Intelligent automatic visual inspection system and method for automobile brake disc

Abstract

The invention relates to the field of visual inspection, and discloses an intelligent automatic visual inspection system and method for an automobile brake disc, and the method comprises the steps: carrying out the time-sharing sampling imaging of a friction working surface of the brake disc through a multi-station imaging device when the brake disc enters a detection station along with a conveying mechanism; analyzing and separating the main texture direction, periodic distribution and continuity characteristics in the image group around regular turning textures formed in the brake disc machining process; performing stratified analysis on abnormal signs by combining distribution characteristics and gray disturbance amplitude differences of residual information under different spatial scales; non-structural fluctuation caused by local reflection or imaging noise is eliminated by analyzing the difference between the texture extension trend and the gray scale stability of adjacent areas; and comprehensively judging the distribution position and the influence range of the suspicious area on the working surface in combination with the reference position of the brake disc structure. The method has the advantage of improving the surface quality detection accuracy of the brake disc.

Description

Technical Field

[0001] This invention relates to the field of visual inspection, specifically to an intelligent automatic visual inspection system and method for automotive brake discs. Background Technology

[0002] On mass production lines for automotive brake discs, after finishing, brake discs typically undergo online appearance quality inspection to quickly determine the surface condition of their friction surfaces. In existing inspection scenarios, to balance inspection cycle time and equipment cost, vision inspection systems often employ single-station, fixed-viewpoint two-dimensional imaging to analyze surface images of the brake disc in rotating or stationary states. However, in this specific scenario, the brake disc working surface often exhibits subtle abnormal textures with small local scales and limited grayscale variations, such as shallow scratches caused by momentary tool jumps, slight clamping eccentricity, or brief contact with foreign objects. These anomalies are highly similar in morphology to normal machining textures, and their grayscale distribution variations are often masked by the overall surface texture. On the one hand, increasing the threshold or filtering intensity to suppress background textures can easily lead to the smoothing out of subtle anomalies and missed detection; on the other hand, lowering the threshold to enhance anomaly response can misidentify normal machining textures as defects, affecting the stability of the inspection results. Therefore, it is essential to design an intelligent automatic vision inspection system and method for automotive brake discs that improves the accuracy of brake disc surface quality inspection. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an intelligent automatic visual inspection system and method for automotive brake discs, which has the advantage of improving the accuracy of brake disc surface quality inspection and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goal of improving the accuracy of brake disc surface quality inspection, this invention provides the following technical solution: an intelligent automatic visual inspection method for automotive brake discs, comprising the following steps: When the brake disc enters the inspection station with the conveying mechanism, the multi-station imaging device is used to perform time-division sampling imaging of the friction working surface of the brake disc, verify the coverage relationship of the imaging area, and construct an image group that reflects the gray-scale change characteristics of the same surface area under different observation conditions. Based on the regular turning texture formed during the brake disc machining process, the main texture direction, periodic distribution and continuity features in the image group are analyzed and separated. Texture components consistent with the machining process are backgrounded, and residual information that deviates locally from the texture distribution is retained. By combining the distribution characteristics of residual information at different spatial scales and the differences in gray-scale perturbation amplitude, the abnormal signs are analyzed in layers, and the phenomena of brightness abrupt change, texture interruption and direction shift are compared in each layer to form a set of regional features reflecting the weak quality fluctuations on the brake disc surface. By introducing neighborhood association constraints into the regional feature set, and analyzing the differences between adjacent regions in terms of texture extension trend and grayscale stability, non-structural fluctuations caused by local reflection or imaging noise are eliminated, and suspicious regions that do not conform to the normal processing state are identified. Based on the reference position of the brake disc structure, the distribution location and influence range of the suspicious area on the working surface are comprehensively judged, and the test results for grading the surface quality of the brake disc are output.

[0005] Preferably, the process of constructing an image set reflecting the grayscale variation characteristics of the same surface region under different observation conditions is as follows: As the brake disc rotates at a constant speed with the conveying mechanism and enters the detection station, multiple imaging stations are triggered to simultaneously acquire images according to a preset angle interval. Based on the geometric center and outer edge contour of the brake disc, spatial registration is performed on the images acquired at each workstation to determine the corresponding surface coverage area under different viewpoints. Surface regions with overlapping coverage are indexed and identified. Gray-scale information acquired under different imaging conditions is merged according to time sequence and viewing angle parameters to construct an image group that reflects the gray-scale variation characteristics of the same surface region under different observation conditions.

[0006] Preferably, the process of analytically separating the main texture direction, periodic distribution, and continuous features in the image group is as follows: Perform directional gradient statistics and spectral decomposition on the image group to extract directional feature components that reflect the main direction of the turning texture; Based on the main direction feature components, the periodic distribution characteristics of the texture in the radial and circumferential directions are calculated to establish a periodic description model of regular textures. By analyzing the continuity and breakage of texture lines in space, we can distinguish between continuous texture structures that conform to the processing technology rules and discontinuous texture structures that have abnormal disturbances.

[0007] Preferably, the process of retaining residual information that deviates locally from the texture distribution is as follows: Based on the direction and periodicity model of regular texture, the consistent texture region in the image is fitted and reconstructed to generate a textured background image; The original image and the textured background image are differentially processed to eliminate the interference of irregular machining texture on anomaly detection; The difference results are normalized to obtain residual information that highlights local deviation characteristics.

[0008] Preferably, the process of performing stratified analysis on abnormal signs is as follows: The residual information map is segmented and statistically analyzed at multiple spatial scales, and the gray-level perturbation intensity and distribution density at each scale are calculated respectively. Based on the different disturbance amplitudes and spatial coverage, the abnormal signs are divided into a weak disturbance layer, a local anomaly layer, and a significant anomaly layer, and corresponding response weights are assigned to the abnormal signs at different levels.

[0009] Preferably, the process of forming a set of regional features reflecting the slight quality fluctuations on the brake disc surface is as follows: In each anomaly level, regional features are extracted, including brightness change gradient, texture connectivity change, and texture direction offset. Regional features are jointly described within the same spatial region to form a regional feature vector with multiple features superimposed. Consistency verification is performed on the regional feature vectors to form a set of regional features that reflect the slight quality fluctuations on the brake disc surface.

[0010] Preferably, the process of analyzing the differences in texture extension trends and grayscale stability between adjacent regions is as follows: Using each region in the set of regional features as a node, construct a regional adjacency structure based on spatial adjacency relationships; Compare the texture extension trend and grayscale stability between adjacent regions. When the difference in statistical characteristics between the target region and the adjacent region in at least one of the texture extension trend or grayscale stability indicators exceeds a preset threshold, the anomaly confidence of the target region is increased.

[0011] Preferably, the process for identifying suspicious areas that do not conform to normal processing conditions is as follows: By combining the grayscale consistency of the same area under multi-position imaging conditions, abnormal responses that only appear in a single viewpoint are suppressed. The morphological continuity and directional consistency of abnormal areas are judged, and random noise responses that do not have processing trace characteristics are eliminated; Regions that exhibit stable anomalous characteristics under multiple observation conditions are retained as suspicious regions.

[0012] Preferably, the process for outputting the test results used for grading the surface quality of brake discs is as follows: Based on the center hole of the brake disc, the range of the friction radius, and the mounting reference surface, the location of the suspicious area is structurally located; Analyze the distribution range of the suspicious areas in the radial and circumferential directions and the degree of potential impact on the brake contact surface; Based on the number, distribution concentration, and impact range of suspicious areas, the surface quality of the brake disc is graded and assessed, and the corresponding test results are output.

[0013] An intelligent automotive brake disc automatic vision inspection system includes: Multi-tasking imaging module: During the process of the brake disc entering the inspection station, it completes time-division imaging acquisition of the friction working surface, verifies the coverage relationship of each imaging field of view, and constructs an image group; Texture separation module: Analyzes the regular turning texture features on the surface of the brake disc, performs background processing on texture components consistent with the machining process, and extracts residual information with deviations; Anomaly comparison module: Combining the distribution characteristics of residual information at different spatial scales, it performs hierarchical comparison of brightness, texture and orientation changes to form a set of regional features. Region filtering module: Introduces neighborhood association constraints to eliminate unstructured fluctuations caused by reflection or imaging noise and identify suspicious regions that do not conform to the normal processing state; Quality assessment module: comprehensively assesses the distribution and impact range of suspicious areas, and outputs the test results of brake disc surface quality grading.

[0014] Compared with the prior art, the present invention provides an intelligent automatic visual inspection system and method for automotive brake discs, which has the following beneficial effects: This invention improves the observability of subtle anomalies by leveraging multi-station time-division imaging and regional coverage verification to fully utilize the differences in grayscale and response of the same surface area under different observation conditions. By analyzing and separating the main texture direction, periodicity, and continuity based on the turning texture formed around the processing technology, it effectively suppresses the interference of normal processing textures on the detection results, making the abnormal residual information more prominent. Combined with multi-spatial-scale hierarchical analysis and regional feature construction, it can collaboratively identify various subtle quality fluctuations such as brightness abrupt changes, texture interruptions, and directional shifts, enhancing the detection capability for early and hidden defects. The introduction of neighborhood association constraints and structural consistency discrimination effectively eliminates non-realistic anomalies caused by local reflections, random noise, or imaging instability, improving the stability and reliability of the detection results. By combining suspicious areas with the brake disc structural benchmark and the actual working area, it achieves a comprehensive assessment of the location and potential impact range of anomalies, outputting intuitive and interpretable surface quality grading results. This meets the comprehensive requirements of accuracy, repeatability, and engineering applicability for automatic inspection on production lines, reducing manual re-inspection costs and improving the quality control level of brake discs. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method of the present invention; Figure 2 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0016] 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.

[0017] Example 1: Please refer to Figure 1 As shown in the embodiment of the present invention, an intelligent automatic visual inspection method for automotive brake discs includes the following steps: S1: When the brake disc enters the inspection station with the conveying mechanism, the multi-station imaging device is used to perform time-division sampling imaging of the brake disc friction working surface, and the coverage relationship of the imaging area is verified to construct an image group that reflects the gray-scale change characteristics of the same surface area under different observation conditions.

[0018] The process of constructing an image set in S1 that reflects the grayscale variation characteristics of the same surface region under different observation conditions is as follows: As the brake disc rotates at a constant speed with the conveyor and enters the inspection station, multiple imaging stations are triggered to simultaneously acquire images at preset angle intervals. When the brake disc enters the inspection station, multiple imaging stations arranged on both sides or around the conveyor line perform time-division sampling imaging of the brake disc's friction surface. The conveyor drives the brake disc to rotate at a constant speed, and a position encoder or rotation angle sensor acquires the brake disc's rotation angle information in real time. Based on preset angle interval parameters, the imaging system triggers the corresponding imaging station to acquire images when the brake disc rotates to the corresponding angle position, allowing different stations to acquire image information of the brake disc surface from different observation angles. This enables the acquisition of multi-view image data covering the same friction surface area without stopping the conveyor. Based on the geometric center and outer edge contour of the brake disc, spatial registration is performed on the images acquired at each workstation to determine the corresponding surface coverage area under different viewpoints. For images acquired from multiple imaging workstations, the geometric center position and outer edge contour of the brake disc are used as a unified structural reference to perform spatial registration processing on the images from each workstation. The outer contour of the brake disc is extracted by edge detection and circular fitting, and its center coordinates and radius parameters are calculated to establish a unified coordinate system for the brake disc in the image. Images from different workstations and different viewpoints are mapped to the same polar coordinate or unfolded coordinate space to eliminate positional offsets caused by differences in imaging angles. Through spatial registration, the actual brake disc surface area corresponding to each pixel in different images can be clearly identified, thereby determining the surface coverage relationship between images under different viewpoints. Surface regions with overlapping coverage are indexed and identified. Gray-scale information acquired under different imaging conditions is merged according to time sequence and viewing angle parameters to construct an image group reflecting the gray-scale variation characteristics of the same surface region under different observation conditions. Based on the registered coordinate mapping results, pixel regions belonging to the same actual surface location are assigned the same region index number. For regions with the same index number, their corresponding gray-scale information at different imaging positions and different acquisition times is extracted and merged in a unified manner with the image acquisition timestamp and imaging viewing angle parameters. The gray-scale information of the same surface region acquired under different observation conditions is organized into a set of associated image data to form an image group used to describe the gray-scale variation characteristics of the region.

[0019] S2: Based on the regular turning texture formed during the brake disc machining process, the main texture direction, periodic distribution and continuity features in the image group are analyzed and separated. Texture components consistent with the machining process are backgrounded, and residual information that deviates locally from the texture distribution is retained.

[0020] The process of analytically separating the main texture direction, periodic distribution, and continuous features in the image group in S2 is as follows: The image group is subjected to directional gradient statistics and spectral decomposition to extract directional feature components that reflect the main direction of the turning texture. In each image, the gradient response of pixel gray level in multiple preset directions is calculated, and the gradient magnitude is statistically summarized to form a directional histogram that reflects the energy distribution of the texture direction. After the image is unfolded along the radial and circumferential directions, frequency domain decomposition processing is performed to convert the periodic texture in the spatial domain to the frequency domain representation. By analyzing the energy concentration direction in the frequency domain, the main direction component corresponding to the regular texture formed by turning is extracted. Based on the main direction feature components, the periodic distribution characteristics of the texture in the radial and circumferential directions are calculated, and a periodic description model of regular texture is established. After obtaining the main texture direction feature components, the image group is oriented and rearranged based on this direction, and the texture information is projected to the radial and circumferential directions respectively. For the projected gray-level sequence, the spacing change between adjacent texture peaks and valleys is calculated, and its distribution in different spatial positions is statistically analyzed, so as to obtain the periodic characteristics of the texture in the radial and circumferential directions. By analyzing the stability, repeatability and variation range of the period length, a periodic model for describing regular turning texture is established, so that the model can characterize the typical distribution range of texture period under normal processing conditions. By analyzing the continuity and breakage of texture lines in space, continuous texture structures that conform to the processing technology rules are distinguished from discontinuous texture structures with abnormal disturbances. The extension path of texture lines is tracked along the main texture direction to evaluate their connectivity between adjacent pixels or adjacent regions. The interruption positions, length changes, and offsets of texture lines are statistically analyzed. When texture lines exhibit continuous, smooth extension characteristics in space and conform to the periodic model constraints, they are judged as continuous texture structures that conform to the processing technology rules. When texture lines show obvious breaks, misalignments, or local disorder and do not conform to the periodic model constraints, they are classified as discontinuous texture structures with abnormal disturbances. This achieves effective differentiation between regular turning textures and abnormal texture components.

[0021] The process of preserving residual information that deviates locally from the texture distribution in S2 is as follows: Based on the direction and periodic model of regular texture, the consistent texture region in the image is fitted and reconstructed to generate a texture background image; the image is sampled directionally along the main texture direction, and the gray-level change trend of each sampling position is smoothly fitted under the constraint of the periodic model, so as to reconstruct the texture gray-level distribution that conforms to the normal processing technology. For the region that is determined to be a regular texture structure, the corresponding texture background pixel value is generated by the continuous texture model. For the region that does not participate in the fitting, the original gray-level information is kept unchanged, and a texture background image that mainly reflects the normal processing texture characteristics of the brake disc is obtained. The original image and the textured background image are differentially processed to eliminate the interference of irregular machining texture on anomaly detection. The difference between the original gray value and the textured background gray value is calculated for each corresponding pixel position. This makes the regions that are highly consistent with the regular texture model approach zero response in the difference result, while the regions that deviate from the model produce obvious residual response. Through this difference processing, the periodic gray value changes caused by normal machining texture can be effectively eliminated, and only gray value disturbance information that may be caused by surface defects, machining anomalies or local damage is retained. The difference results are normalized to obtain residual information that highlights local deviation features. Based on the statistical distribution of residual amplitude in the whole residual image, the residuals are linearly or piecewise normalized to map them to a uniform numerical range. At the same time, the influence of extreme values ​​caused by imaging noise or local brightness fluctuations is suppressed. The residual information after normalization can better highlight the regional features that deviate from the regular texture distribution.

[0022] S3: Combining the distribution characteristics of residual information at different spatial scales and the differences in grayscale perturbation amplitude, the abnormal signs are analyzed in layers, and the phenomena of brightness abrupt change, texture interruption and direction shift are compared in each layer to form a set of regional features reflecting the weak quality fluctuations on the brake disc surface.

[0023] The process of stratified analysis of abnormal signs in S3 is as follows: The residual information map is segmented and statistically analyzed at multiple spatial scales, and the gray-level perturbation intensity and distribution density at each scale are calculated. Based on the geometric dimensions and texture feature dimensions of the brake disc friction working surface, different sizes of analysis windows are set, and the residual information map is traversed block by block. At each spatial scale, the average perturbation intensity, maximum perturbation amplitude, and number of significant residual pixels per unit area of ​​the residual gray level in each segment are calculated to characterize the strength and spatial distribution density of the abnormal gray-level perturbation at that scale. Through multi-scale segmented statistics, small, local, and large-scale abnormal signs can be captured simultaneously. Based on the different disturbance amplitudes and spatial coverage, anomalous signs are divided into a weak disturbance layer, a local anomaly layer, and a significant anomaly layer, and corresponding response weights are assigned to different levels of anomalous signs. Anomalies with small disturbance amplitudes, scattered distributions, and only appearing within a small-scale window are classified as weak disturbance layers. Anomalies that are continuously distributed within a medium-scale window and have disturbance intensity significantly higher than background noise are classified as local anomaly layers. Anomalies that form patches within a large-scale window and have disturbance amplitudes significantly higher than normal texture fluctuations are classified as significant anomaly layers. Different response weights are assigned to different levels of anomalous signs for regional feature fusion and anomaly confidence calculation, which can distinguish surface quality fluctuations of different severity.

[0024] The process of forming a set of regional features reflecting the weak mass fluctuations on the brake disc surface in S3 is as follows: In each anomaly level, regional features are extracted, including brightness change gradient, texture connectivity change, and texture direction offset. Within each anomaly level, the brightness change gradient is calculated for the corresponding region to characterize the magnitude and direction of grayscale abrupt changes within the region. Based on the texture continuity analysis results, the connectivity change of texture lines within the region is evaluated to reflect whether there are interruptions or disorder in the texture. The offset between the actual texture direction and the regular texture direction within the region is calculated along the main texture direction to quantify the degree of texture direction offset. Through feature extraction, the anomaly region can be comprehensively characterized from three dimensions: brightness, structure, and direction. Regional features are jointly described within the same spatial region to form a regional feature vector with multiple features superimposed. Multiple features extracted within the same region are combined according to a unified feature dimension order, and the response weights corresponding to the anomaly level to which the region belongs are combined to weight each feature component to form a regional feature vector with multiple features superimposed. Different types of anomaly features can be uniformly expressed in the same feature space. Consistency checks are performed on the regional feature vectors to form a set of regional features reflecting the slight quality fluctuations on the brake disc surface. By comparing the changes in the corresponding feature vectors of the same region under adjacent imaging views or adjacent time sampling, the stability of the feature vectors in terms of numerical amplitude and structural composition is evaluated. When the feature vector of a certain region remains relatively consistent in multiple observations, it is retained as a valid regional feature. When the feature vector fluctuates drastically and lacks consistency, it is discarded. The regional feature vectors retained after consistency checks together constitute a set of regional features reflecting the slight quality fluctuations on the brake disc surface.

[0025] S4: Introduce neighborhood association constraints to the regional feature set. By analyzing the differences between adjacent regions in terms of texture extension trend and grayscale stability, eliminate unstructured fluctuations caused by local reflection or imaging noise, and identify suspicious regions that do not conform to the normal processing state.

[0026] The process in S4 that analyzes the differences in texture extension trends and grayscale stability between adjacent regions is as follows: Using each region in the set of regional features as a node, a regional adjacency structure based on spatial adjacency is constructed; based on the position coordinates of the region in the registration coordinate system, its adjacent regions in the radial and circumferential directions are determined, regions with shared boundaries or distances less than a preset spatial threshold are identified as adjacent nodes, and adjacency relationships are established between nodes to form a regional adjacency structure that reflects the spatial adjacency relationship between regions. By comparing the texture extension trend and grayscale stability between adjacent regions, when the statistical feature difference between the target region and its adjacent regions in at least one of the texture extension trend or grayscale stability indicators exceeds a preset threshold, the anomaly confidence of the target region is increased. For each target region, the statistical difference between it and its adjacent regions in terms of texture extension trend and grayscale stability indicators is calculated. Texture extension trend is used to characterize the spatial extension consistency of texture lines, and grayscale stability is used to characterize the smoothness of grayscale changes within the region. When the statistical feature difference between the target region and its adjacent regions in at least one of the above two indicators exceeds a preset threshold, it is determined that the target region has an abnormal enhancement phenomenon relative to its neighboring regions, and the anomaly confidence of the target region is increased accordingly. By introducing neighborhood difference analysis, the misjudgment of isolated noise regions can be effectively suppressed, while highlighting real abnormal regions that are inconsistent with the performance of the surrounding regions.

[0027] The process for identifying suspicious areas in S4 that do not conform to normal processing conditions is as follows: By combining the grayscale consistency of the same area under multi-position imaging conditions, abnormal responses that only appear in a single viewpoint are suppressed. For images of the same surface area acquired at different imaging positions and observation angles, the corresponding grayscale statistical features of the area are extracted and their changes under different viewpoints are compared and analyzed. When a certain area shows abnormal grayscale response only under a single imaging viewpoint, while the grayscale features remain consistent with the background in other viewpoints, the abnormal response is judged as an unstable anomaly caused by local reflection or viewpoint-related imaging factors, and the anomaly confidence of the area is suppressed, thereby reducing the impact of single-viewpoint misjudgment on the detection results. The morphological continuity and directional consistency of abnormal regions are judged, and random noise responses that do not have the characteristics of machining traces are eliminated. The spatial morphological structure of abnormal regions is analyzed, and the continuity of their boundaries, the stability of their shapes, and their connection with the surrounding texture structures are evaluated. Combined with the main texture direction model, it is determined whether the texture direction in the abnormal region is consistent with or reasonably offset from the turning direction formed during the brake disc machining process. For abnormal regions with scattered shapes, discontinuous boundaries, or directional characteristics that obviously do not conform to the machining trace rules, they are judged as random noise responses and eliminated, thereby avoiding mistaking non-structural disturbances for machining defects. Regions exhibiting stable anomalous characteristics under multiple observation conditions are retained as suspicious regions. When a region exhibits stable anomalous grayscale characteristics, texture structure changes, or directional shift characteristics at different imaging positions, different observation angles, or different sampling times, the region is identified as a stable anomalous region that does not conform to the normal processing state. Stable anomalous regions are output as suspicious regions and used for distribution analysis and quality grading determination in conjunction with the brake disc structure reference position.

[0028] S5: Combining the reference position of the brake disc structure, comprehensively determine the distribution location and influence range of the suspicious area on the working surface, and output the test results for the grading of the brake disc surface quality.

[0029] The process of outputting the test results for brake disc surface quality grading in S5 is as follows: Based on the brake disc center hole, friction radius range, and mounting reference surface, the location of suspicious areas is structurally located. After extracting the suspicious areas, the geometric center of the brake disc is determined as the unified coordinate origin by fitting the center of the brake disc center hole contour. Combined with the equipment calibration parameters, the mounting reference surface direction of the brake disc is identified, and a disc surface reference coordinate system with radial direction and circumferential angle as coordinate axes is established. Under this coordinate system, the pixel coordinates of each suspicious area are mapped to the corresponding radial position interval and circumferential angle range, and its spatial relationship with the friction working area, non-working area, and edge transition area is marked, thereby realizing a structured description of the location of the suspicious area. The distribution range of suspicious areas in the radial and circumferential directions and their potential impact on the brake contact surface are analyzed. Based on the structured positioning results, the coverage ratio of each suspicious area in the radial direction is statistically analyzed to determine whether it is located within the main braking friction radius. The continuous angular length and distribution pattern of suspicious areas in the circumferential direction are analyzed to distinguish between local isolated anomalies and continuous anomalies that expand in the circumferential direction. Combined with the contact pressure distribution characteristics of the brake disc during actual braking, suspicious areas located in high contact frequency areas and with large areas or continuous distribution characteristics are given higher impact weights, while anomalous areas located at the edge or non-main contact areas with small areas and scattered distribution are given lower impact weights, thus forming a comprehensive evaluation result that reflects the degree of potential impact on braking performance. Based on the number, distribution concentration, and impact range of suspicious areas, the surface quality of the brake disc is graded and assessed, and corresponding test results are output. Based on the analysis results, the total number, radial concentration, circumferential continuity, and impact weight of suspicious areas on the brake disc surface are comprehensively statistically analyzed to form a surface quality evaluation index set. This index set is matched with pre-set quality grading rules. For example, brake discs with no or only low-impact, scattered anomalies are judged as qualified, those with localized moderate-impact anomalies are judged as re-inspectable, and those with high-impact, continuously distributed anomalies are judged as unqualified. According to the quality grade, the corresponding test conclusions and suspicious area location information are output, providing clear and interpretable test results for production line quality judgment, rework decisions, or manual review.

[0030] Example 2, as Figure 2 As shown, an intelligent automotive brake disc automatic vision inspection system includes: Multi-tasking imaging module: During the process of the brake disc entering the inspection station, it completes time-division imaging acquisition of the friction working surface, verifies the coverage relationship of each imaging field of view, and constructs an image group; Texture separation module: Analyzes the regular turning texture features on the surface of the brake disc, performs background processing on texture components consistent with the machining process, and extracts residual information with deviations; Anomaly comparison module: Combining the distribution characteristics of residual information at different spatial scales, it performs hierarchical comparison of brightness, texture and orientation changes to form a set of regional features. Region filtering module: Introduces neighborhood association constraints to eliminate unstructured fluctuations caused by reflection or imaging noise and identify suspicious regions that do not conform to the normal processing state; Quality assessment module: comprehensively assesses the distribution and impact range of suspicious areas, and outputs the test results of brake disc surface quality grading.

[0031] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0032] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent automatic visual inspection method for automotive brake discs, characterized in that, Includes the following steps: When the brake disc enters the inspection station with the conveying mechanism, the multi-station imaging device is used to perform time-division sampling imaging of the friction working surface of the brake disc, verify the coverage relationship of the imaging area, and construct an image group that reflects the gray-scale change characteristics of the same surface area under different observation conditions. Based on the regular turning texture formed during the brake disc machining process, the main texture direction, periodic distribution and continuity features in the image group are analyzed and separated. Texture components consistent with the machining process are backgrounded, and residual information that deviates locally from the texture distribution is retained. By combining the distribution characteristics of residual information at different spatial scales and the differences in gray-scale perturbation amplitude, the abnormal signs are analyzed in layers, and the phenomena of brightness abrupt change, texture interruption and direction shift are compared in each layer to form a set of regional features reflecting the weak quality fluctuations on the brake disc surface. By introducing neighborhood association constraints into the regional feature set, and analyzing the differences between adjacent regions in terms of texture extension trend and grayscale stability, non-structural fluctuations caused by local reflection or imaging noise are eliminated, and suspicious regions that do not conform to the normal processing state are identified. Based on the reference position of the brake disc structure, the distribution location and influence range of the suspicious area on the working surface are comprehensively judged, and the test results for grading the surface quality of the brake disc are output.

2. The intelligent automatic visual inspection method for automotive brake discs according to claim 1, characterized in that, The process of constructing an image set that reflects the grayscale variation characteristics of the same surface region under different observation conditions is as follows: As the brake disc rotates at a constant speed with the conveying mechanism and enters the detection station, multiple imaging stations are triggered to simultaneously acquire images according to a preset angle interval. Based on the geometric center and outer edge contour of the brake disc, spatial registration is performed on the images acquired at each workstation to determine the corresponding surface coverage area under different viewpoints. Surface regions with overlapping coverage are indexed and identified. Gray-scale information acquired under different imaging conditions is merged according to time sequence and viewing angle parameters to construct an image group that reflects the gray-scale variation characteristics of the same surface region under different observation conditions.

3. The intelligent automatic visual inspection method for automotive brake discs according to claim 2, characterized in that, The process of analytically separating the main texture direction, periodic distribution, and continuous features in the image group is as follows: Perform directional gradient statistics and spectral decomposition on the image group to extract directional feature components that reflect the main direction of the turning texture; Based on the main direction feature components, the periodic distribution characteristics of the texture in the radial and circumferential directions are calculated to establish a periodic description model of regular textures. By analyzing the continuity and breakage of texture lines in space, we can distinguish between continuous texture structures that conform to the processing technology rules and discontinuous texture structures that have abnormal disturbances.

4. The intelligent automatic visual inspection method for automotive brake discs according to claim 3, characterized in that, The process of preserving residual information that deviates locally from the texture distribution is as follows: Based on the direction and periodicity model of regular texture, the consistent texture region in the image is fitted and reconstructed to generate a textured background image; The original image and the textured background image are differentially processed to eliminate the interference of irregular machining texture on anomaly detection; The difference results are normalized to obtain residual information that highlights local deviation characteristics.

5. The intelligent automatic visual inspection method for automotive brake discs according to claim 4, characterized in that, The process of stratified analysis of abnormal signs is as follows: The residual information map is segmented and statistically analyzed at multiple spatial scales, and the gray-level perturbation intensity and distribution density at each scale are calculated respectively. Based on the different disturbance amplitudes and spatial coverage, the abnormal signs are divided into a weak disturbance layer, a local anomaly layer, and a significant anomaly layer, and corresponding response weights are assigned to the abnormal signs at different levels.

6. The intelligent automatic visual inspection method for automotive brake discs according to claim 5, characterized in that, The process of forming a set of regional features reflecting the slight mass fluctuations on the brake disc surface is as follows: In each anomaly level, regional features are extracted, including brightness change gradient, texture connectivity change, and texture direction offset. Regional features are jointly described within the same spatial region to form a regional feature vector with multiple features superimposed. Consistency verification is performed on the regional feature vectors to form a set of regional features that reflect the slight quality fluctuations on the brake disc surface.

7. The intelligent automatic visual inspection method for automotive brake discs according to claim 6, characterized in that, The process of analyzing the differences in texture extension trends and grayscale stability between adjacent regions is as follows: Using each region in the set of regional features as a node, construct a regional adjacency structure based on spatial adjacency relationships; Compare the texture extension trend and grayscale stability between adjacent regions. When the difference in statistical characteristics between the target region and the adjacent region in at least one of the texture extension trend or grayscale stability indicators exceeds a preset threshold, the anomaly confidence of the target region is increased.

8. The intelligent automatic visual inspection method for automotive brake discs according to claim 7, characterized in that, The process for identifying suspicious areas that do not conform to normal processing conditions is as follows: By combining the grayscale consistency of the same area under multi-position imaging conditions, abnormal responses that only appear in a single viewpoint are suppressed. The morphological continuity and directional consistency of abnormal areas are judged, and random noise responses that do not have processing trace characteristics are eliminated; Regions that exhibit stable anomalous characteristics under multiple observation conditions are retained as suspicious regions.

9. The intelligent automatic visual inspection method for automotive brake discs according to claim 8, characterized in that, The process for outputting test results used for grading the surface quality of brake discs is as follows: Based on the center hole of the brake disc, the range of the friction radius, and the mounting reference surface, the location of the suspicious area is structurally located; Analyze the distribution range of the suspicious areas in the radial and circumferential directions and the degree of potential impact on the brake contact surface; Based on the number, distribution concentration, and impact range of suspicious areas, the surface quality of the brake disc is graded and assessed, and the corresponding test results are output.

10. An intelligent automatic visual inspection system for automotive brake discs, applied to the method described in any one of claims 1-9, characterized in that, include: Multi-tasking imaging module: During the process of the brake disc entering the inspection station, it completes time-division imaging acquisition of the friction working surface, verifies the coverage relationship of each imaging field of view, and constructs an image group; Texture separation module: Analyzes the regular turning texture features on the surface of the brake disc, performs background processing on texture components consistent with the machining process, and extracts residual information with deviations; Anomaly comparison module: Combining the distribution characteristics of residual information at different spatial scales, it performs hierarchical comparison of brightness, texture and orientation changes to form a set of regional features. Region filtering module: Introduces neighborhood association constraints to eliminate unstructured fluctuations caused by reflection or imaging noise and identify suspicious regions that do not conform to the normal processing state; Quality assessment module: comprehensively assesses the distribution and impact range of suspicious areas, and outputs the test results of brake disc surface quality grading.

Images