Automobile brake assembly abnormity identification method and system based on image identification

Abstract

The invention provides an automobile brake assembly abnormity identification method and system based on image identification, and the method comprises the steps: firstly collecting an original image of a hub bearing nut entering an assembly station, recording batch information, extracting visual fingerprint features through a deep learning model, and comparing the visual fingerprint features with pre-stored defect features to complete risk assessment. Then images before and after anchoring are collected at the same position before and after tightening and anchoring operation is executed on the hub bearing nut, a difference image is generated through image registration and difference operation, microscopic physical change characteristics caused by anchoring operation are analyzed through a machine learning model, and a quality judgment result is output. And the system synthesizes an early risk assessment result and a quality judgment result to identify whether the assembly state is abnormal or not. And if an exception is found, packaging all the information into an exceptional data packet, calculating an encrypted hash value, and storing the encrypted hash value into the distributed block chain. According to the method, abnormity recognition and tracing in the brake assembling process are achieved, and the quality control level of automobile safety critical components is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition, and in particular relates to a method and system for identifying automobile brake assembly anomalies based on image recognition. Background Art

[0002] In the automotive industry, the assembly quality of the brake system is directly related to vehicle safety, and the correct installation of the wheel hub bearing nut is particularly critical. A significant challenge with existing technologies is the difficulty in accurately and in real time identifying minor anomalies during brake assembly, particularly those subtle defects that are imperceptible to the naked eye but could pose serious safety risks. Traditional quality inspection methods typically rely on manual visual inspection or simple mechanical measurements, which struggle to capture subtle variations and potential risks during the assembly process.

[0003] Especially on high-speed production lines, where inspection time is limited, manual inspection is susceptible to factors such as fatigue and subjective judgment, resulting in a high rate of missed inspections. Conventional automated inspection equipment can often only detect obvious geometric deviations or surface defects, but cannot effectively identify subtle deformations and abnormal stress distribution that occur during the assembly process. Although these hidden problems may not be apparent during the assembly stage, they can gradually worsen during vehicle use, ultimately leading to brake system failure. Summary of the Invention

[0004] The present invention provides a method and system for identifying automobile brake assembly abnormalities based on image recognition to solve the above technical problems.

[0005] In a first aspect, the present invention provides a method for identifying anomalies in automobile brake assembly based on image recognition, the method comprising the following steps:

[0006] The beneficial effects of the present invention are:

[0007] By integrating deep learning with image differential analysis technology, this invention enables precise identification and full-process traceability of minor anomalies during the automotive brake assembly process, significantly improving the quality control of critical automotive safety components. Compared to traditional methods that rely on manual inspection or simple mechanical measurement, this invention can automatically capture microscopic physical changes imperceptible to the naked eye, improving detection accuracy to submillimeter levels and significantly reducing missed detection and false positive rates. In particular, through precise registration and differential analysis of pre- and post-anchoring images, the present invention can visually reveal subtle deformations and stress distribution anomalies generated during the tightening process. These hidden defects, often overlooked in traditional inspections, may be the source of future safety hazards. The present invention's early risk assessment mechanism, through visual fingerprint feature extraction, can identify potential issues before parts enter assembly, shifting quality control forward and avoiding the waste of resources caused by unqualified parts entering subsequent processes. Furthermore, the present invention innovatively incorporates blockchain technology into the quality traceability system, ensuring the authenticity, integrity, and immutability of anomaly data, providing a reliable basis for quality accountability determination and batch recalls. In practical applications, this invention not only improves the automation level and inspection efficiency of the production line, but also forms a self-evolving quality control system through data accumulation and model optimization, which can continuously improve the ability to identify new defects. This comprehensive, high-precision, and traceable quality control solution provides strong technical support for the automotive manufacturing industry, effectively ensuring the reliability of the braking system and vehicle safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 This is a flow chart of a method for identifying abnormalities in automobile brake assembly based on image recognition in one embodiment of the present application.

[0009] Figure 2 This is a schematic diagram of the system structure of an automobile brake assembly abnormality identification system in one embodiment of the present application. DETAILED DESCRIPTION

[0010] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0011] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0012] Figure 1 FIG. 1 is a flow chart of a method for identifying abnormalities in the assembly of automobile brakes based on image recognition in one embodiment. It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps in the above process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but may be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps. Figure 1 As shown, the present invention discloses a method for identifying abnormalities in automobile brake assembly based on image recognition, which specifically includes the following steps:

[0013] S101. Collect the original image of the wheel hub bearing nut entering the assembly station, and identify and record the batch information of the wheel hub bearing nut.

[0014] When the wheel hub bearing nut enters the assembly station, a high-resolution industrial camera captures it from multiple angles under standardized lighting conditions, capturing raw images that capture the nut's surface texture, geometry, and material characteristics. Optical character recognition (OCR) or a QR code scanner simultaneously reads the batch identification on the nut's surface, associating information such as the batch number, production date, and supplier code with the image data. For example, when the batch number "HN2024A001" is detected, the camera captures the front and side views of the nut at a resolution of 2048×2048 pixels, ensuring image clarity sufficient for subsequent analysis.

[0015] S102. Use a deep learning model to extract the visual fingerprint features of the hub bearing nut from the original image, and compare the visual fingerprint features with the pre-stored visual defect features to complete an early risk assessment of the hub bearing nut.

[0016] The deep learning model first preprocesses the original image, including Gaussian filtering for denoising, histogram equalization, and illumination normalization. It then uses the Canny edge detection algorithm to accurately segment the nut's region of interest. The preprocessed image is fed into a pretrained convolutional autoencoder network, which consists of a 5-layer encoder and 5-layer decoder. A nonlinear transformation is performed using the ReLU activation function, and a 512-dimensional latent space feature vector is extracted from the intermediate bottleneck layer as the initial visual fingerprint. Principal component analysis is then applied to reduce the feature dimension to 128, retaining 95% of the variance information. The optimized visual fingerprint features are then compared with the defect features stored in the historical risk database using a weighted cosine similarity calculation. When the similarity exceeds a threshold of 0.85, it is marked as a preliminary abnormal risk; otherwise, it is marked as normal.

[0017] S103. Before the hub bearing nut is tightened and anchored, a pre-anchoring image is captured; after the same hub bearing nut is tightened and anchored, a post-anchoring image is captured at the same shooting position.

[0018] Among them, at the critical time point when the nut is tightened and anchored, a high-precision camera system in a fixed position is used to capture images. Before anchoring, the camera captures a reference image when the nut is not in contact with the tool, recording the initial state of the nut, including surface gloss, texture distribution, and geometric outline. After the anchoring operation is completed, the camera captures the post-anchoring image at exactly the same shooting position, angle, and lighting conditions to ensure that the spatial coordinate deviation of the two shots is controlled within 0.1 mm. The high consistency of the camera position is guaranteed by a precise mechanical positioning device and a laser guidance system. For example, when the torque wrench completes a tightening operation of 35 N·m, the camera immediately triggers the shooting to capture the microscopic changes on the nut surface caused by the force.

[0019] S104. Perform image registration and differential operations on the pre-anchoring image and the post-anchoring image to generate a differential image, use a machine learning model to analyze the microscopic physical change characteristics caused by the anchoring operation in the differential image, and output a quality judgment result of the anchoring operation.

[0020] Accurately registering the pre- and post-anchoring images is a key step in detecting microscopic changes. SURF (Speeded Robust Features) and ORB algorithms are first used to detect local feature points in key structural regions of the two images. Correspondence between these feature points is established using the FLANN matcher. RANSAC is then used to remove mismatched points, retaining accurate correspondences with a confidence level above 0.9. A homography transformation matrix is ​​calculated based on these feature points, aligning the post-anchoring image with the pre-anchoring image at the subpixel level. After alignment, a normalized grayscale difference operation is performed to calculate the grayscale value difference at each pixel position, generating an initial difference image containing both positive and negative change information. A Gaussian filter is then applied to smooth the noise, and an adaptive threshold is used to binarize the image to highlight areas of significant change. The resulting difference image clearly displays microscopic physical features such as surface deformation, tool contact marks, and material plastic deformation caused by the anchoring operation.

[0021] The difference image is fed into an anchor anomaly recognition model built on the DenseNet architecture. This model is trained using supervised learning on a dataset containing 5,000 examples of different anchoring conditions. The model's deep convolutional layers extract multi-level feature maps representing the tool contact area morphology, the degree of minor material plastic deformation, surface gloss, and texture variations. Each feature map contains 64 to 512 distinct feature channels. Fully connected layers and a Softmax classification layer output discrete categories of anchor quality: "acceptable," "minor anomaly," and "serious anomaly," along with a confidence score. A parallel attention mechanism generates a heatmap to precisely locate the coordinates of anomaly features in the difference image. If the judgment result is an unacceptable condition, the model decodes the feature map and outputs a specific anomaly description, such as "uneven thread contact" or "overtightening causing material deformation." All outputs are integrated into a comprehensive anchoring operation quality assessment report.

[0022] S105. Integrate the evaluation results of the early risk assessment and the quality judgment results to identify whether there is any assembly abnormality in the assembly state of the hub bearing nut.

[0023] The abnormality decision engine, built based on weighted fuzzy logic, receives early risk assessment results and anchor quality judgment results as input variables. Fuzzy membership functions are preset for different input combinations. For example, if the risk assessment is "high risk" and the quality judgment is "severely abnormal," the membership function value is set to 0.9. A rule base containing 25 if-then rules is established, such as "IF Early Risk is Medium AND Anchor Quality is Slightly Abnormal THEN Assembly Status is Suspicious." The decision engine converts the input values ​​into fuzzy sets through a fuzzification process and performs fuzzy reasoning using the Mamdani inference method. Finally, the center of gravity method is used to defuzzify the input values ​​to obtain a comprehensive assessment score ranging from 0 to 100. The score is compared with preset thresholds: scores below 30 are considered normal, scores between 30 and 70 are considered suspicious and require re-inspection, and scores above 70 are considered abnormal and require action.

[0024] S106. If there is an assembly anomaly in the assembly state of the hub bearing nut, all the collected image information, batch information, early risk assessment results, and quality judgment results of the hub bearing nut are packaged into an abnormal data packet, the encrypted hash value of the abnormal data packet is calculated, and the abnormal data packet and the encrypted hash value are stored in a preset distributed blockchain.

[0025] When an assembly anomaly is detected, all relevant data is first timestamped and synchronized to ensure temporal consistency across image information, batch information, assessment results, and judgment results. The synchronized key anomaly data is encoded in the Apache Avro serialization format, generating a compact binary anomaly data packet typically between 2 and 5MB in size. The data packet is encrypted using the Keccak-256 hash function from the SHA-3 algorithm family, generating a 64-bit hexadecimal fixed-length hash value, such as "a1b2c3d4e5f6...", which serves as the packet's unique content fingerprint. A smart contract encapsulates the encrypted hash value and the anomaly data packet into a blockchain transaction, which is verified using the elliptic curve digital signature algorithm. The final data is stored on a pre-defined consortium or private blockchain. Each block contains a timestamp, the previous block's hash value, and transaction data, ensuring data immutability and complete traceability, providing a reliable chain of evidence for subsequent investigations of quality issues.

[0026] In one embodiment, extracting the visual fingerprint features of the hub bearing nut from the original image using a deep learning model includes the following steps:

[0027] Perform preprocessing operations on the original image, including image denoising, illumination normalization, and accurate segmentation of the region of interest based on contour detection;

[0028] The preprocessed original image is input into the pre-trained convolutional autoencoder network, and the encoder part of the convolutional autoencoder network performs nonlinear compression transformation on the original image;

[0029] After performing nonlinear compression transformation on the original image, the latent space feature vector of the original image is extracted from the middle bottleneck layer of the convolutional autoencoder network as the initial visual fingerprint of the hub bearing nut;

[0030] The principal component analysis method is applied to the initial visual fingerprint for dimensional optimization and feature selection to enhance the initial visual fingerprint's ability to express defect-sensitive features and form the visual fingerprint characteristics of the hub bearing nut.

[0031] In this embodiment, the preprocessing of the original image is a key basic step to ensure the accuracy of subsequent analysis. First, the image is denoised and a Gaussian filter is used to eliminate the random noise generated by the camera sensor. The filter kernel size is set to 5×5, the standard deviation σ=1.2, and the filtering formula is: . Then, an illumination normalization operation is performed, and the image brightness distribution is adjusted through the histogram equalization algorithm so that the pixel values ​​are evenly distributed in the range of 0-255, eliminating the impact of different lighting conditions on image quality. Finally, the region of interest is segmented, and the Canny edge detection algorithm is used to identify the nut outline. The dual threshold parameters are set to 50 and 150, and the interior of the outline is filled through morphological operations to accurately extract the nut area. For example, for a 1024×1024 pixel original image, a clear 512×512 pixel nut area image is obtained after preprocessing, the background noise is reduced by 80%, and the brightness standard deviation is controlled within 15, providing high-quality input data for subsequent feature extraction.

[0032] The preprocessed image is fed into a pretrained convolutional autoencoder network for deep feature learning. This network employs a symmetric encoder-decoder architecture. The encoder consists of five convolutional layers, each using a 3×3 convolution kernel with a stride of 2 and a Reluctant Unit (ReLU) activation function. The first layer outputs 64 feature maps, followed by 128, 256, 512, and 1024 feature maps. Each convolutional layer is followed by a batch normalization layer and a max pooling layer, progressively reducing the spatial resolution while increasing the feature depth. Through this nonlinear compression transformation, the 512×512 input image is progressively compressed into a high-dimensional feature representation of 16×16×1024. The network, pretrained on a dataset of 10,000 nut samples, learns an abstract representation of the nut's surface texture, geometry, and material properties, effectively capturing key visual features that influence assembly quality.

[0033] After the convolutional autoencoder completes the nonlinear compression transformation, a latent space feature vector is extracted from the network's intermediate bottleneck layer as the nut's initial visual fingerprint. The bottleneck layer, located at the last layer of the encoder, contains 16×16×1024 neurons. A global average pooling operation is used to compress the spatial dimensions to produce a 1024-dimensional feature vector. This 1024-dimensional vector contains high-level semantic information about the nut image, with each dimension representing a specific visual pattern or feature combination. For example, some dimensions may correspond to the regularity of the thread, while others may reflect surface roughness or material gloss. t-SNE visualization analysis revealed that nuts of similar quality clustered in the latent space, while nuts with different defect types were clearly separated in space, demonstrating the effectiveness and discriminative power of the feature vector.

[0034] The principal component analysis method is applied to the 1024-dimensional initial visual fingerprint for dimensionality optimization and feature selection to enhance the ability to express defect-sensitive features. First, the covariance matrix of the eigenvector is calculated. , then solve the eigenvalues ​​and eigenvectors, and sort them in descending order of eigenvalue size. Select the first k principal components whose cumulative variance contribution rate reaches 95%, usually the k value is between 128-256. The transformation formula is ,in The principal component matrix is: Through PCA transformation, the original 1024-dimensional features are compressed to 128 dimensions while retaining the most important change information.

[0035] In one embodiment, comparing visual fingerprint features with pre-stored visual defect features to complete early risk assessment of the hub bearing nut includes the following steps:

[0036] Accessing a historical risk database, where visual defect features corresponding to nut samples of various defect types are pre-stored in the historical risk database;

[0037] The weighted cosine similarity is used to calculate the matching score between the visual fingerprint features and the visual defect features;

[0038] Combine the matching scores and the multi-level matching score thresholds set for different defect types to determine whether the visual fingerprint feature has a statistically significant association with any defect type;

[0039] If there is a significant statistical correlation, the corresponding wheel hub bearing nut will be marked with a preliminary abnormal early risk assessment result based on the associated defect type and the confidence interval of the matching score, and the visual fingerprint feature will be updated to the historical risk database;

[0040] If there is no significant statistical association or the matching score is lower than the preset warning threshold, it is an early risk assessment result that the corresponding wheel hub bearing nut mark is initially normal.

[0041] In this embodiment, the historical risk database adopts a distributed storage architecture and pre-stores the visual defect features corresponding to nut samples of 15 common defect types. The database is stored by defect type, including thread damage, surface cracks, uneven material, dimensional deviation, corrosion spots, etc. Each defect type contains a 128-dimensional feature vector of 500-2000 samples. The database adopts a B+ tree index structure, supports fast retrieval and range query, and the average query time is controlled within 10 milliseconds. The access process obtains relevant feature data through the SQL query statement "SELECT feature_vector, defect_type, confidence_level FROM defect_database WHERE defect_type IN ('thread_damage', 'surface_crack', 'material_uneven')". The database also maintains statistical information for each defect type, including the mean, standard deviation and distribution parameters of the feature vector. Through an efficient database access mechanism, a complete reference benchmark data set is provided for subsequent similarity calculations.

[0042] The weighted cosine similarity algorithm is used to calculate the matching degree between the visual fingerprint features of the current nut and the various defect features in the database. The weighted cosine similarity formula is:

[0043]

[0044] in is the current nut feature vector, is the defect sample feature vector, is the weight coefficient for the i-th dimension feature. The weight coefficient is determined based on the feature dimension's ability to discriminate between different defect types. Calculated using the information gain algorithm, the weight for important feature dimensions can reach 2.5, while the weight for minor dimensions is 0.3. During the calculation process, the similarity is calculated for all samples of each defect type, and the average is taken as the overall matching score for that defect type.

[0045] The statistical correlation between visual fingerprint features and each defect type is determined based on matching scores and a multi-level threshold system. Three matching score thresholds are set for different defect types: high-risk threshold T_h=0.85, medium-risk threshold T_m=0.70, and low-risk threshold T_l=0.55. Statistical correlation is determined through hypothesis testing, with the null hypothesis H_0 being "no correlation between the current sample and the defect type" and the alternative hypothesis H_1 being "a significant correlation exists." The t statistic is calculated. , where μ_0 is the historical average similarity for that defect type, σ_0 is the standard deviation, and n is the number of samples. When the t-value is greater than the critical value t_0.05 = 1.96, the null hypothesis is rejected, and a statistically significant association is considered. For example, the matching score between a nut and thread damage is 0.82, exceeding the high-risk threshold of 0.85, and the t-value is 2.34 > 1.96, indicating a significant association. The confidence interval is also considered. When the 95% confidence interval of the matching score is [0.78, 0.86], the reliability of the association is further confirmed.

[0046] When a statistically significant association is detected, detailed early risk assessment results are marked for the nut based on the associated defect type and the confidence interval of the matching score. The risk level is divided into four levels: extremely high risk (matching score>0.90), high risk (0.80-0.90), medium risk (0.65-0.80), and low risk (0.55-0.65). The assessment results include defect type, risk level, confidence level, and recommended measures. For example, a thread damage association with a matching score of 0.87 is marked as "Defect type: thread damage, risk level: high risk, confidence level: 92%, recommendation: prioritize checking thread integrity". At the same time, the visual fingerprint features of the current nut are updated as new samples to the historical risk database, and the statistical parameters of the defect type are updated using incremental learning. The update formula is , .

[0047] When the matching scores of the visual fingerprint features with all defect types are lower than the preset warning threshold of 0.55, or there is no significant statistical correlation, the corresponding nut is marked with a preliminary normal early risk assessment result. The normal state judgment adopts a multiple verification mechanism. First, it is checked whether the matching scores of all 15 defect types are lower than their respective low-risk thresholds, and then it is verified whether the highest matching score is lower than the global warning threshold of 0.55. In the statistical correlation test, the t-statistics of all defect types are less than the critical value of 1.96, and the p-value is greater than 0.05, and the null hypothesis is accepted. The normal state mark contains detailed information: "Risk level: normal, highest matching score: 0.42 (surface roughness), confidence: 96%, recommendation: continue assembly according to the standard process". To ensure the reliability of the judgment, the Mahalanobis distance is also calculated To measure the distance between the current sample and the center of the normal sample distribution, the normal state is further confirmed when D_M<3.0.

[0048] In one embodiment, performing image registration and difference calculation on the pre-anchoring image and the post-anchoring image to generate a difference image includes the following steps:

[0049] The accelerated robust features and ORB algorithm are used to detect local feature points in the key structure areas of the image before anchoring and the image after anchoring respectively;

[0050] FLANN matching is performed based on the descriptor of local feature points to establish the initial correspondence between the local feature points in the image before and after anchoring. The wrong matching point pairs are eliminated through the random sampling consistency algorithm to obtain the accurately corresponding feature points after screening.

[0051] Based on the precise correspondence of feature points, the homography transformation model is used to align the anchored image with the pre-anchored image at the sub-pixel level.

[0052] Perform a normalized grayscale value subtraction operation on the same pixel position of the aligned pre-anchor image and post-anchor image to obtain an initial difference image containing positive and negative difference information;

[0053] The initial difference image is subjected to Gaussian filtering to smooth the noise, and the adaptive threshold binarization method is applied to highlight the microscopic physical change characteristics caused by the anchoring operation to obtain the difference image.

[0054] In this implementation, SURF (Speeded Robust Features) and ORB (Oriented FAST and Rotated BRIEF) algorithms are applied to the key structural areas of the images before and after anchoring to detect local feature points, providing basic anchor points for subsequent image registration. The SURF algorithm first constructs a scale space and detects interest points through the determinant value of the Hessian matrix. The calculation formula is: ,when The ORB algorithm combines FAST corner detection and BRIEF descriptor to detect corner points through Harris corner response function. Stable feature points are screened, where M is the structure tensor matrix and k is typically set to 0.04. In practical applications, the SURF algorithm detects approximately 150 feature points in the threaded area of ​​the nut, and the ORB algorithm detects approximately 200 feature points in the surface texture area. The combination of these two algorithms ensures stable detection of a sufficient number of feature points under varying lighting and angles, laying a solid foundation for accurate registration.

[0055] Based on the 128-dimensional SURF descriptor and 32-dimensional ORB descriptor of the detected local feature points, the FLANN (Fast Nearest Neighbor Search Library) matching algorithm is used to establish the initial correspondence between the feature points in the image before and after anchoring. FLANN uses KD tree or LSH hash table for fast nearest neighbor search. The matching criterion is the minimum Euclidean distance. The distance calculation formula is: To improve matching quality, a Lowe ratio test is used. A match is considered valid when the ratio of the nearest neighbor distance to the next nearest neighbor distance is less than 0.7. Initial matching typically produces 300-500 matching point pairs, but this includes a large number of false matches. False matches are then removed using the RANSAC (Random Sample Consensus) algorithm. During this iterative process, four matching points are randomly selected to calculate the homography matrix. The number of inliers is counted, and points with a reprojection error of less than 3 pixels are considered inliers.

[0056] Based on the selected precise corresponding feature points, the homography transformation model is used to achieve sub-pixel precise alignment of the image after anchoring and the image before anchoring. The homography matrix H is a 3×3 homogeneous transformation matrix, and the transformation relationship is:

[0057] ,in .

[0058] The eight unknown parameters are solved by the least squares method, and the objective function is In order to achieve sub-pixel accuracy, bilinear interpolation is used to resample pixel values. The interpolation formula is:

[0059]

[0060] where α and β are decimals. The transformed registration accuracy typically reaches within 0.1 pixel, effectively eliminating small displacements, rotations, and scale changes between the two shots, creating conditions for accurately detecting the real physical changes caused by the anchoring operation.

[0061] After precise alignment, the normalized grayscale value subtraction operation is performed on the images before and after anchoring to generate an initial differential image containing positive and negative difference information. First, the grayscale of the two images is normalized to ensure that the pixel value range is consistent. The normalization formula is: . Then perform the difference operation at the same pixel location , and get the difference value in the range of [-255, 255]. For the convenience of display and subsequent processing, the difference value is mapped to the range of [0, 255]. The mapping formula is A positive difference value (bright area) indicates that the area has become brighter after anchoring, which usually corresponds to material compression or tool contact marks; a negative difference value (dark area) indicates a darkening, which may reflect material deformation or shadow changes.

[0062] The initial difference image is smoothed by Gaussian filtering to eliminate noise interference, and then adaptive threshold binarization is applied to highlight the significant microscopic physical change features. Gaussian filtering uses a 5×5 convolution kernel with a standard deviation of σ=1.5, and the filter weights are calculated according to a two-dimensional Gaussian distribution. , effectively suppressing high-frequency noise while maintaining edge information. Adaptive threshold binarization uses local statistical characteristics to determine the threshold, and calculates the mean and standard deviation of each pixel in its neighborhood. The threshold formula is , where μ is the local mean, σ is the local standard deviation, and k is a tuning parameter, usually set to 0.2. When the pixel value is greater than the local threshold, it is set to 255 (white), otherwise it is set to 0 (black).

[0063] In one embodiment, using a machine learning model to analyze microscopic physical change characteristics caused by the anchoring operation in the differential image and outputting a quality judgment result of the anchoring operation includes the following steps:

[0064] The difference image is input into an anchoring anomaly recognition model built on a densely connected network. The anchoring anomaly recognition model is pre-trained through supervised learning using difference image samples containing different anchoring states.

[0065] The deep convolutional layer of the anchoring anomaly recognition model is used to extract hierarchical feature maps from the differential image, which represent the morphology of the tool contact area, the degree of material micro-plastic deformation, and the surface gloss and texture changes.

[0066] Based on the hierarchical feature map, the fully connected layer and softmax classification layer of the anchoring anomaly recognition model are used to output discrete category judgments of the anchoring quality. At the same time, through the parallel regression branch or attention mechanism in the anchoring anomaly recognition model, the confidence score of the anchoring quality judgment and the heat map indicating the location of abnormal micro features in the difference image are output;

[0067] If the discrete category judgment result is a non-qualified state, a text description of the microscopic physical change characteristics associated with the non-qualified state is decoded from the hierarchical feature map and output;

[0068] The text description and all output results of the anchoring anomaly recognition model are integrated into the quality judgment result of the anchoring operation.

[0069] In this embodiment, the differential image input is based on the anchor anomaly recognition model built on the DenseNet architecture, which uses a densely connected block structure to achieve feature reuse and gradient flow optimization. The model contains 4 dense blocks, and each dense block has direct connections between layers. The connection method is ,in represents the concatenation of the outputs of all previous layers, The composite function for layer l includes batch normalization, ReLU activation, and a 3×3 convolution operation. The model was pre-trained for supervised learning using a dataset containing 8,000 differential image samples of different anchoring states, covering 12 categories: normal anchoring, overtightening, undertightening, and eccentric contact. The cross-entropy loss function was used during training.

[0070] The deep convolutional layer of the anchoring anomaly recognition model uses a multi-scale feature extraction strategy to extract hierarchical feature maps from the differential image layer by layer, representing the morphology of the tool contact area, material plastic deformation, surface gloss, and texture changes. The first convolution layer extracts 64 7×7 basic feature maps to capture edge and basic texture information; the second layer outputs 128 5×5 feature maps to identify the geometric shape of tool contact; the third layer generates 256 3×3 feature maps to detect the subtle deformation mode of the material; the fourth layer generates 512 feature maps to analyze the variation of surface gloss. The calculation formula for each layer of feature map is: ,in is the ReLU activation function, is the convolution kernel weight, Represents a convolution operation. Through dense connections, deep feature maps are able to access all shallow-layer information, forming a rich feature representation. For example, when detecting overtightened threads, shallow feature maps reveal the sharpening of contact edges, while deep feature maps capture changes in texture density caused by material compression.

[0071] Based on the extracted hierarchical feature map, the fully connected layer and the Softmax classification layer output the discrete category judgment of the anchoring quality, and the parallel regression branch and attention mechanism are used to generate the confidence score and the abnormal location heat map. The fully connected layer maps the 960-dimensional feature vector to the prediction score of 12 categories, and the Softmax function calculates the probability of each category. ,in is the original score of the i-th category. The final category is the category with the highest probability. The parallel regression branch outputs the confidence score through the sigmoid function The regression value is mapped to the interval [0,1]. The attention mechanism uses the Grad-CAM algorithm to generate the heat map, and the calculation formula is ,in is the weight of the k-th feature map, is the corresponding feature map.

[0072] When the discrete category judgment result is a non-qualified state, a detailed text description of the microscopic physical change characteristics associated with the non-qualified state is decoded from the hierarchical feature map and output. The decoding process uses a feature-text mapping network based on the attention mechanism, which is pre-trained on the correspondence between the feature map and the description text. First, the feature map is compressed into a feature vector by global average pooling. , and then input into LSTM decoder to generate text description. The hidden state update formula of LSTM is , the probability distribution of the output vocabulary is For example, for the "overtightening" condition, the decoding output is "obvious indentation appears in the thread contact area, the material undergoes plastic deformation, the surface gloss decreases by 15%, the texture density increases, and it is recommended to check the torque setting value."

[0073] The text description and all outputs of the anchor anomaly recognition model are integrated into a complete anchoring operation quality assessment report, creating a standardized quality assessment document. This integration process organizes information according to a predefined template structure, including six sections: basic information, classification results, confidence assessment, anomaly location, feature description, and action recommendations. Basic information records the inspection time, image resolution, and processing time; classification results display the probability distribution of 12 categories and the final judgment; confidence assessment provides a quantitative indicator of judgment reliability; anomaly location precisely identifies the problem area using heat map coordinates; feature descriptions use natural language to detail the detected physical changes; and action recommendations provide action guidance based on the anomaly type and severity. For example, a complete report format might be "Inspection time: 2024-03-15 14:32:05, Classification result: Overtightening (probability 0.89), Confidence: 89%, Anomaly location: (128,156) 25-pixel radius area, Feature description: Thread root indentation depth of 0.02 mm exceeds the standard, Action recommendation: Adjust torque to the standard value of 30±2 N·m and reassemble."

[0074] In one embodiment, the steps of combining the early risk assessment result and the quality judgment result to identify whether there is an assembly abnormality in the assembly state of the hub bearing nut include the following:

[0075] An exception decision engine is built based on weighted fuzzy logic. The input of the exception decision engine is the early risk assessment results and quality judgment results.

[0076] Preset fuzzy membership functions and IF-THEN rule bases for different combinations of early risk assessment results and quality judgment results;

[0077] Combining the fuzzy membership function and the IF-THEN rule base, and through the fuzzification, fuzzy reasoning, and defuzzification processes of the abnormal decision engine, the comprehensive evaluation score of the hub bearing nut assembly status is calculated;

[0078] The comprehensive evaluation score is compared with the preset multi-level threshold value, and whether there is any assembly abnormality in the assembly state of the hub bearing nut is determined based on the comparison result.

[0079] In this embodiment, an abnormal decision engine is constructed based on weighted fuzzy logic theory, which receives early risk assessment results and quality judgment results as dual input variables for comprehensive decision analysis. The decision engine adopts the Mamdani fuzzy reasoning system architecture, which includes four core components: fuzzification interface, rule base, inference engine and defuzzification interface. The input range of early risk assessment results is [0,1], where 0 represents no risk and 1 represents extremely high risk; the input range of quality judgment results is also [0,1], where 0 represents fully qualified and 1 represents serious abnormality. The engine uses a weighted mechanism to handle the difference in importance of the two inputs, and the weight distribution formula is: The decision engine outputs a comprehensive evaluation score ranging from 0 to 100, with higher scores indicating a greater likelihood of assembly anomaly. By leveraging the uncertainty processing capabilities of fuzzy logic, the engine effectively integrates two different types of evaluation information, overcoming the limitations of traditional binary logic in handling edge cases and providing more flexible and accurate decision support for complex assembly quality assessments.

[0080] Detailed fuzzy membership functions and a complete IF-THEN rule base are preset for different combinations of early risk assessment and quality judgment results. Early risk assessment uses a triangular membership function to define five linguistic variables:

[0081] Very low risk

[0082] Low risk

[0083] Medium risk

[0084] High risk

[0085] Extremely high risk

[0086] Quality judgment results adopt the same five-level classification structure. The rule base contains 25 if-then rules, covering all possible input combinations, such as "IF the early risk is high AND the quality judgment is moderately abnormal THEN the assembly status is severely abnormal." Each rule is assigned a corresponding weight coefficient, with key rules having a weight of up to 1.0 and minor rules having a weight of 0.6.

[0087] Combining the preset fuzzy membership function and rule base, the abnormal decision engine calculates the comprehensive evaluation score of the hub bearing nut assembly status through three stages: fuzzification, fuzzy reasoning, and defuzzification. The fuzzification stage converts the input precise numerical value into a fuzzy set and calculates the membership value of each linguistic variable. The fuzzy reasoning stage uses the minimum-maximum synthesis operation to calculate the output fuzzy set for each activated rule. The formula is ,in is the activation strength of the i-th rule, is the corresponding output membership function. Defuzzification uses the centroid method to calculate the final accurate output value, the formula is ,in is a discrete point in the output universe.

[0088] The calculated comprehensive evaluation score is compared with the preset multi-level threshold system, and the assembly status of the hub bearing nut is accurately judged based on the comparison results. The multi-level threshold system includes four key nodes: normal threshold , Suspicious Threshold , abnormal threshold , severe abnormality threshold The judgment logic is: when It is considered normal and no special treatment is required; when It is judged as suspicious and manual re-inspection is recommended; When it is judged as a slight abnormality, it needs to be reassembled; when When it is judged as a moderate abnormality, replacement of parts is required; when When a fault is detected, it is considered a serious abnormality, requiring shutdown and inspection. Threshold settings are based on ROC curve analysis and cost-benefit evaluation to ensure a false positive rate of less than 3%. For example, if the comprehensive evaluation score is 72.3, exceeding the abnormality threshold of 65, the system will determine a moderate abnormality and automatically trigger the part replacement process.

[0089] In one embodiment, packaging all collected image information, batch information, early risk assessment results, and quality judgment results of the hub bearing nut into an abnormal data packet, calculating an encrypted hash value of the abnormal data packet, and storing the abnormal data packet and the encrypted hash value in a preset distributed blockchain includes the following steps:

[0090] All collected image information, batch information, early risk assessment results, and quality judgment results of wheel hub bearing nuts are time-stamped and synchronized, and integrated into key abnormality data;

[0091] Encode the integrated key exception data in Avro serialization format to form a binary exception data packet;

[0092] Apply a cryptographic hash function based on the SHA-3 algorithm family to operate on the abnormal data packet to obtain a fixed-length cryptographic hash value. The cryptographic hash value is the unique content fingerprint of the abnormal data packet on the blockchain.

[0093] The encrypted hash value and the abnormal data packet are encapsulated and stored in the preset distributed blockchain.

[0094] In this implementation, all collected data related to hub bearing nuts undergoes precise timestamp synchronization to ensure temporal consistency and integrity. Timestamp synchronization utilizes the UTC Coordinated Universal Time standard, achieving millisecond-level accuracy. All image information includes the original image, pre- and post-anchoring images, and differential images. Each image is accompanied by a capture timestamp, camera parameters, and image quality metrics. Batch information includes the nut number, production date, supplier code, and material specifications. Early risk assessment results include risk level, confidence score, and defect type prediction. Quality judgment results include classification results, anomaly descriptions, and handling recommendations. The integration process organizes key anomaly data into a unified format using a predefined JSON data structure. The integrated key anomaly data is efficiently encoded using the Apache Avro serialization format, generating a compact binary anomaly data packet. Avro utilizes a schema evolution mechanism, first defining a data schema containing field names, data types, and constraints. For example, image data is defined as bytes, timestamps as long, and text descriptions as string. The serialization process converts the JSON-formatted anomaly data into binary format, using variable-length encoding to compress integers and strings, typically achieving a compression ratio of 60-70%. The encoding algorithm removes duplicate data, and the LZ4 compression algorithm is applied to image data to further reduce its size. The serialized binary data packets have cross-platform compatibility and efficient read and write performance, supporting fast data transmission and storage operations, and providing an optimized data format for distributed storage on the blockchain.

[0095] The Keccak-256 hash function based on the SHA-3 algorithm family is used to perform encryption operations on the serialized abnormal data packets to generate a unique content fingerprint of fixed length. The SHA-3 algorithm uses a sponge construction, which includes two stages: absorption and extrusion. The state array size is 1600 bits. In the absorption stage, the input data is divided into blocks, each block is 576 bits, and the permutation function is used to generate a unique content fingerprint. Perform 24 rounds of transformation, and the transformation formula is ,in is the current state, The extrusion phase extracts a 256-bit hash value from the final state, outputting it as 64 hexadecimal characters. The hash calculation process exhibits an avalanche effect, where small changes in the input data result in significant differences in the output hash value, ensuring reliable data integrity verification. The resulting hash value serves as a unique identifier for the data packet on the blockchain, enabling fast data retrieval and integrity verification.

[0096] The calculated cryptographic hash value is encapsulated with the abnormal data packet and then stored in a pre-defined distributed blockchain network for tamper-proof data storage. The encapsulation process constructs a blockchain transaction structure consisting of four components: a transaction header, a hash value, a data packet, and a digital signature. The transaction header records the timestamp, version number, and data size; the hash value is stored as a content index in the transaction's key fields; the abnormal data packet is Base64-encoded and stored in the transaction's data payload; and the digital signature is generated using the Elliptic Curve Digital Signature Algorithm (ECDSA). The blockchain network utilizes a consortium chain architecture with five validating nodes and employs the PBFT consensus algorithm to ensure data consistency. The storage process first broadcasts the encapsulated transaction to the network. Each node verifies the transaction's validity and signature correctness. Once consensus is reached, the transaction is packaged into a new block. Each block contains a block header and a list of transactions. The block header includes the previous block's hash, Merkle root, and a timestamp, forming an immutable chain structure. After storage, the abnormal data is assigned a unique blockchain address and transaction ID, enabling rapid retrieval and verification via hash value, providing reliable technical support for quality traceability and auditing.

[0097] In one embodiment, the method comprises the following steps:

[0098] receiving a traceback request from an authorized user and including one or more query identifiers;

[0099] Retrieving matching candidate transaction records from an on-chain index of the distributed blockchain using an application program interface interacting with the distributed blockchain and utilizing a query identifier in the traceability request;

[0100] For each candidate transaction record retrieved, verify the integrity of the candidate transaction record block hash chain and the validity of the transaction signature;

[0101] Extract the encrypted and stored abnormal data packets and the corresponding encrypted hash values ​​from the verified candidate transaction records.

[0102] In this embodiment, the traceability system receives query requests from authorized users, each containing one or more query identifiers used to locate specific exception records. User authentication utilizes a JWT (JSON Web Token)-based authentication mechanism to ensure that only authorized users can access sensitive quality data. Query identifiers support a variety of formats, including nut batch numbers, production time ranges, exception type codes, SHA-3 hash values, or blockchain transaction IDs. Requests are made using a RESTful API interface with the standard format of POST / api / trace, with the request body containing query parameters and user credentials. For example, a typical traceability request might be "batch_id: HN2024A001, time_range: 2024-03-15 to 2024-03-16, defect_type: thread_damage." The system performs parameter validation and format checks on received requests to ensure the validity and integrity of the query identifiers. Request processing utilizes an asynchronous queue mechanism, supporting concurrent processing of multiple query requests and maintaining an average response time of less than 200 milliseconds.

[0103] Through an API that interacts with the distributed blockchain, the query identifier in the traceability request is used to retrieve matching candidate transaction records from the blockchain's on-chain index. The blockchain index utilizes a multi-level index structure consisting of a primary index, a time index, a batch index, and a hash index, supporting efficient multi-dimensional queries. The primary index is sorted by transaction ID, the time index is constructed using a B+ tree structure based on block timestamps, the batch index uses a hash table to map batch numbers to transaction records, and the hash index directly maps SHA-3 values ​​to specific transactions. The retrieval algorithm first selects the optimal index path based on the query type. The API uses the gRPC protocol to communicate with blockchain nodes. The query format is "SELECT transaction_id, block_hash, timestamp FROM blockchain_index WHERE batch_id = 'HN2024A001' AND timestamp BETWEEN '2024-03-15' AND '2024-03-16'". The retrieval process supports fuzzy matching and range queries, and pre-screens candidate records using Bloom filters to reduce unnecessary disk access. The search results are arranged in chronological order, making it easier for users to track the process and trend of anomalies along the timeline.

[0104] Each retrieved candidate transaction record undergoes rigorous block hash chain integrity verification and transaction digital signature validity checks to ensure data authenticity and integrity. Block hash chain verification begins with the genesis block, verifying the correctness of each block's hash calculation. Recursive verification ensures that the entire chain from the genesis block to the target block is intact. Transaction signature verification utilizes the Elliptic Curve Digital Signature Algorithm (ECDSA). The verification process consists of two steps: first, calculating the hash value of the transaction data, then verifying the signature using the public key. The verification process also includes a timestamp check to ensure that the transaction time is within a reasonable range and prevent replay attacks. For a query result containing 15 candidate records, the verification process checks all 15 transaction signatures and the hash chains of the associated blocks. The verification result includes detailed status information, such as "Block Hash Verification: Passed, Signature Verification: Passed, Timestamp Verification: Passed," providing users with a clear indication of data credibility.

[0105] Extract the encrypted and stored exception data packets and the corresponding encrypted hash values ​​from the verified candidate transaction records to complete the complete recovery and parsing of the data. The extraction process first obtains the Base64-encoded exception data packet from the data payload field of the blockchain transaction, and then performs a decoding operation to obtain the original Avro serialized binary data. The integrity of the extracted data is then verified by recalculating the SHA-3 hash value and comparing it with the stored hash value. After the integrity verification is passed, the Avro deserializer is used to restore the binary data into a structured exception record, which contains complete information such as image data, batch information, risk assessment results, and quality judgment results. Finally, a complete exception record report is generated, which contains information such as the original data, verification status, and extraction time, providing complete and reliable data support for quality analysis and problem tracing.

[0106] The present invention also discloses a vehicle brake assembly abnormality identification system based on image recognition, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. The system is characterized in that when the processor executes the computer program, the vehicle brake assembly abnormality identification method based on image recognition as described in any one of the above embodiments is implemented.

[0107] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0108] Among them, the memory can be an internal storage unit of a computer device, such as a hard disk or memory of a computer device, or an external storage device of a computer device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD) or flash memory card (FC) equipped on the computer device. In addition, the memory can also be a combination of an internal storage unit and an external storage device of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or is to be output. This application does not impose any restrictions on this.

[0109] The present invention also discloses a computer-readable storage medium having instructions stored thereon. When the instructions are executed by a processor, the processor is configured to execute the method for identifying automobile brake assembly abnormalities based on image recognition as described in any one of the above embodiments.

[0110] Among them, the computer program can be stored in a machine-readable medium, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or certain middleware, etc. The machine-readable medium includes any entity or device that can carry computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the machine-readable medium includes but is not limited to the above-mentioned components.

[0111] Among them, through this computer-readable storage medium, the automobile brake assembly abnormality identification method based on image recognition in the above embodiment is stored in the computer-readable storage medium, and is loaded and executed on the processor to facilitate the storage and application of the above method.

[0112] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0113] The one or more embodiments of this application are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this application should be included in the scope of protection of this application.

Claims

1. A method for identifying abnormalities in automobile brake assembly based on image recognition, characterized in that: The steps include: Collect the original image of the wheel hub bearing nut entering the assembly station, and identify and record the batch information of the wheel hub bearing nut; A deep learning model is used to extract the visual fingerprint features of the wheel hub bearing nut from the original image and compare the visual fingerprint features with pre-stored visual defect features to complete an early risk assessment of the wheel hub bearing nut; Before the hub bearing nut is tightened and anchored, a pre-anchoring image is acquired; after the same hub bearing nut is tightened and anchored, a post-anchoring image is acquired at the same shooting position; Perform image registration and differential calculations on the pre-anchoring image and the post-anchoring image to generate a differential image. Use a machine learning model to analyze the microscopic physical change characteristics caused by the anchoring operation in the differential image and output a quality judgment result of the anchoring operation. Combining the early risk assessment results with the quality judgment results, it is determined whether there is any assembly abnormality in the assembly state of the hub bearing nut; If there is an assembly abnormality in the assembly status of the hub bearing nut, all the collected image information, batch information, early risk assessment results and quality judgment results of the hub bearing nut are packaged into an abnormal data packet, the encrypted hash value of the abnormal data packet is calculated, and the abnormal data packet and the encrypted hash value are stored in the preset distributed blockchain.

2. The method for identifying automobile brake assembly anomalies based on image recognition according to claim 1, characterized in that: The method of extracting the visual fingerprint features of the hub bearing nut from the original image using the deep learning model includes the following steps: Perform preprocessing operations on the original image, including image denoising, illumination normalization, and accurate segmentation of the region of interest based on contour detection; The preprocessed original image is input into the pre-trained convolutional autoencoder network, and the encoder part of the convolutional autoencoder network performs nonlinear compression transformation on the original image; After performing nonlinear compression transformation on the original image, the latent space feature vector of the original image is extracted from the middle bottleneck layer of the convolutional autoencoder network as the initial visual fingerprint of the hub bearing nut; The principal component analysis method is applied to the initial visual fingerprint for dimensional optimization and feature selection to enhance the initial visual fingerprint's ability to express defect-sensitive features and form the visual fingerprint characteristics of the hub bearing nut.

3. The method for identifying automobile brake assembly anomalies based on image recognition according to claim 2, characterized in that: The comparison of the visual fingerprint features with the pre-stored visual defect features to complete the early risk assessment of the hub bearing nut includes the following steps: Accessing a historical risk database, where visual defect features corresponding to nut samples of various defect types are pre-stored in the historical risk database; The weighted cosine similarity is used to calculate the matching score between the visual fingerprint features and the visual defect features; Combine the matching scores and the multi-level matching score thresholds set for different defect types to determine whether the visual fingerprint feature has a statistically significant association with any defect type; If there is a significant statistical correlation, the corresponding wheel hub bearing nut will be marked with a preliminary abnormal early risk assessment result based on the associated defect type and the confidence interval of the matching score, and the visual fingerprint feature will be updated to the historical risk database; If there is no significant statistical association or the matching score is lower than the preset warning threshold, it is an early risk assessment result that the corresponding wheel hub bearing nut mark is initially normal.

4. The method for identifying automobile brake assembly anomalies based on image recognition according to claim 1, characterized in that: The performing of image registration and difference operation on the pre-anchoring image and the post-anchoring image to generate a difference image comprises the following steps: The accelerated robust features and ORB algorithm are used to detect local feature points in the key structure areas of the image before anchoring and the image after anchoring respectively; FLANN matching is performed based on the descriptor of local feature points to establish the initial correspondence between the local feature points in the image before and after anchoring. The wrong matching point pairs are eliminated through the random sampling consistency algorithm to obtain the accurately corresponding feature points after screening. Based on the precise correspondence of feature points, the homography transformation model is used to align the anchored image with the pre-anchored image at the sub-pixel level. Perform a normalized grayscale value subtraction operation on the same pixel position of the aligned pre-anchor image and post-anchor image to obtain an initial difference image containing positive and negative difference information; The initial difference image is subjected to Gaussian filtering to smooth the noise, and the adaptive threshold binarization method is applied to highlight the microscopic physical change characteristics caused by the anchoring operation to obtain the difference image.

5. The method for identifying automobile brake assembly anomalies based on image recognition according to claim 4, characterized in that: The method of using a machine learning model to analyze the microscopic physical change characteristics caused by the anchoring operation in the differential image and outputting a quality judgment result of the anchoring operation includes the following steps: The difference image is input into an anchoring anomaly recognition model built on a densely connected network. The anchoring anomaly recognition model is pre-trained through supervised learning using difference image samples containing different anchoring states. The deep convolutional layer of the anchoring anomaly recognition model is used to extract hierarchical feature maps from the differential image, which represent the morphology of the tool contact area, the degree of material micro-plastic deformation, and the surface gloss and texture changes. Based on the hierarchical feature map, the fully connected layer and softmax classification layer of the anchoring anomaly recognition model are used to output discrete category judgments of the anchoring quality. At the same time, through the parallel regression branch or attention mechanism in the anchoring anomaly recognition model, the confidence score of the anchoring quality judgment and the heat map indicating the location of abnormal micro features in the difference image are output; If the discrete category judgment result is a non-qualified state, a text description of the microscopic physical change characteristics associated with the non-qualified state is decoded from the hierarchical feature map and output; The text description and all output results of the anchoring anomaly recognition model are integrated into the quality judgment result of the anchoring operation.

6. The method for identifying automobile brake assembly anomalies based on image recognition according to claim 1, characterized in that: The method of identifying whether there is an assembly abnormality in the hub bearing nut assembly state by integrating the assessment result of the early risk assessment and the quality judgment result comprises the following steps: An exception decision engine is built based on weighted fuzzy logic. The input of the exception decision engine is the early risk assessment results and quality judgment results. Preset fuzzy membership functions and IF-THEN rule bases for different combinations of early risk assessment results and quality judgment results; Combining the fuzzy membership function and the IF-THEN rule base, and through the fuzzification, fuzzy reasoning, and defuzzification processes of the abnormal decision engine, the comprehensive evaluation score of the hub bearing nut assembly status is calculated; The comprehensive evaluation score is compared with the preset multi-level threshold value, and whether there is any assembly abnormality in the assembly state of the hub bearing nut is determined based on the comparison result.

7. The method for identifying automobile brake assembly anomalies based on image recognition according to claim 1, characterized in that: The steps of packaging all collected image information, batch information, early risk assessment results, and quality judgment results of the hub bearing nut into an abnormal data packet, calculating an encrypted hash value of the abnormal data packet, and storing the abnormal data packet and the encrypted hash value in a preset distributed blockchain include the following steps: All collected image information, batch information, early risk assessment results, and quality judgment results of wheel hub bearing nuts are time-stamped and synchronized, and integrated into key abnormality data; Encode the integrated key exception data in Avro serialization format to form a binary exception data packet; Apply a cryptographic hash function based on the SHA-3 algorithm family to operate on the abnormal data packet to obtain a fixed-length cryptographic hash value. The cryptographic hash value is the unique content fingerprint of the abnormal data packet on the blockchain. The encrypted hash value and the abnormal data packet are encapsulated and stored in the preset distributed blockchain.

8. The method for identifying automobile brake assembly anomalies based on image recognition according to claim 7, characterized in that: The method comprises the following steps: receiving a traceback request from an authorized user and including one or more query identifiers; Retrieving matching candidate transaction records from an on-chain index of the distributed blockchain using an application program interface interacting with the distributed blockchain and utilizing a query identifier in the traceability request; For each candidate transaction record retrieved, verify the integrity of the candidate transaction record block hash chain and the validity of the transaction signature; Extract the encrypted and stored abnormal data packets and the corresponding encrypted hash values ​​from the verified candidate transaction records.

9. An automobile brake assembly anomaly recognition system based on image recognition, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for identifying automobile brake assembly abnormality based on image recognition according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by a processor, the processor is configured to execute the method for identifying automobile brake assembly abnormality based on image recognition according to any one of claims 1 to 8.

Images