A kind of intelligent detection method for the thickness of glue film of stabilizer bar bushing
Patent Information
- Application Number
- CN202611462929.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-20
- Publication Date
- 2026-10-23
AI Technical Summary
[0005]为了解决现有方法在光晕导致的附加灰度能量被计入重心计算时，无法准确区分光晕拖尾区域与光条正常边缘过渡区域，从而导致光条中心线定位偏移、膜厚测量值系统性失真的问题，本发明提供了一种衬稳定杆衬套喷胶膜厚智能检测方法，包括以下步骤：
本发明从光条左右两侧灰度衰减行为的独立量化出发，经衰减因子生成、自适应阈值比对和跨列空间一致性校验的逐层递进，将真实光晕拖尾从光条正常边缘和瞬时形变中准确分离。对被确认存在真实光晕拖尾的一侧边缘像素，基于预设灰度条件进行筛选后实施方向性权重系数压缩，光晕越严重的一侧压缩幅度越大，光条核心区域及正常边缘侧保持原始权重系数不变。经过加权重心计算，光条中心线行坐标不再向光晕拖尾侧偏移，膜厚测量值的准确性与重复性得到有效提升。
Smart Images
Figure CN122961320A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of optical inspection technology. More specifically, this invention relates to an intelligent method for detecting the thickness of the adhesive film sprayed onto the bushing of a stabilizer bar. Background Technology
[0002] On automated production lines for automotive stabilizer bar bushings, hot melt adhesive needs to be sprayed onto the inner wall of the rubber bushing to enhance the bonding strength of subsequent pressing processes. The thickness and uniformity of the adhesive film are key quality indicators determining bonding performance. Currently, line laser triangulation is commonly used in industrial settings for non-contact online film thickness detection. A line laser profile sensor projects a laser line onto the inner wall of the bushing, and a camera captures an image of the laser stripes modulated by the inner wall profile. The centroid method is used to calculate the center line of each laser stripe, extracting the sub-pixel level center line, which corresponds to the height information at that position. The height data collected from the bushing's rotation are stitched together and compared with a pre-calibrated reference profile of the unadhesive-free bushing; the difference between the two is the adhesive film thickness.
[0003] In actual production, hot melt adhesive, after being sprayed from the nozzle, does not completely flow evenly, resulting in irregular, orange-peel-like micro-undulations on the adhesive surface. The bushing may have residual mold release agent oil stains from the previous process, and the adhesive may also contain a very thin liquid film formed by a small amount of unevaporated solvent. When a laser beam is projected onto these localized areas, significant diffuse reflection occurs at the peaks of the orange-peel undulations, while the oil stains and liquid film areas produce localized highlights with near-specular reflection. These physical phenomena collectively cause a diffuse halo with a gray value only slightly higher than the background to appear outside the true boundary of the laser stripe image. When the thickness of the residual oil film in a certain localized area of the bushing's inner wall reaches several micrometers, the width of the halo extending outward from the edge of the main light stripe can reach more than 30% of the normal light stripe width.
[0004] When extracting the center line of a light stripe, the centroid method indiscriminately includes all pixels within the cross-section of the light stripe in the weighted average calculation. Although the individual grayscale levels of low-grayscale pixels in the halo region are much lower than those in the core region of the light stripe, the halo is spatially continuous, and the cumulative grayscale contribution of these pixels will shift the calculated center position of the light stripe towards the halo. In a detection system with a pixel equivalent of 20 micrometers per pixel, when the center line positioning offset reaches one pixel, the film thickness measurement value will produce a systematic offset of 20 micrometers. This offset has systematic characteristics and cannot be eliminated by averaging multiple measurements; it will continue to appear in every frame of the image. The more abundant the orange peel texture and the more oil residue in the area, the more obvious the measurement offset. This can lead to qualified products being misjudged as defective products, or defective products with localized glue thickness defects being pulled back into the qualified range by the measurement value and thus rejected. Summary of the Invention
[0005] To address the problem that existing methods cannot accurately distinguish between the halo tail region and the normal edge transition region of the light stripe when the additional gray energy caused by the halo is included in the centroid calculation, thus leading to the offset of the light stripe centerline and systematic distortion of the film thickness measurement, this invention provides an intelligent detection method for the adhesive film thickness of the bushing stabilizing rod, comprising the following steps: Acquire laser stripe images of the inner wall of the bushing to be tested, and extract grayscale sequences column by column within the pre-labeled region of interest in the laser stripe images and locate the peak positions; Perform differential operations from the peak position to the left and right sides, independently quantize the grayscale decay rate on the left and right sides to generate decay factors, and compare the left and right decay factors with the adaptive threshold respectively. Based on the comparison results, cross-column spatial consistency verification is performed. The weight coefficients of edge pixels on the side confirmed to have real halo trails are compressed, while the original weight coefficients are maintained for the core area of the light stripe and the normal edge side. Based on the compressed weighting coefficients and the original weighting coefficients, a weighted center calculation is performed to obtain the center line coordinates of the light stripe, calculate the film thickness value, and output a judgment signal to drive the sorting execution mechanism.
[0006] Preferably, before locating the peak position, the grayscale sequence needs to be filtered, wherein the filter kernel width is calculated based on the physical spot width of the line laser and the pixel equivalent of the camera.
[0007] Preferably, after filtering, the gray values of a preset number of pixels at both ends of the grayscale sequence are taken and the arithmetic mean is calculated. The calculated arithmetic mean is used as the background base gray value of the corresponding column.
[0008] Preferably, after locating the peak position, an anomaly check is performed on the current column to remove invalid columns. Specifically, the anomaly check involves using half the sum of the background grayscale and the effective grayscale value of the peak as a threshold line. Starting from the peak position, the system searches for the first pixel with a grayscale value less than this threshold line to the left and right. The pixel span between these two pixels is recorded as the half-width ratio (WHM). If the WHM exceeds a preset physical width limit, the column is marked as invalid. If the WHM does not exceed the preset physical width limit, the column is marked as valid.
[0009] Preferably, when performing differential calculations from the peak position to the left and right sides, the transition attenuation region is first determined. Specifically, the peak position is moved pixel by pixel to the left and right sides respectively, and the first-order difference of the gray values of adjacent pixels is calculated. When the absolute value of the first-order difference of two consecutive pixels is less than or equal to 1, it is determined that the gray value change has entered the smooth area. The first position that meets this condition is marked as the preliminary end boundary of the transition zone on that side edge. The peak position is shifted outward by 1 pixel as the starting point of the transition zone. The area between the starting point of the transition zone and the preliminary end boundary is defined as the transition attenuation region.
[0010] Preferably, one way to independently quantize the grayscale decay rates on the left and right sides in the transition decay region to generate a decay factor is to calculate a second-order difference sequence in the transition decay region and extract the maximum value of the absolute value of the second-order difference as the decay factor on that side.
[0011] Preferably, another implementation method for independently quantizing the grayscale decay rates on the left and right sides within the transition decay region to generate a decay factor is as follows: calculate the absolute value of the first-order difference of each pixel within the transition decay region, divide the absolute value by the difference between the peak grayscale and the background base grayscale to obtain the normalized gradient value, sum all the normalized gradient values within the transition decay region to obtain the gradient integral area, and use the gradient integral area as the decay factor on that side.
[0012] Preferably, before comparing the left and right attenuation factors with the adaptive threshold respectively, the adaptive threshold is first determined based on the statistical characteristics of the whole image. Specifically, the process involves: traversing all columns marked as valid in the entire image, merging the left and right attenuation factors of each column into the whole image factor set, calculating the median value of the whole image factor set, and using the median value as the adaptive threshold of the current frame image.
[0013] Preferably, the cross-column space consistency verification specifically involves: comparing the left and right attenuation factors of the current column with the adaptive threshold respectively, shifting the comparison results column by column into the shift register, counting the number of high-level flags representing abnormalities in the shift register, and outputting the suppression enable signal on the corresponding side when the number of high-level flags is greater than or equal to the preset trigger threshold.
[0014] Preferably, before compressing the edge pixels on one side that are confirmed to have a real halo trail, it is first determined whether the edge pixels on that side meet the compression triggering requirements based on a preset grayscale condition. The preset grayscale condition is: the effective grayscale value of the current pixel is lower than the preset ratio of the peak effective grayscale value, wherein the effective grayscale value is the original grayscale value of the current pixel minus the grayscale value of the background base of that column, and the peak effective grayscale value is equal to the peak grayscale value minus the grayscale value of the background base of that column.
[0015] The present invention has the following beneficial effects: This invention starts with the independent quantification of grayscale attenuation behavior on both sides of the light stripe. Through a progressive process of attenuation factor generation, adaptive threshold comparison, and cross-column consistency verification, it accurately separates the true halo tail from the normal edge and instantaneous deformation of the light stripe. For edge pixels on the side confirmed to have a true halo tail, directional weight coefficient compression is applied after filtering based on preset grayscale conditions. The more severe the halo, the greater the compression. The original weight coefficients of the light stripe core area and the normal edge side remain unchanged. After weighted center calculation, the row coordinates of the light stripe centerline no longer shift towards the halo tail side, effectively improving the accuracy and repeatability of film thickness measurements. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating the steps of an intelligent detection method for the adhesive film thickness of a stabilizer bar bushing according to an embodiment of the present invention. Figure 2 This is a comparison chart of detection results provided in an embodiment of the present invention. Detailed Implementation
[0017] To facilitate understanding of the solution, the application scenario of an intelligent detection method for the adhesive film thickness of a stabilizer bar bushing is first introduced here, specifically: A smart detection method for the adhesive film thickness of automotive stabilizer bar bushings is applied to the adhesive spraying inspection station on an automated production line for automotive stabilizer bar bushings. This inspection station is located after the hot melt adhesive spraying station. The bushing to be tested is picked up from the spraying station by a robotic arm and placed into a rotating fixture at the inspection station. Driven by a servo motor, the rotating fixture rotates the bushing one revolution at a constant angular velocity. During rotation, a line laser profile sensor fixedly installed above the inspection station continuously projects laser lines onto the area to be tested on the inner wall of the bushing and collects the reflected laser stripe images. Each frame of laser stripe image collected by the line laser profile sensor is transmitted to an industrial control computer in real time. The industrial control computer, which runs all the algorithm logic of this method, extracts the center line of the laser stripe and calculates the adhesive film thickness for each frame of laser stripe image, generating a pass / fail judgment signal, which is then sent to the PLC controller. Based on the pass / fail determination signal, the PLC controller controls the pneumatic pusher cylinder to extend or remain retracted when the bushing arrives at the sorting position along the conveyor line. This pushes the unqualified bushing into the waste collection channel, while the qualified bushing enters the unloading channel along the conveyor line.
[0018] The following is combined with Figure 1 A specific implementation method for intelligent detection of adhesive film thickness in a stabilizer bar bushing is described, including the following steps: S1: Acquire a laser stripe image of the inner wall of the bushing to be tested, and extract grayscale sequences column by column within the pre-labeled region of interest in the laser stripe image and locate the peak position.
[0019] This application realizes non-contact detection of the thickness of the adhesive film on the inner wall of the bushing based on the principle of line laser triangulation. The basis of line laser triangulation is laser stripe image processing, so it is necessary to acquire laser stripe images first.
[0020] After the inspection station is powered on, the industrial control computer sends a start acquisition command to the line laser profile sensor. After receiving the command, the line laser profile sensor acquires each frame of laser stripe image and transmits the acquired laser stripe image to the industrial control computer in real time.
[0021] After receiving each frame of laser stripe image, the industrial control computer extracts grayscale sequences column by column within a pre-defined region of interest. Each grayscale sequence corresponds to the contour height distribution of a sampling position on the inner wall of the bushing. The position with the highest grayscale value in the sequence is the peak position of the light stripe, which serves as the spatial reference origin for all subsequent differential operations. If the peak position is misaligned, the subsequent division of the transition attenuation region and the calculation of the attenuation factor will also be misaligned. Simultaneously, the grayscale sequence may contain columns that cannot provide effective light stripe cross-sectional information due to double-peak aliasing or light stripe breakage. The abnormal grayscale distribution of these columns will contaminate subsequent calculations of the entire film thickness. Therefore, before proceeding with the differential operation, two preliminary tasks need to be completed: peak positioning and invalid column removal. The accuracy of peak positioning depends on two conditions: first, the grayscale sequence must not carry noise spikes after filtering; second, the effective grayscale value of the peak must not be distorted after the background grayscale is correctly subtracted. Based on this, after extracting the grayscale sequence column by column, filtering is performed first, then the background grayscale is extracted, then the peak position is located, and finally, anomaly verification is performed on the current column.
[0022] The following sections will elaborate on each of the above processing steps.
[0023] S11: Complete the calibration of the region of interest.
[0024] The region of interest (ROI) defines the pixel row range for peak search. If the peak search exceeds this range, false light bars far from the main light bar will be included in the search, causing the peak position to be located on the wrong light bar. Therefore, the ROI must be calibrated before batch detection begins.
[0025] In one specific embodiment, a standard bushing without adhesive is placed at the inspection station, and a line laser profile sensor projects a laser line and acquires an image. On the industrial control computer's visual interface, a manually selected pixel row range corresponding to the physical depth of the bushing's inner wall is defined. The coordinates of the upper and lower boundary rows obtained from this selection are written into the initialization configuration file of the inspection program. During subsequent batch inspections, the industrial control computer reads this range from the configuration file, and peak searches are strictly limited to this range, performed line by line.
[0026] Secondary reflection streaks caused by excessive release agent usually appear far below the main streaks, outside the region of interest, and therefore are not included in the peak search range.
[0027] Step S12: Filter the grayscale sequence.
[0028] The inherent shot noise of the camera is amplified in subsequent difference operations. When the unsmoothed grayscale sequence is directly differenced, noise spikes are mistaken for grayscale attenuation features. Therefore, after extracting the grayscale sequence column by column, the grayscale sequence is filtered.
[0029] In one specific embodiment, the physical spot width is divided by the pixel equivalent to obtain the number of pixels corresponding to the spot, and the smallest odd number greater than the number of pixels corresponding to the spot is taken as the filter kernel width. The physical spot width is the nominal light stripe projection width of the line laser, and the pixel equivalent is the physical size corresponding to a single pixel of the camera. Both are read from the factory calibration file of the line laser contour sensor.
[0030] For each pixel in the grayscale sequence, take the grayscale values of all pixels in the neighborhood with a width equal to the width of the filter kernel and take the arithmetic mean of the grayscale values of the pixel itself as the center. Use the arithmetic mean as the new grayscale value of the pixel after filtering.
[0031] After filtering, the grayscale changes of the light stripe cross section tend to be smoothed, and isolated grayscale abrupt changes caused by shot noise are suppressed.
[0032] Step S13: Extract the background base grayscale.
[0033] The grayscale response generated by ambient light and camera dark current in areas without laser illumination will be superimposed on the effective grayscale energy of the light stripe. If this is not subtracted, all subsequent judgments based on the effective grayscale value will be shifted. Therefore, after filtering, the background base grayscale must be extracted from the filtered grayscale sequence first.
[0034] In one specific embodiment, ten consecutive pixels are taken from the beginning of the filtered grayscale sequence and ten consecutive pixels are taken from the end of the filtered grayscale sequence. The grayscale values of these twenty pixels are then arithmetically averaged, and the arithmetic mean is used as the background base grayscale of that column.
[0035] The first and last ten pixels are located outside the laser beam's illumination range; their grayscale values do not include the energy reflected by the beam and are contributed only by ambient light and the camera's dark current. Because the intensity of ambient light varies at different sampling positions on the inner wall of the bushing, and the camera's dark current response fluctuates across different pixel rows, the average grayscale values of the first and last ten pixels in each column are not the same. Only by independently extracting the background grayscale values column by column can the grayscale response level of each column without laser illumination be reflected.
[0036] Step S14: Locate the peak position.
[0037] The peak position is the starting point for subsequent difference operations on the left and right sides. After the background grayscale is extracted, the peak position is located by traversing the filtered grayscale sequence and comparing each extreme value.
[0038] In one specific embodiment, starting from the beginning row of the region of interest, the algorithm moves downwards pixel by pixel, comparing the grayscale value of the current pixel with the recorded maximum grayscale value, and retaining the row coordinates of the pixel with the larger grayscale value. After traversal, the retained row coordinates are used as the peak position, and the grayscale value corresponding to that position is used as the peak grayscale. The peak grayscale is subtracted from the background base grayscale to obtain the effective peak grayscale value.
[0039] After peak positioning, the column coordinates of the energy center of the light stripe were determined.
[0040] Step S15: Perform anomaly checks on the current column to remove invalid columns.
[0041] Once the peak position is determined, if the cross-section of the current column shows double-peak overlap or broken stripes, that column can no longer provide effective stripe energy distribution information. If the calculation proceeds directly to the subsequent centroid calculation, the abnormal grayscale distribution will contaminate the entire film thickness calculation result. Therefore, it is necessary to perform anomaly verification on the current column.
[0042] In one specific embodiment, half the sum of the background base grayscale and the peak effective grayscale is used as a threshold line. Starting from the peak position, the first pixel with a grayscale value less than this threshold line is searched to the left and right. The pixel span between these two pixels is taken as the half-width at half-height (WHH). If the WHH exceeds a preset physical limit width threshold, the column is marked as invalid. If the WHH does not exceed the preset physical limit width threshold, the column is marked as valid.
[0043] After anomaly verification, erroneous data caused by bimodal overlap or broken light stripes are excluded from subsequent calculations, and the invalid column is not included in the centroid calculation in subsequent steps.
[0044] S2: Perform differential operation from the peak position to the left and right sides, independently quantize the grayscale decay rate of the left and right sides to generate decay factors, and compare the left and right decay factors with the adaptive threshold respectively.
[0045] Once the peak position is determined, the pattern of grayscale drop from the peak to the background on both sides of the light stripe directly determines the subsequent calculation method for the attenuation factor. The grayscale drop at the edge of a normal light stripe is steep, while the grayscale drop at the edge of a halo tail is gentle; the grayscale change rate in the transition attenuation region differs significantly between the two. If grayscale statistics are directly performed on the entire light stripe, this difference will be masked. Therefore, it is necessary to perform differential calculations independently from the peak position to both sides, quantifying the grayscale attenuation rate of each separately.
[0046] Based on this, the algorithm moves pixel by pixel from the peak position to the left and right, calculating the first-order difference to determine the transition attenuation region. Once the transition attenuation region is determined, the attenuation factors on both the left and right sides are calculated independently within this region. After the attenuation factors are generated, an adaptive threshold is determined based on the statistical characteristics of the entire image, and the left and right attenuation factors are compared with this adaptive threshold respectively.
[0047] The following sections will elaborate on each preprocessing step.
[0048] S21: Determine the transition attenuation region.
[0049] The transitional attenuation region is the physical domain for calculating the attenuation factor. If this region is too wide, background noise will be included in the attenuation analysis; if it is too narrow, the tail features of the halo trail will be truncated. Therefore, it is necessary to determine the transitional attenuation region first.
[0050] In one specific embodiment, the system moves pixel by pixel from the peak position to the left and right, calculating the first-order difference between the grayscale values of adjacent pixels. This process continues outwards while calculating the first-order difference. When the absolute value of the first-order difference between two consecutive pixels is less than or equal to one, it is determined that the grayscale change has entered a plateau region. An integer grayscale value and an absolute value of the difference between adjacent pixels less than or equal to one indicate that the grayscale no longer physically undergoes a step change.
[0051] The first position that meets this condition is marked as the initial end boundary of the transition zone on that side. The peak position is shifted outward by one pixel as the start point of the transition zone, and the area between the start point and the initial end boundary is defined as the transition attenuation region. The above operations are performed independently on both the left and right sides.
[0052] The width of the transition attenuation region at the edge of a normal light stripe is significantly narrower than that at the edge of a halo trail. This difference in width provides the boundary of the scope for the subsequent calculation of the attenuation factor.
[0053] S22: Independently quantize the grayscale decay rates on both the left and right sides within the transition decay region to generate a decay factor.
[0054] After the transition attenuation region is determined, the attenuation factor is calculated within this region. Considering the differences in light stripe morphology under different operating conditions, this application provides two calculation paths for the attenuation factor.
[0055] In one specific embodiment, the system moves pixel by pixel from the peak position to the left and right within the transition attenuation region, calculates the second-order difference sequence of gray values of adjacent pixels, and extracts the maximum value of the absolute value of the second-order difference from the second-order difference sequence as the attenuation factor on that side.
[0056] The second-order difference maximum value is larger at normal steep edges, while it is smaller at halo tail edges. Through this path calculation, the attenuation factor directly reflects the curvature characteristics of the area with the most drastic gray-level changes within the transition attenuation region.
[0057] Considering the sensitivity of a single extreme point to noise, in another specific embodiment, the absolute value of the first-order difference of each pixel is calculated in the transition attenuation region. The absolute value of the first-order difference is divided by the difference between the peak gray level and the background base gray level to obtain the normalized gradient value. All normalized gradient values in the transition attenuation region are summed to obtain the gradient integral area, and the gradient integral area is used as the attenuation factor on that side.
[0058] Normal, steep edges have a narrow transition decay region, highly concentrated gradient values, and a small gradient integration area. Halo trailing edges, on the other hand, have a wide transition decay region, a larger number of pixels in the trailing area, more dispersed gradient values, and a larger integration area. Through this path calculation, the decay factor distinguishes normal edges from halo trails from an energy accumulation perspective, effectively eliminating random noise interference.
[0059] S23: Determine the adaptive threshold based on the statistical characteristics of the whole graph, and compare the left and right attenuation factors with the adaptive threshold respectively.
[0060] After the attenuation factor is generated, a criterion needs to be determined to distinguish between normal edges and halo tails. Fixed thresholds are prone to misjudgment under different ambient light conditions, so an adaptive threshold based on the statistical characteristics of the entire image is adopted.
[0061] In one specific embodiment, all valid columns of the entire image are traversed, and the left and right attenuation factors of each valid column are merged into the full image factor set. The median value of the full image factor set is then calculated, and this median value is used as the adaptive threshold of the current frame image.
[0062] After the adaptive threshold is determined, the attenuation factor on the left side is compared with the adaptive threshold. If the attenuation factor on the left side is less than or equal to the adaptive threshold, it is determined that there is a possible halo tail on the left side of the column; if the attenuation factor on the left side is greater than the adaptive threshold, it is determined that the left side of the column is normal. The comparison method for the attenuation factor on the right side with the adaptive threshold is the same as for the left side. If the attenuation factor on the right side is less than or equal to the adaptive threshold, it is determined that there is a possible halo tail on the right side of the column; if the attenuation factor on the right side is greater than the adaptive threshold, it is determined that the right side of the column is normal.
[0063] S3: Combine the comparison results to perform cross-column spatial consistency verification, compress the weight coefficients of edge pixels on the side confirmed to have real halo trailing, and maintain the original weight coefficients for the core area of the light stripe and the normal edge side.
[0064] A low attenuation factor in column light stripes could be caused by genuine halo tailing or by a momentary, natural narrowing of the stripe width due to minute fiber-reflection bumps on the adhesive surface. These two factors differ significantly in spatial distribution. Genuine halo tailing is caused by physical factors such as orange peel texture, oil stains, or liquid films on the adhesive surface, which have a certain spatial distribution. Therefore, genuine halo tailing exhibits a continuous and consistent abnormal attenuation characteristic across multiple adjacent columns. In contrast, the momentary deformation caused by fiber-reflection bumps manifests as isolated abrupt changes in a single column or a very small number of columns. If weighted compression is applied indiscriminately to all columns with low attenuation factors, the momentary deformation will be misjudged as a halo, introducing new calculation biases at normal edges.
[0065] Therefore, after obtaining the comparison results of the single-column attenuation factor and the adaptive threshold, it is necessary to perform cross-column spatial consistency verification between this comparison result and the comparison results of adjacent columns, using the spatial continuity of the real halo tail to distinguish it from isolated instantaneous deformation. After the cross-column spatial consistency verification is completed, the suppression enable signal of the corresponding side is output. This suppression enable signal identifies the side where the real halo tail exists, but not all edge pixels on this side belong to the halo tail. The halo tail is only distributed in the low-to-medium energy region of the light stripe, and the sharp and effective signals in the high-energy region do not belong to the halo tail. Therefore, before the weight coefficient compression, it is necessary to filter the edge pixels on this side based on the preset grayscale conditions, and only the pixels confirmed to belong to the halo tail are subject to weight coefficient compression, while the original weight coefficients are maintained for the core area of the light stripe and the normal edge side.
[0066] Based on this, the comparison results are shifted column by column into a shift register, and the number of high-level flags representing anomalies in the shift register is counted. When the number of high-level flags is greater than or equal to a preset trigger threshold, a suppression enable signal for the corresponding side is output. After obtaining the suppression enable signal, it is determined whether the edge pixels on that side meet the compression triggering requirements based on preset grayscale conditions. If the preset grayscale conditions are met, the edge pixels on the side confirmed to have a real halo trail are compressed using weighted coefficients, while the original weighted coefficients are maintained for the light stripe core area and the normal edge side.
[0067] The following sections will elaborate on each preprocessing step.
[0068] S31: Move the comparison result into the shift register for consistency verification.
[0069] The comparison result between the single-column attenuation factor and the adaptive threshold can only indicate whether the grayscale attenuation on one side of the current column is gradual, but it cannot distinguish whether this gradual attenuation is caused by a real halo tail or by an instantaneous change in the width of the light stripe. A real halo tail appears continuously between multiple adjacent columns, while an instantaneous deformation appears isolatedly only in a single column or a very few columns. Therefore, it is necessary to shift the comparison results column by column into a shift register, and use the shift register to perform a majority vote on the comparison results of multiple consecutive columns. Only when the proportion of suspected halo tails in multiple consecutive columns reaches a preset condition is it confirmed that a real halo tail exists on that side, and a suppression enable signal is output.
[0070] In one specific embodiment, two independent N-level shift registers are defined, where N equals the number of columns confirming morphological continuity. The two shift registers track the continuity of the halo on the left and right sides, respectively. When traversing the image column by column, the comparison result on the left side of the current column is shifted into the left shift register. If the comparison result indicates a suspected halo tail, a high-level flag of 1 is shifted in; if the comparison result is normal, a low-level flag of 0 is shifted in. As new column data is shifted in, the earliest shifted flag is automatically shifted out. The number of high-level flags in the left shift register is counted. When the number of high-level flags is greater than or equal to a preset trigger threshold, a left-side suppression enable signal is output. When the number of high-level flags is less than the preset trigger threshold, the left-side suppression enable signal remains invalid. The operation of the right shift register is completely symmetrical to that of the left. Through majority voting, misjudgments caused by noise or dust in individual columns within a continuous halo region will not interrupt the suppression enable signal, and isolated instantaneous deformations will not trigger the suppression enable signal due to insufficient high-level flags.
[0071] The number of morphological continuity confirmation columns involved in the above embodiments is determined through offline calibration. The specific calibration process is as follows: Before formal batch testing, multiple typical images containing different types and degrees of halo phenomena are collected, and the column range where the halo trail is located is manually marked on each image. Different candidate values are used as the number of morphological continuity confirmation columns, and the detection algorithm is run separately to obtain their respective automatically determined suppression regions. The automatically determined suppression regions are compared with the manually marked column ranges, and the number of overlapping columns and the number of falsely affected columns in the automatically determined suppression regions that exceed the manually marked column ranges are counted. The candidate value with the highest number of overlapping columns and the lowest number of falsely affected columns is selected as the fixed parameter value for the number of morphological continuity confirmation columns and written into the initialization configuration file of the detection program. If two candidate values have the same number of overlapping columns, the candidate value with fewer falsely affected columns is selected. In actual production lines, the number of morphological continuity confirmation columns is usually between 3 and 5.
[0072] S32: Determine whether the edge pixels on this side meet the compression triggering requirements based on preset grayscale conditions.
[0073] Consistency checks confirm the spatial continuity of the halo trail, but not all pixels on that side are affected by the halo trail. Even if the attenuation factor is small, sharp, effective signals in high-energy regions are not considered halo trails. Therefore, filtering based on preset grayscale conditions is necessary.
[0074] In one specific embodiment, the original grayscale value of each pixel on the side edge is obtained, and the background base grayscale value of that column is subtracted from the original grayscale value to obtain the effective grayscale value of the pixel. The peak grayscale value is obtained, and the background base grayscale value of that column is subtracted from the peak grayscale value to obtain the peak effective grayscale value. The effective grayscale value is compared with a preset ratio of the peak effective grayscale value. If the effective grayscale value is lower than the preset ratio of the peak effective grayscale value, the pixel meets the compression triggering requirement; if the effective grayscale value is higher than or equal to the preset ratio of the peak effective grayscale value, the pixel does not meet the compression triggering requirement.
[0075] Pixels that meet the compression triggering requirements are located in the low-energy region of the light stripe, while pixels that do not meet the requirements are located in the high-energy region. Subsequent weighted coefficient compression only applies to pixels that meet the compression triggering requirements.
[0076] S33: Compress the weight coefficients of edge pixels that meet the conditions.
[0077] Pixels that meet the compression triggering requirements are located in the low-energy region of the light stripe. The grayscale contribution of these pixels includes spurious components of the halo tail, and their weight coefficients need to be compressed. The compression logic is that the more severe the halo tail on one side, the smaller its attenuation factor, and the lower the weight coefficient of the pixels on that side should be compressed.
[0078] In one specific embodiment, the initial weight coefficient of each pixel is first set to a value of one. For left-edge pixels that output a left-side suppression enable signal and meet the compression triggering requirements, the left-side attenuation factor is used as the numerator, and the sum of the right-side attenuation factor and the minimum protection constant is used as the denominator. The ratio of the numerator to the denominator is calculated. This ratio is compared with the constant one, and the smaller value is taken as the attenuation coefficient of the left-edge pixel. The weight coefficient of the left-edge pixel is then modified from one to this attenuation coefficient.
[0079] For right edge pixels that output right-side suppression enable signals and meet compression trigger requirements, the weight coefficients are set using the same method.
[0080] For pixels at peak positions, the weighting coefficient remains constant at one, regardless of the state of the suppression enable signals on the left and right sides. For the core region of the light stripe and the normal edge side, the initial weighting coefficient remains unchanged.
[0081] The physical meaning of the attenuation coefficient is that the smaller the attenuation factor on one side relative to the other side, the closer the calculated ratio is to zero, the lower the weight coefficient of the pixel on that side is compressed, and the more significantly its grayscale contribution is reduced. When the attenuation factors on the left and right sides are close, the calculated ratio is close to one, the attenuation coefficient is close to one, and the weight coefficient is basically not compressed. If the attenuation factor on one side is greater than that on the other side, the calculated ratio is greater than one, and the smaller of the two values is used to limit the attenuation coefficient to one, while the weight coefficient remains unchanged. After the weight coefficient is compressed, the false grayscale contribution of the halo trail side is reduced proportionally, while the grayscale contribution of the normal edge side and the core area of the light stripe remains intact.
[0082] S4: Based on the compressed weighting coefficients and the original weighting coefficients, perform weighted center calculation to obtain the center line coordinates of the light stripe, calculate the film thickness value, and output a judgment signal to drive the sorting execution mechanism.
[0083] After step S3 completes the weighting coefficient compression, the grayscale contribution of the pixels on the halo trail side has been weakened, but the final position of the light stripe centerline is still undetermined. The row coordinates of the light stripe centerline are the only basis for calculating the film thickness value. Without these coordinates, subsequent film thickness calculations and qualification judgments cannot be performed.
[0084] Therefore, based on the compressed weighting coefficients output in step S3 and the original weighting coefficients retained, it is necessary to perform a weighted center calculation on the cross-section of the light stripe to obtain the row coordinates of the light stripe centerline. After obtaining the row coordinates of the light stripe centerline, they are compared with the reference row coordinates of the adhesive-free bushing to calculate the film thickness value at the current sampling position. Then, the thickness distribution data of the entire circle is judged to be qualified, and the judgment signal is output to drive the sorting execution mechanism to complete the physical sorting.
[0085] The following sections will elaborate on each processing step.
[0086] S41: Perform weighted centering calculation to obtain the centerline coordinates of the light stripe.
[0087] The compressed weighting coefficients and the preserved original weighting coefficients together constitute the weight distribution of the light stripe cross section, but this weight distribution itself cannot directly give the position of the light stripe centerline. The position of the light stripe centerline needs to be determined by weighted centerline calculation. The logic of weighted centerline calculation is to use the effective gray value of each pixel as the energy weight and the row coordinate as the spatial position, and combine the energy weight and the spatial position to obtain the row coordinates of the light stripe energy center.
[0088] In one specific embodiment, the effective grayscale value of each pixel in the effective column is obtained. The effective grayscale value of each pixel is multiplied by its weight coefficient, and all products are summed to obtain a weighted grayscale sum. The effective grayscale value, its weight coefficient, and its row coordinates are multiplied together, and all products are summed to obtain a weighted coordinate sum. The weighted coordinate sum is divided by the weighted grayscale sum to obtain the row coordinates of the light bar centerline, which reflect the energy center position of the current column of light bars.
[0089] The contribution of the halo pixels on the compressed side to the weighted grayscale sum and weighted coordinate sum decreases proportionally, and no longer shifts the center line to that side.
[0090] S42: Calculate the film thickness value based on the centerline coordinates of the light stripe.
[0091] Once the row coordinates of the light stripe centerline are determined, only the position of the energy center of the current column of light stripes is obtained, but this position is not directly equal to the adhesive film thickness value. The calculation of the adhesive film thickness value requires using the bushing inner wall contour in the uncoated state as a reference, and obtaining it by comparing the difference between the current measured row coordinates and the reference row coordinates.
[0092] In one specific embodiment, during batch testing, the row coordinates of the center line of the light stripe in the current valid column are obtained. These row coordinates are then compared with the reference row coordinates of the adhesive-free bushing, and the difference between them is calculated. This difference is multiplied by the pixel equivalent to obtain the adhesive film thickness value at the current valid column. This process is repeated column by column to obtain the adhesive film thickness values at all valid columns after one revolution of the bushing.
[0093] The reference row coordinates of the glue-free bushings used above were obtained through calibration during the equipment commissioning phase. The specific calibration method is as follows: using a standard bushing without adhesive spraying, perform the complete process from step S1 to step S41, and store the obtained row coordinates of the center lines of each column of light stripes as reference values.
[0094] S43: Determine the quality of the thickness distribution data of the entire circle and output a determination signal to drive the sorting execution mechanism.
[0095] A single thickness value can only reflect the film condition at a local sampling point and cannot directly determine whether the entire bushing is qualified. The thickness distribution data of the entire circle may contain multiple abnormalities such as local overthickness, local underthickness, and overall offset. These abnormalities need to be processed by a unified qualification judgment logic before the judgment signal driving the sorting actuator can be output.
[0096] In one specific embodiment, the film thickness values corresponding to all valid columns collected after one revolution of the bushing are stitched together to obtain the film thickness distribution data for the entire circle. It is determined whether each film thickness value in the entire circle film thickness distribution data exceeds the acceptable range, and the number of sampling points exceeding the acceptable range is counted. If the number of sampling points exceeding the acceptable range is greater than a preset upper limit for the number of defective points, it is judged as unacceptable, and an unacceptable judgment signal is output. The arithmetic mean of the entire circle film thickness distribution data is calculated. If the arithmetic mean exceeds the acceptable range, it is also judged as unacceptable, and an unacceptable judgment signal is output. If neither of the two judgment conditions is met, an acceptable judgment signal is output.
[0097] The pass / fail signal is sent to the PLC controller. Upon receiving the fail signal, the PLC controller extends the pneumatic pusher cylinder as the bushing reaches the sorting position on the conveyor line, pushing the fail-quality bushing into the waste collection channel. Upon receiving the pass signal, the PLC controller keeps the pneumatic pusher cylinder in the retracted state, and the bushing enters the unloading channel along the conveyor line. After pass / fail judgment and signal output, the pass and fail bushings are physically separated.
[0098] To demonstrate the effectiveness of the solution, relevant experiments were conducted. Below are the images obtained from the experiments: Figure 2 The image shows a comparison of the detection results. It can be seen from the image that when the laser stripe is in the non-halo area, the extraction results of both the present method and the traditional method can fit the main light stripe well. However, when entering the halo trailing area, due to the accumulation of energy in the downward diffuse reflection trailing, the trajectory of the traditional centroid method becomes severely out of control and drifts downward as a whole. In contrast, the trajectory of the present method remains smooth and stable, fitting the true center line without being affected by the halo.
[0099] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for intelligent detection of adhesive film thickness in a stabilizer bar bushing, characterized in that, include: Acquire laser stripe images of the inner wall of the bushing to be tested, and extract grayscale sequences column by column within the pre-labeled region of interest in the laser stripe images and locate the peak positions; Perform differential operations from the peak position to the left and right sides, independently quantize the grayscale decay rate on the left and right sides to generate decay factors, and compare the left and right decay factors with the adaptive threshold respectively. Based on the comparison results, cross-column spatial consistency verification is performed. The weight coefficients of edge pixels on the side confirmed to have real halo trails are compressed, while the original weight coefficients are maintained for the core area of the light stripe and the normal edge side. Based on the compressed weighting coefficients and the original weighting coefficients, a weighted center calculation is performed to obtain the center line coordinates of the light stripe, calculate the film thickness value, and output a judgment signal to drive the sorting execution mechanism.
2. The intelligent detection method for the adhesive film thickness of a stabilizer bar bushing according to claim 1, characterized in that, Before locating the peak position, the grayscale sequence needs to be filtered. The width of the filter kernel is calculated based on the physical spot width of the line laser and the pixel equivalent of the camera.
3. The intelligent detection method for the adhesive film thickness of the bushing of the stabilizer bar according to claim 2, characterized in that, After filtering, the gray values of a preset number of pixels at both ends of the gray sequence are taken and the arithmetic mean is calculated. The calculated arithmetic mean is used as the background base gray value of the corresponding column.
4. The intelligent detection method for the adhesive film thickness of the bushing of the stabilizer bar according to claim 3, characterized in that, After locating the peak position, anomaly detection is performed on the current column to remove invalid columns. Specifically, the anomaly detection involves using half the sum of the background grayscale and the effective grayscale value of the peak as a threshold line. Starting from the peak position, the system searches for the first pixel with a grayscale value less than this threshold line to the left and right. The pixel span between these two pixels is recorded as the half-width ratio (WHM). If the WHM exceeds a preset physical width limit, the column is marked as invalid. If the WHM does not exceed the preset physical width limit, the column is marked as valid.
5. The intelligent detection method for the adhesive film thickness of the bushing of the stabilizer bar according to claim 1, characterized in that, When performing differential operations from the peak position to the left and right, the transition attenuation region is first determined. Specifically, the operation involves moving pixel by pixel from the peak position to the left and right, calculating the first-order difference of the gray values of adjacent pixels, and determining that the gray value change has entered a smooth area when the absolute value of the first-order difference of two consecutive pixels is less than or equal to 1. The first position that meets this condition is marked as the preliminary end boundary of the transition zone on that side. The peak position is shifted outward by 1 pixel as the starting point of the transition zone, and the area between the starting point of the transition zone and the preliminary end boundary is defined as the transition attenuation region.
6. The intelligent detection method for the adhesive film thickness of the bushing of the stabilizer bar according to claim 5, characterized in that, One way to generate an attenuation factor by independently quantizing the grayscale attenuation rates on both sides within the transition attenuation region is to calculate a second-order difference sequence within the transition attenuation region and extract the maximum value of the absolute value of the second-order difference as the attenuation factor on that side.
7. The intelligent detection method for the adhesive film thickness of the bushing of the stabilizer bar according to claim 5, characterized in that, Another way to generate a decay factor by independently quantizing the grayscale decay rates on the left and right sides within the transition decay region is as follows: calculate the absolute value of the first-order difference of each pixel within the transition decay region, divide the absolute value by the difference between the peak grayscale and the background base grayscale to obtain the normalized gradient value, sum all the normalized gradient values within the transition decay region to obtain the gradient integral area, and use the gradient integral area as the decay factor on that side.
8. The intelligent detection method for the adhesive film thickness of the bushing of the stabilizer bar according to claim 1, characterized in that, Before comparing the left and right attenuation factors with the adaptive threshold respectively, the adaptive threshold is first determined based on the statistical characteristics of the whole image. Specifically, the process involves traversing all columns marked as valid in the entire image, merging the left and right attenuation factors of each column into the whole image factor set, calculating the median value of the whole image factor set, and using the median value as the adaptive threshold of the current frame image.
9. The intelligent detection method for the adhesive film thickness of the bushing of the stabilizer bar according to claim 1, characterized in that, The cross-column space consistency check is specifically as follows: the left and right attenuation factors of the current column are compared with the adaptive threshold respectively, the comparison results are shifted into the shift register column by column, the number of high-level flags representing abnormalities in the shift register is counted, and when the number of high-level flags is greater than or equal to the preset trigger threshold, the corresponding suppression enable signal is output.
10. The intelligent detection method for the adhesive film thickness of a stabilizer bar bushing according to claim 1, characterized in that, Before compressing the edge pixels on one side that are confirmed to have a real halo trail, it is first determined whether the edge pixels on that side meet the compression triggering requirements based on a preset grayscale condition. The preset grayscale condition is: the effective grayscale value of the current pixel is lower than the preset ratio of the peak effective grayscale value, wherein the effective grayscale value is the original grayscale value of the current pixel minus the background base grayscale value of that column, and the peak effective grayscale value is equal to the peak grayscale value minus the background base grayscale value of that column.