquisition time of two adjacent scanning mechanism poses as interpolation nodes, the scanning mechanism poses are interpolated, and the interpolation results are used as the row-by-row scanning mechanism poses according to the row number.
[0029] The pose sampling time is used as the interpolation node, and cubic spline interpolation is performed on each attitude parameter and displacement parameter respectively. The interpolation result is taken according to the acquisition time of each scan line and recorded as the pose of the scanning mechanism for that scan line. Spline interpolation is used instead of linear interpolation to ensure the continuity of the first and second derivatives of the pose at the nodes, and to avoid the introduction of abrupt changes in velocity or acceleration by interpolation.
[0030] In this embodiment, the establishment of state transition relationships based on the pose changes of the scanning mechanism in adjacent scan lines, and the formation of a recursive observation structure by the local slope observation set and the state transition relationships, includes: Using the line connecting the poses of the scanning mechanisms in the first and last scan lines as the reference straight line, the ideal scanning mechanism pose sequence is calculated by dividing the line into equal parts according to the row number. The scanning mechanism pose of each scan line is subtracted from the ideal scanning mechanism pose sequence parameter by parameter to obtain the scanning mechanism pose residual. The scanning mechanism pose residual is then decomposed into line height offset and line tilt. The measured pose of the scanning mechanism in each row is subtracted from the ideal pose of the same row in the ideal scanning mechanism pose sequence parameter by parameter to obtain the pose residual of the scanning mechanism. The translational residual perpendicular to the rubber lining surface is recorded as the row height offset, which indicates how much the row as a whole is raised or lowered relative to the reference. The two rotational residuals around the scanning direction and around the direction perpendicular to the scanning direction are multiplied by the lateral distance of each section measuring point in the row from the center line of the row and the value is recorded point by point as the in-row tilt.
[0031] Based on the difference between the row height offset of the current scan line and the difference between the row tilt of the previous scan line, a state transition relationship is established to map the row height distribution of the previous scan line to the current scan line. The establishment process is as follows: Take the difference in row height offset between two adjacent scan lines to obtain the overall height offset of the current scan line relative to the previous scan line; take the difference in inline tilt between two adjacent scan lines, and take the value point by point according to the lateral distance of each profile measuring point in the current scan line to obtain the inline tilt increment; align the row height distribution of the previous scan line with the profile measuring points of the current scan line according to the column number mirror according to the inter-row transfer coefficient; after alignment, first add the overall height offset to the height of the profile measuring point at the corresponding position of the previous scan line, and then add the inline tilt increment at that profile measuring point to obtain the prior height of that profile measuring point in the current scan line; register the alignment method and the two superimposed values together as a state transition relationship.
[0032] Based on the geometric relationship that the directions of travel of two adjacent scan lines in the reciprocating grid scanning path are opposite, the profile measurement points of the current scan line are mirrored with the profile measurement points of the previous scan line according to the column number to obtain the inter-row transfer coefficient. In reciprocating grid scanning, adjacent rows travel in opposite directions. The first profile measurement point of the current scan row and the last profile measurement point of the previous scan row are in the same horizontal position. They are mirrored according to the column number. This geometric relationship is recorded as the inter-row transfer coefficient so that the row height distribution of the previous row can be correctly aligned to the corresponding position of the current row.
[0033] Align the line height distribution of the previous scan line with the profile measurement points of the current scan line point by point according to the inter-line transfer coefficient. Superimpose the difference between the line height offset and the difference between the in-line tilt according to the state transition relationship to obtain the prior height distribution of the current scan line. Then, obtain the prior slope sequence according to the ratio of the height difference between adjacent profile measurement points to the lateral spacing. The height distribution of the previous scan row is transferred point by point to the column number of the current row according to the transfer coefficient; then the two differences mentioned above are superimposed to obtain the expected height distribution of the current row when there are no new observations or anomalies, i.e., the prior height distribution; the slope of the prior height distribution is calculated point by point according to the ratio of the height difference between adjacent profile measurement points to the in-row step distance of the reciprocating grid scan path to obtain the prior slope sequence.
[0034] The prior slope sequence and the local slope observation set of the current scan row are arranged side by side according to column number to form a recursive observation structure.
[0035] The process of finding the prior slope is exactly the same as the process of constructing the local slope observation set (the height difference between adjacent profile measurement points divided by the horizontal spacing). Therefore, the prior slope sequence and the local slope observation set of the current scan row naturally correspond point by point on the index (the position of the two profile measurement points in the row is determined by the same row step distance).
[0036] In this embodiment, the step of advancing the scanning line by line along the scanning direction according to the recursive observation structure and calculating the line height distribution of each scanning line to obtain the flatness height field includes: Using the local slope observation set of the first scan line as the initial state, the product of the local slope observation value and the row step distance of the reciprocating grid scan path is accumulated along the scanning direction to obtain the row height distribution of the first scan line. The local slope observation value is the ratio of the height difference between adjacent profile measurement points to the lateral spacing, and is itself a slope quantity. Multiplying each local slope observation value by the in-row step distance, the dimension is restored from height to length to height, that is, the height increment between two adjacent profile measurement points at that point. Accumulating these height increments point by point along the scanning direction, we can obtain the height of each profile measurement point in the first scan row relative to the starting point of that row.
[0037] The upper limit of bending tolerance is determined based on the mean and standard deviation of the second difference of the full scan line. The profile measurement points that exceed the upper limit of bending tolerance are identified as abrupt profile measurement points, and the profile measurement points that do not exceed the upper limit of bending tolerance are identified as flat profile measurement points. All abrupt profile measurement points constitute an active set. After taking the absolute value of the second difference of the full scan line, calculate its mean and standard deviation. Use the mean plus three times the standard deviation as the upper limit of the bending tolerance. If the upper limit is exceeded, it indicates that the slope has abruptly changed. Usually, it is the edge of the overlap, bulge or depression. It is recorded as the abrupt change profile measurement point. The remaining profile measurement points are recorded as flat profile measurement points. All abrupt change profile measurement points constitute the active set.
[0038] The row height of each column in the current scan row is taken as an unknown. The unknown is merged according to the hard constraint that the row height between two adjacent flat profile measurement points is equal. The unknown is substituted into the recursive observation structure to perform constrained least squares solution to obtain the row height distribution of the current scan row. During the solution process, the relationship between the height difference between two adjacent columns divided by the in-row step distance is substituted into the slope position in the structure. The unknowns are thus added to the equation system, and the number of unknowns is roughly equivalent to the number of rows and columns scanned. Hard constraints reduce the number of unknowns. Between two adjacent flat profile measurement points is a gradually changing section, where the height of each column is taken as the same unknown. Therefore, multiple unknowns in this section are combined into one, allowing the equation system to determine a unique solution. The solution aims only at minimizing the sum of squared observed residuals and does not include weights, penalty coefficients, or regularization terms.
[0039] The slope sequence is obtained from the row height distribution of the current scan row by the ratio of the height difference between adjacent profile measurement points to the row step distance of the reciprocating grid scanning path. The slope sequence is then subtracted from the local slope observation set to obtain the observation deviation. The upper limit of the deviation is determined by the mean and standard deviation of the observation deviation of the entire scan row. Profile measurement points that exceed the upper limit of the deviation are merged into the active set at both ends. The solution and expansion are repeated until no more profile measurement points are added to the active set. Determining the rationality of the current solution relies on restoring it to its slope and comparing it with the measured slope. The restoration process is exactly the same as when constructing the local slope observation set: the height difference between adjacent profile measurement points is divided by the lateral distance between these two points, thereby obtaining the slope sequence for the row height distribution of the current scan line; then, the slope sequence is subtracted from the local slope observation set for each profile measurement point, and the difference is the observation bias. A large observation bias indicates that the reconstructed height at that point does not match the measured tilt, usually meaning that the area has been misjudged as a flat section.
[0040] From the second scan line to the last scan line, the hard constraint solution and active set expansion are repeated for each scan line. The row height distribution of each scan line is used as the input for the recursion of the next scan line. Once the current scan row is solved, its row height distribution becomes the basis for the calculation of the next scan row. The next scan row is processed using the same method: first, the state transition relationship is established based on the difference between the row height offset and the in-row tilt of two adjacent scan rows, then hard constraints are applied, constrained least squares are performed, and the active set is expanded.
[0041] The flatness height field is obtained by compiling the row height distribution of each scanned row according to the row number and column number.
[0042] In this embodiment, the multi-scale decomposition of the flatness height field to obtain a multi-scale component set includes: For each scan line, the row height distribution of the same scan line is multiplied point by point with the row height distribution sequence after the same scan line is translated by several profile measurement points, and the average value of all products is obtained to obtain the autocorrelation coefficient under each translation amount. The autocorrelation function values under all translation amounts constitute the autocorrelation function of each scan line. The second-order difference is calculated point by point for the autocorrelation function. The inflection points of the two second-order differences are taken, and scale division points are generated according to the translation of the two inflection points. The smaller scale division point corresponds to the boundary between roughness and waviness, and the larger scale division point corresponds to the boundary between waviness and shape. For each scan line of the flatness height field, the height sequence of that line is multiplied point by point by the sequence after shifting the height sequence of that line by several profile measurement points, and the average of all products is calculated to obtain the autocorrelation coefficient under that shift. By traversing all shifts, the autocorrelation function of that scan line is obtained. Then, the second difference of the autocorrelation function is calculated point by point, and the shifts at the points where the second difference changes from positive to negative and from negative to positive are taken as the smaller-scale segmentation points and the larger-scale segmentation points, respectively.
[0043] Morphological filtering is applied to the flatness height field using structural elements whose size is larger than the larger scale segmentation point, and the filtering result is registered as a shape component field. Morphological filtering of the flatness height field is performed using structural elements with dimensions larger than the larger scale division point. Specifically, the structural element performs opening and closing operations on the flatness height field in sequence. Since the structural element size is larger than the larger scale division point, it means that only undulations with wavelengths longer than the waviness can be retained, thus obtaining the macroscopic shape of the entire lining.
[0044] Morphological filtering is applied to the flatness height field using structural elements whose dimensions are between two scale division points. The difference between the filtering result and the shape component field is recorded as the waviness component field, and the difference between the flatness height field and the filtering result is recorded as the roughness component field. Using structural elements with dimensions between two scale division points, a combination of opening and closing filters of the same type is applied. The resulting result contains both shape and waviness components. Therefore, subtracting the shape component field from the filtered result yields the waviness component field. The flatness height field contains three components: shape, waviness, and roughness. Therefore, subtracting the filtered result from the flatness height field yields the roughness component field.
[0045] The shape component field, roughness component field, and waviness component field are combined into a multi-scale component set.
[0046] In this embodiment, the step of calculating flatness, waviness amplitude, and roughness amplitude according to multi-scale component sets, dividing the uneven region according to the shape component field in the multi-scale component sets, and extracting the features of the uneven region to form a flatness feature set includes: Flatness of the shape component field is calculated using the minimum region method; Set up a parallel plane on each side of all the profile measurement points in the shape component field and continuously adjust their orientations until the two planes exactly encompass all the profile measurement points and the spacing is minimized; this minimum spacing is the flatness.
[0047] The ridges and valleys of the waviness component field are extracted according to the morphological skeleton. The height difference between the profile measurement points between adjacent ridges and valleys is calculated and the median is taken to obtain the waviness amplitude. Skeleton extraction shrinks each undulation in the waviness component field to a center line of a single measurement point width, with the high points of the undulations being ridge lines and the low points being valley lines. The difference between the height of the ridge line and the height of the valley line at the same row and column position is taken to obtain the waviness undulation amount at that point. The median of all undulation amounts is taken as the waviness amplitude to avoid individual deep valleys from skewing the results.
[0048] The roughness component field is scanned sequentially along the scanning direction to calculate the arithmetic mean deviation and root mean square deviation of the profile, thus forming the roughness amplitude. The profile arithmetic mean deviation is the average of the absolute values of the height deviation of each section measuring point within the scan line from the centerline of the line, and the profile root mean square deviation is the root mean square of the same deviation; the two together constitute the roughness range of the scan line, which is calculated and summarized after each line.
[0049] Based on the shape component field, profile measuring points whose height is greater than the height of adjacent profile measuring points in the row and column are selected as raised seed profile measuring points, and profile measuring points whose height is less than the height of adjacent profile measuring points in the row and column are selected as recessed seed profile measuring points. Adjacent in rows and columns refers to the cross-sectional measurement points that are adjacent to each other on the left and right within the same scan row and in the same column number on adjacent scan rows; the point whose height is higher than all adjacent in rows and columns is the highest point in the local area and is recorded as a raised seed cross-sectional measurement point; the point whose height is lower than all adjacent in rows and columns is the lowest point in the local area and is recorded as a recessed seed cross-sectional measurement point.
[0050] The growth height tolerance is determined based on the mean and standard deviation of the heights of adjacent profile measurement points in the rows and columns of the seed profile measurement points. Regional growth is then performed. Profile measurement points whose height difference with the protruding seed profile measurement points does not exceed the growth height tolerance and whose rows and columns are adjacent are merged into the protruding area. Profile measurement points whose height difference with the concave seed profile measurement points does not exceed the growth height tolerance and whose rows and columns are adjacent are merged into the concave area, thus forming a concave-convex area. Starting from each seed point, the process extends to adjacent profile measurement points, incorporating adjacent profile measurement points with the same extremum sign as the current concave-convex region, until the adjacent profile measurement points have opposite signs or there are no more profile measurement points in that direction. The set of included profile measurement points is then registered as a concave-convex region.
[0051] The sum of the mean of the absolute values of the waviness component field of the entire area and three times the standard deviation is taken as the upper limit of the fluctuation tolerance. Any point on a certain profile where the absolute value of the waviness component field exceeds the upper limit of the fluctuation tolerance is registered as a seed point.
[0052] The difference between the height of the maximum and minimum profile measuring points within the concave-convex region is recorded as the peak-valley height. The area of a single grid point is obtained by multiplying the row spacing of the reciprocating grid scanning path by the row step distance of the reciprocating grid scanning path. The projected area is obtained by multiplying the number of profile measuring points within the concave-convex region by the area of a single grid point. The peak-valley height and the projected area are recorded as the concave-convex region features. Peak-valley height is the difference in height between the highest and lowest profile measuring points within the same concave-convex area, measuring the depth of the protrusion or depression at that location.
[0053] Flatness, waviness amplitude, roughness amplitude, and unevenness features are combined into a flatness feature set.
[0054] Flatness characterizes macroscopic shape deviations, ripple amplitude distribution characterizes mesoscale periodic fluctuations, roughness amplitude distribution characterizes microscopic texture fluctuations, and concave-convex area characteristics characterize local significant deviations.
[0055] In this embodiment, the step of clustering uneven areas according to the flatness feature set to obtain clusters, determining the defect type corresponding to the clusters, and forming a defect map includes: For each uneven area in the flatness feature set, calculate the peak and valley height, projected area, row span, column span, aspect ratio, and ripple wavelength to form a morphological feature vector; Valley height and projected area are directly taken from the concave and convex features of the flatness feature set; row span is the product of the number of scan rows crossed by the concave and convex areas and the row spacing of the reciprocating grid scan path; column span is the product of the number of columns crossed by the concave and convex areas and the row spacing of the reciprocating grid scan path; aspect ratio is the ratio of row span to column span; ripple wavelength is the undulation wavelength of the concave and convex areas in the ripple component field, which takes information from the second scale and does not overlap with the shape scale.
[0056] The mean and standard deviation of each component of the morphological feature vector across all concave and convex regions are normalized, and the Euclidean distance between any two concave and convex regions is calculated as the morphological distance. Clustering is performed based on morphological distance to obtain a clustering tree; Initially, each concave and convex region is considered as a cluster. In each round, the pair of clusters with the smallest inter-cluster distance are merged. The inter-cluster distance is the minimum value of the pairwise morphological distance between the concave and convex regions in the two clusters. The objects merged in each round and the merge distance are recorded in order to form a clustering tree.
[0057] The clustering tree is truncated at the point where the increase in merging distance between adjacent levels is the largest, and the clusters are divided accordingly. Compare the increments of the merging distance between two adjacent levels from top to bottom. The level with the largest increment corresponds to the level where the number of clusters changes abruptly. The cluster obtained by breaking the clustering tree at that point is the cluster cluster.
[0058] By comparing the median of each component of the morphological feature vector of the concave and convex regions within each cluster, the order of the values of aspect ratio, peak and valley height, and ripple wavelength is used to determine the defect type of the cluster. The clusters were named lap joint steps, bulges, depressions, and folds. The cluster with the largest aspect ratio corresponds to the lap joint steps that extend slenderly along the scanning direction; the cluster with the largest peak-valley height and an aspect ratio close to one corresponds to the rounded bulges; the cluster with the smallest peak-valley height and a large proportion of the projected area corresponds to the shallow and wide depressions; and the cluster with the smallest ripple wavelength corresponds to the folds with dense ripples.
[0059] By using the row spacing and in-row step distance of the reciprocating grid scanning path, the row number and column number of the concave and convex areas are converted into planar coordinates. The defect type, concave and convex areas, planar coordinates and morphological feature vectors are registered together to form a defect map.
[0060] The row coordinate is obtained by multiplying the row number by the row spacing of the reciprocating grid scan path, and the column coordinate is obtained by multiplying the column number by the row spacing of the reciprocating grid scan path. The defect type, concave and convex areas, planar coordinates and morphological feature vectors are registered together to form a defect map with location.
[0061] In this embodiment, the step of classifying the flatness level based on the defect map and flatness feature set, and outputting the flatness evaluation result, includes: The rubber lining surface is divided into assessment units according to the row number of the lap joint step in the defect map, with each assessment unit located between two adjacent lap joint steps. The lap joint is a continuous step formed at the joint of two lining panels. It has been marked by row number in the defect map. The row number is used as the boundary, and the section of lining between two adjacent lap joints is an evaluation unit.
[0062] Based on the flatness feature set, the median of flatness, waviness amplitude, roughness amplitude, and unevenness features is extracted according to the evaluation unit; The median of each component—flatness, waviness amplitude, roughness amplitude, and unevenness feature—is calculated as the ratio of its ascending rank in all evaluation units to the total number of evaluation units, thus obtaining the score for each component. The geometric mean of the scores for each component is then calculated to synthesize the comprehensive flatness index. The rating is determined based on the median and upper quartile of the overall flatness index of all assessment units. Assessment units above the upper quartile are rated as excellent, those between the median and the upper quartile are rated as good, and those below the median are rated as acceptable. Arrange all the assessment units in ascending order of their overall flatness index, and divide the assessment units into three segments using the median and the upper quartile as the dividing lines: those ranked above three-quarters are rated as excellent, those between one-half and three-quarters are rated as good, and those below one-half are rated as acceptable.
[0063] The overall flatness index, the scores of each component, and the defect types of each assessment unit are recorded together, and the flatness assessment results are output.
[0064] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.
[0065] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0066] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.
Claims
1. A system for evaluating the surface flatness of rubber linings based on laser contour scanning, characterized in that, include: Point cloud construction module: used to perform contour scanning on the surface of rubber lining to obtain the height of profile measurement points, echo intensity and scanning mechanism pose; perform consistency correction on the height of profile measurement points based on echo intensity, and construct measurement point cloud based on the corrected height of profile measurement points and scanning mechanism pose; Recursive observation module: It is used to construct a local slope observation set based on the height difference and lateral spacing of adjacent profile measurement points in the measurement point cloud, interpolate the pose of the scanning mechanism of adjacent scan lines to obtain the pose of the scanning mechanism line by line, establish the state transition relationship based on the change of the pose of the scanning mechanism of adjacent scan lines, and form a recursive observation structure by the local slope observation set and the state transition relationship. Smoothness calculation module: used to advance along the scanning direction of the recursive observation structure line by line, calculate the line height distribution of each scanning line, and obtain the smoothness height field; Scale decomposition module: used to perform multi-scale decomposition on the flatness height field to obtain multi-scale component sets; Feature Calculation Module: Used to calculate flatness, waviness amplitude and roughness amplitude according to multi-scale component sets, divide the concave and convex areas according to the shape component field in the multi-scale component set, and extract the features of the concave and convex areas to form a flatness feature set; Defect Mapping Module: Used to cluster uneven areas according to the flatness feature set to obtain clusters, determine the defect type corresponding to the clusters, and form a defect map; Evaluation Results Module: Used to classify levels based on defect maps and flatness feature sets, and output flatness evaluation results.
2. The system for evaluating the surface flatness of rubber linings based on laser contour scanning according to claim 1, characterized in that, The process involves performing a contour scan on the surface of the rubber lining to obtain the height of the profile measurement points, the echo intensity, and the pose of the scanning mechanism. Based on the echo intensity, a consistency correction is applied to the height of the profile measurement points. A measurement point cloud is then constructed based on the corrected height of the profile measurement points and the pose of the scanning mechanism, including: The coverage width of the laser on the surface of the rubber lining is collected, and the side length of the coverage width in the inter-row direction is taken as the row spacing of the reciprocating grid scanning path. The projection size of a single pixel on the surface of the rubber liner is acquired by the area frame camera, and the component of the projection size in the scanning direction is taken as the in-row step of the reciprocating grid scanning path. The reciprocating grid scanning path is determined by the row spacing and the in-row step distance. The height of the profile measurement points is calculated by laser triangulation along the reciprocating grid scanning path. The echo intensity of each profile measurement point is collected by an area array camera. The pose of the scanning mechanism is collected by an encoder and an inertial measurement unit. For each scan line along the scanning direction, a polynomial fit is performed on the neighborhood of the profile measurement point. The fitting result is used as the local smoothing height, and the difference between the height of the profile measurement point and the local smoothing height is used as the local residual. Intensity bins are divided according to the equal frequency of the echo intensity. The median of the local residuals of the intensity bins is calculated, and a broken line is plotted between the median of the residuals and the echo intensity. Ordinal regression is applied to the broken line to construct the intensity-residual relationship curve. Subtract the value of the corresponding point in the strength-residual relationship curve from the height of the profile measuring point to obtain the corrected height. Statistically analyze the local residual distribution in the strength box, mark suspicious profile measuring points according to the local residual distribution, and add quality marks to the profile measuring points. The coordinates of the profile measurement points are constructed by using the row number, column number, corrected height, row spacing and in-row step distance of the reciprocating grid scanning path. The pose of the scanning mechanism is interpolated according to the acquisition time. The coordinates of the profile measurement points are then transformed to a unified measurement coordinate system based on the interpolated pose of the scanning mechanism to form a measurement point cloud.
3. The system for evaluating the surface flatness of rubber linings based on laser contour scanning according to claim 1, characterized in that, The method of constructing a local slope observation set based on the height difference and lateral spacing of adjacent profile measurement points in the measurement point cloud, and interpolating the poses of the scanning mechanism in adjacent scanning rows to obtain the poses of the scanning mechanism line by line, includes: Calculate the ratio of the height difference of each pair of adjacent profile measurement points in each scan line of the measurement point cloud to the in-line step distance of the reciprocating grid scan path, and use the obtained ratio as the local slope observation value. The entire pair of adjacent profile measurement points is traversed row by row. All local slope observations constitute the local slope observation set. The adjacent profile measurement point pairs with at least one end bearing the profile measurement point quality mark are removed from the local slope observation set. Using the acquisition time of two adjacent scanning mechanism poses as interpolation nodes, the scanning mechanism poses are interpolated, and the interpolation results are used as the row-by-row scanning mechanism poses according to the row number.
4. The system for evaluating the surface flatness of rubber linings based on laser contour scanning according to claim 1, characterized in that, The state transition relationship is established based on the pose changes of the scanning mechanism in adjacent scan rows, and the recursive observation structure is formed by the local slope observation set and the state transition relationship, including: Using the line connecting the poses of the scanning mechanisms in the first and last scan lines as the reference straight line, the ideal scanning mechanism pose sequence is calculated by dividing the line into equal parts according to the row number. The scanning mechanism pose of each scan line is subtracted from the ideal scanning mechanism pose sequence parameter by parameter to obtain the scanning mechanism pose residual. The scanning mechanism pose residual is then decomposed into line height offset and line tilt. Based on the difference between the row height offset of the current scan line and the difference between the row tilt of the previous scan line, a state transition relationship is established to map the row height distribution of the previous scan line to the current scan line. Based on the geometric relationship that the directions of travel of two adjacent scan lines in the reciprocating grid scanning path are opposite, the profile measurement points of the current scan line are mirrored with the profile measurement points of the previous scan line according to the column number to obtain the inter-row transfer coefficient. Align the line height distribution of the previous scan line with the profile measurement points of the current scan line point by point according to the inter-line transfer coefficient. Superimpose the difference between the line height offset and the difference between the in-line tilt according to the state transition relationship to obtain the prior height distribution of the current scan line. Then, obtain the prior slope sequence according to the ratio of the height difference between adjacent profile measurement points to the lateral spacing. The prior slope sequence and the local slope observation set of the current scan row are arranged side by side according to column number to form a recursive observation structure.
5. The system for evaluating the surface flatness of rubber linings based on laser contour scanning according to claim 1, characterized in that, The process of advancing the observation structure along the scanning direction line by line, calculating the line height distribution of each scanning line, and obtaining the flatness height field includes: Using the local slope observation set of the first scan line as the initial state, the product of the local slope observation value and the row step distance of the reciprocating grid scan path is accumulated along the scanning direction to obtain the row height distribution of the first scan line. The upper limit of bending tolerance is determined based on the mean and standard deviation of the second difference of the full scan line. The profile measurement points that exceed the upper limit of bending tolerance are identified as abrupt profile measurement points, and the profile measurement points that do not exceed the upper limit of bending tolerance are identified as flat profile measurement points. All abrupt profile measurement points constitute an active set. The row height of each column in the current scan row is taken as an unknown. The unknown is merged according to the hard constraint that the row height between two adjacent flat profile measurement points is equal. The unknown is substituted into the recursive observation structure to perform constrained least squares solution to obtain the row height distribution of the current scan row. The slope sequence is obtained from the row height distribution of the current scan row by the ratio of the height difference between adjacent profile measurement points to the row step distance of the reciprocating grid scanning path. The slope sequence is then subtracted from the local slope observation set to obtain the observation deviation. The upper limit of the deviation is determined by the mean and standard deviation of the observation deviation of the entire scan row. Profile measurement points that exceed the upper limit of the deviation are merged into the active set at both ends. The solution and expansion are repeated until no more profile measurement points are added to the active set. From the second scan line to the last scan line, the hard constraint solution and active set expansion are repeated for each scan line. The row height distribution of each scan line is used as the input for the recursion of the next scan line. The flatness height field is obtained by compiling the row height distribution of each scanned row according to the row number and column number.
6. The system for evaluating the surface flatness of rubber linings based on laser contour scanning according to claim 1, characterized in that, The multi-scale decomposition of the flatness height field yields a multi-scale component set, including: For each scan line, the row height distribution of the same scan line is multiplied point by point with the row height distribution sequence after the same scan line is translated by several profile measurement points, and the average value of all products is obtained to obtain the autocorrelation coefficient under each translation amount. The autocorrelation function values under all translation amounts constitute the autocorrelation function of each scan line. The second-order difference is calculated point by point for the autocorrelation function. The inflection points of the two second-order differences are taken, and scale division points are generated according to the translation of the two inflection points. The smaller scale division point corresponds to the boundary between roughness and waviness, and the larger scale division point corresponds to the boundary between waviness and shape. Morphological filtering is applied to the flatness height field using structural elements whose size is larger than the larger scale segmentation point, and the filtering result is registered as a shape component field. Morphological filtering is applied to the flatness height field using structural elements whose dimensions are between two scale division points. The difference between the filtering result and the shape component field is recorded as the waviness component field, and the difference between the flatness height field and the filtering result is recorded as the roughness component field. The shape component field, roughness component field, and waviness component field are combined into a multi-scale component set.
7. The system for evaluating the surface flatness of rubber linings based on laser contour scanning according to claim 6, characterized in that, The flatness, waviness amplitude, and roughness amplitude are calculated using multi-scale component sets. Concave and convex regions are delineated according to the shape component field within the multi-scale component sets, and features of these regions are extracted to form a flatness feature set, including: Flatness of the shape component field is calculated using the minimum region method; The ridges and valleys of the waviness component field are extracted according to the morphological skeleton. The height difference between the profile measurement points between adjacent ridges and valleys is calculated and the median is taken to obtain the waviness amplitude. The roughness component field is scanned sequentially along the scanning direction to calculate the arithmetic mean deviation and root mean square deviation of the profile, thus forming the roughness amplitude. Based on the shape component field, profile measuring points whose height is greater than the height of adjacent profile measuring points in the row and column are selected as raised seed profile measuring points, and profile measuring points whose height is less than the height of adjacent profile measuring points in the row and column are selected as recessed seed profile measuring points. The growth height tolerance is determined based on the mean and standard deviation of the heights of adjacent profile measurement points in the rows and columns of the seed profile measurement points. Regional growth is then performed. Profile measurement points whose height difference with the protruding seed profile measurement points does not exceed the growth height tolerance and whose rows and columns are adjacent are merged into the protruding area. Profile measurement points whose height difference with the concave seed profile measurement points does not exceed the growth height tolerance and whose rows and columns are adjacent are merged into the concave area, thus forming a concave-convex area. The difference between the height of the maximum and minimum profile measuring points within the concave-convex region is recorded as the peak-valley height. The area of a single grid point is obtained by multiplying the row spacing of the reciprocating grid scanning path by the row step distance of the reciprocating grid scanning path. The projected area is obtained by multiplying the number of profile measuring points within the concave-convex region by the area of a single grid point. The peak-valley height and the projected area are recorded as the concave-convex region features. Flatness, waviness amplitude, roughness amplitude, and unevenness features are combined into a flatness feature set.
8. The system for evaluating the surface flatness of rubber linings based on laser contour scanning according to claim 1, characterized in that, The process of clustering uneven areas according to a flatness feature set to obtain clusters, determining the defect type corresponding to each cluster, and forming a defect map includes: For each uneven area in the flatness feature set, calculate the peak and valley height, projected area, row span, column span, aspect ratio, and ripple wavelength to form a morphological feature vector; The mean and standard deviation of each component of the morphological feature vector across all concave and convex regions are normalized, and the Euclidean distance between any two concave and convex regions is calculated as the morphological distance. Clustering is performed based on morphological distance to obtain a clustering tree; The clustering tree is truncated at the point where the increase in merging distance between adjacent levels is the largest, and the clusters are divided accordingly. By comparing the median of each component of the morphological feature vector of the concave and convex regions within each cluster, the order of the values of aspect ratio, peak and valley height and ripple wavelength is compared to determine the defect type of the cluster and the clusters are named as lap joint steps, bulges, depressions and wrinkles. By using the row spacing and in-row step distance of the reciprocating grid scanning path, the row number and column number of the concave and convex areas are converted into planar coordinates. The defect type, concave and convex areas, planar coordinates and morphological feature vectors are registered together to form a defect map.
9. The system for evaluating the surface flatness of rubber linings based on laser contour scanning according to claim 8, characterized in that, The method of classifying levels based on defect maps and flatness feature sets, and outputting flatness evaluation results, includes: The rubber lining surface is divided into assessment units according to the row number of the lap joint step in the defect map, with each assessment unit located between two adjacent lap joint steps. Based on the flatness feature set, the median of flatness, waviness amplitude, roughness amplitude, and unevenness features is extracted according to the evaluation unit; The median of each component—flatness, waviness amplitude, roughness amplitude, and unevenness feature—is calculated as the ratio of its ascending rank in all evaluation units to the total number of evaluation units, thus obtaining the score for each component. The geometric mean of the scores for each component is then calculated to synthesize the comprehensive flatness index. The rating is determined based on the median and upper quartile of the overall flatness index of all assessment units. Assessment units above the upper quartile are rated as excellent, those between the median and the upper quartile are rated as good, and those below the median are rated as acceptable. The overall flatness index, the scores of each component, and the defect types of each assessment unit are recorded together, and the flatness assessment results are output.
