A ceramic product defect detection and analysis method and system

Abstract

The present invention discloses a method and system for defect detection and analysis of ceramic products, which relates to the field of intelligent detection and analysis technology, including: collecting multimodal data of ceramic products and performing data preprocessing; using image analysis methods to identify defect areas on the ceramic surface after preprocessing, and marking their positions and ranges; building a classification model, and using the identified defect area data as input to classify defect types and extract the characteristics of each defect; comparing the classified defect characteristics with a preset production process parameter rule library to analyze the possible causes of defect formation; and generating a detection report based on the classification and analysis results. The present invention realizes the accurate identification and classification of surface defects of ceramic products. It combines 2D images with 3D morphological data, greatly improving the comprehensiveness and accuracy of detection; at the same time, through the cause analysis model, the defect characteristics are associated with the production process parameters, providing clear cause inferences and optimization suggestions.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent detection and analysis, and in particular to a method and system for detecting and analyzing defects in ceramic products. Background Art

[0002] As important industrial materials and daily consumer goods, the manufacturing quality of ceramic products directly affects their performance and market competitiveness. In the production process of ceramic products, due to factors such as fluctuations in raw material properties, inaccurate production process control, and unstable equipment operation, surface and internal defects such as cracks, pores, and glaze flaws are prone to occur. These defects not only affect the appearance and physical properties of the product, but may also shorten the product's service life and pose safety hazards. Traditional ceramic product quality inspection methods are mainly based on manual visual inspection, combined with a small amount of simple physical inspection equipment, relying on the experience and subjective judgment of inspectors, and are difficult to meet the modern ceramic industry's needs for large-scale, high-precision, and high-efficiency inspections.

[0003] In recent years, with the rapid development of computer vision technology, image processing technology and machine learning, ceramic product defect detection methods based on image analysis have been initially applied. This type of method collects surface data or morphological information of ceramic products through industrial cameras, laser scanners and other equipment, and uses image processing technologies such as edge detection and region segmentation to achieve defect recognition. However, when dealing with complex ceramic surface characteristics (such as high gloss and complex textures), these methods often have problems such as low detection accuracy and high false detection rate. In addition, existing methods usually only stay at the stage of simple defect recognition and classification, fail to deeply analyze the causes of defects, and lack effective association with production process parameters, making it difficult to provide reliable decision support for subsequent process optimization.

[0004] The existing technology has obvious deficiencies in the following aspects: First, data collection is single and mostly relies on two-dimensional images, and fails to effectively combine multimodal information such as three-dimensional morphological data, resulting in poor detection of complex defects (such as deep cracks or surface depressions); Second, there is a lack of comprehensive preprocessing of the collected data, which is easily affected by noise interference or changes in illumination, weakening the robustness of defect detection; Third, the classification models are mostly based on simple feature engineering, and fail to make full use of advanced technologies such as deep learning to extract the global and local features of defects, resulting in insufficient classification accuracy and generalization ability; Fourth, the existing technology cannot effectively associate defect features with production process parameters, lacks cause analysis and process optimization functions, and it is difficult to achieve closed-loop management from defect detection to quality improvement. Summary of the invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is: how to improve the accuracy and efficiency of ceramic product defect detection, and at the same time achieve in-depth analysis of the causes of defects and generation of production process optimization suggestions.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In the first aspect, an embodiment of the present invention provides a ceramic product defect detection and analysis method, including collecting multimodal data of ceramic products and performing data preprocessing; using an image analysis method to identify defect areas on the ceramic surface after preprocessing, and marking their positions and ranges; constructing a classification model, and using the identified defect area data as input to classify the defect types and extract the characteristics of each defect; comparing the classified defect characteristics with a preset production process parameter rule library to analyze the possible causes of the defect formation; and generating a detection report based on the classification and analysis results.

[0009] As a preferred solution of the ceramic product defect detection and analysis method of the present invention, the collecting of multimodal data of the ceramic product includes collecting 2D image data and 3D morphology data of the surface of the ceramic product, wherein:

[0010] Collect 2D image data of the ceramic surface through industrial cameras to obtain the texture, color and gloss characteristics of the ceramic surface;

[0011] The 3D topographic data of ceramic products is collected by laser scanning equipment to obtain the geometric shape and height information of the ceramic surface.

[0012] As a preferred solution of the ceramic product defect detection and analysis method described in the present invention, the data preprocessing includes: denoising the collected 2D image data and using an adaptive filtering algorithm to eliminate noise interference; performing surface fitting on the collected 3D morphology data and smoothing the morphology data of the ceramic surface to remove collection errors; standardizing the 2D image data and 3D morphology data to have a unified resolution and coordinate reference system; extracting preliminary features of the ceramic surface, including texture features, geometric features and color features, as input data for subsequent defect analysis.

[0013] As a preferred solution of the ceramic product defect detection and analysis method of the present invention, the identifying defective areas on the ceramic surface includes performing edge detection processing on the 2D image data, using the Canny algorithm to identify edge features of the ceramic surface, and extracting edge information related to the defect;

[0014] Perform regional segmentation processing on the 3D morphology data, and use a segmentation algorithm based on region growth to segment the abnormal geometric morphology area on the ceramic surface;

[0015] The edge information of the 2D image data is fused with the segmentation result of the 3D morphology data to generate a multimodal feature map containing the defect area;

[0016] By setting the morphological parameter threshold, the suspected defect area in the multimodal feature map is screened to determine the boundary, location and range of the defect area.

[0017] As a preferred solution of the ceramic product defect detection and analysis method described in the present invention, wherein: the construction of the classification model includes constructing a convolutional neural network classification model, which is composed of an input layer, a feature extraction layer, a fully connected classification layer and an output layer, wherein: the input layer receives 2D image data and 3D morphology data, and extracts color, texture and geometric features respectively;

[0018] The feature extraction layer extracts local and global features through convolution and pooling operations; the fully connected classification layer integrates the extracted features to classify the defective areas;

[0019] The output layer outputs the classification results and feature parameters, including the type, location, size, shape parameters and depth information of the defect;

[0020] In the feature extraction layer, a convolution operation is performed, and its calculation formula is:

[0021] ;

[0022] in, After the convolution operation, the and Channel The output eigenvalues ​​of For input data at location Pixel value or point cloud height value; is the convolution kernel at position and Channel The weight matrix of ; For Channel The bias term of

[0023] The eigenvalues ​​after the convolution operation are transformed nonlinearly through the ReLU activation function, and the calculation formula is:

[0024] ;

[0025] The convolution feature values ​​are pooled to reduce the data dimension. The pooling calculation formula is:

[0026] ;

[0027] in, is the output feature value after pooling, are the row and column indices of the input feature matrix, is the input data value within the pooling window;

[0028] The output of the feature extraction layer is normalized, and the normalization formula is:

[0029] ;

[0030] in, is the normalized eigenvalue, is the input feature value, is the mean of the eigenvalues, is the variance of the eigenvalue, To prevent small positive numbers from dividing by zero;

[0031] In the fully connected classification layer, the classification probability of each defect type is calculated. The classification probability calculation formula is:

[0032] ;

[0033] in, For input data Belongs to category The probability of is the actual category label of the input data, For the classification network, The score, is the total number of categories;

[0034] In the output layer, the shape parameters of the defect area are calculated, and the shape factor calculation formula is:

[0035] ;

[0036] in, is the shape factor, is the area of ​​the defect region, is the perimeter of the defect area.

[0037] As a preferred embodiment of the ceramic product defect detection and analysis method of the present invention, the possible causes of defect formation analysis include extracting the main features of the defect, including the type, location, size, depth and shape factor of the defect, based on the defect type and characteristic parameters output by the classification model;

[0038] Compare the extracted defect features with the production process parameter rule base, which includes: the range of firing temperature and time, the ratio and particle size distribution of ceramic raw materials, the glaze thickness and uniformity parameters, and the pressure range and uniformity during the pressing process;

[0039] During the comparison process, the possible causes of the defect formation are evaluated based on the following matching rules:

[0040] When the depth value of the defect feature Exceeding the set threshold and the firing temperature Deviation When there is a linear correlation, it is inferred that the defect may be caused by abnormal firing temperature;

[0041] When the shape factor of the defect feature Less than threshold , and the glaze thickness Deviation When it exceeds the specified range, it is inferred that the defect may be caused by uneven glazing;

[0042] When the position distribution of defect features and the pressing pressure The unevenness When matching, it is inferred that the defect may be caused by abnormalities in the pressing process;

[0043] Use the causal reasoning model to correlate the comparison results with the preset rules and calculate the possibility of defect formation using the following formula:

[0044] ;

[0045] in, Cause Defect characteristics Cause The probability of causing Cause Defect characteristics In the cause The probability of the following occurring; Cause Cause The prior probability of is the number of possible causes; Defect characteristics In the cause The probability of the following occurring; Cause Cause The prior probability of

[0046] Based on the analysis results, possible causes of defects are output and ranked according to their likelihood, generating recommendations for optimizing the production process.

[0047] As a preferred solution of the ceramic product defect detection and analysis method of the present invention, wherein: the generation of the detection report according to the classification and analysis results includes summarizing the classification results, characteristic parameters and cause analysis results of the defects;

[0048] Generate a test report based on the summary information, which includes defect type, location, size, cause and corresponding production process optimization suggestions;

[0049] Display the test report in a visual form and output it in the form of electronic documents;

[0050] Upload the generated test report to the cloud management platform to support real-time viewing and sharing.

[0051] In a second aspect, an embodiment of the present invention provides a ceramic product defect detection and analysis system, comprising:

[0052] Data acquisition and preprocessing module: collect multimodal data of ceramic products and perform data preprocessing;

[0053] An image analysis module is used to identify defective areas on the ceramic surface using image analysis methods on the preprocessed data and mark their locations and ranges;

[0054] The classification and feature extraction module is used to build a classification model and use the identified defect area data as input to classify the defect types and extract the features of each defect;

[0055] The defect cause analysis module is used to compare the classified defect characteristics with the preset production process parameter rule library to analyze the possible causes of the defect;

[0056] The test report generation module is used to generate a test report based on the classification and analysis results.

[0057] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0058] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the method described above when executed by a processor.

[0059] Beneficial effects of the present invention:

[0060] 1. The present invention utilizes automated data collection and processing procedures, as well as a deep learning-based defect classification model, to quickly complete defect identification, classification, and feature extraction, significantly reducing detection time and meeting the online detection needs of large quantities of ceramic products;

[0061] 2. By comparing the classified defect characteristics with the preset production process parameter rule library, the present invention can deeply analyze the possible causes of defect formation, provide a cause probability calculation model based on causal reasoning, and provide a scientific basis for the quality control of ceramic products;

[0062] 3. The test report generated by the present invention contains the defect type, location, size and cause analysis results, and is accompanied by production process optimization suggestions for specific causes, such as adjusting the firing temperature, improving the glaze uniformity or optimizing the pressing pressure, providing clear guidance for production management and helping to reduce the defect rate;

[0063] 4. The present invention realizes a complete process from data collection, defect detection to cause analysis and process optimization, forming a systematic ceramic product quality management solution, and providing technical support for the intelligent and refined production of the ceramic manufacturing industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0065] Figure 1 An overall flow chart of a ceramic product defect detection and analysis method provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0067] Example 1, reference Figure 1 , as an embodiment of the present invention, provides a ceramic product defect detection and analysis method, comprising:

[0068] S1: Collect multimodal data of ceramic products and perform data preprocessing.

[0069] Collect 2D image data and 3D topography data of the ceramic product surface, including:

[0070] Collect 2D image data of the ceramic surface through industrial cameras to obtain the texture, color and gloss characteristics of the ceramic surface;

[0071] The 3D topographic data of ceramic products is collected by laser scanning equipment to obtain the geometric shape and height information of the ceramic surface.

[0072] The collected 2D image data is denoised and an adaptive filtering algorithm is used to eliminate noise interference. Adaptive filtering is an algorithm that dynamically adjusts filtering parameters and can make real-time adjustments based on noise characteristics. It retains key texture information while denoising, which can significantly reduce noise interference in 2D images while avoiding the loss of features of defects such as tiny cracks.

[0073] The collected 3D morphological data is subjected to surface fitting processing to smooth the morphological data of the ceramic surface to remove the collection error; surface fitting is a technology for smoothing 3D morphological data, using mathematical models (such as bicubic spline interpolation) to reduce the collection error while retaining the continuity of concave and convex features to facilitate subsequent geometric feature extraction.

[0074] Standardize 2D image data and 3D topography data to have a uniform resolution and coordinate reference system;

[0075] Extract preliminary features of the ceramic surface, including texture features, geometric features and color features, as input data for subsequent defect analysis.

[0076] It should be noted that industrial cameras are used to collect 2D image data of the ceramic surface to ensure high-precision capture of small defects (such as hairline cracks or slight glaze flaws). The camera adopts a fixed light source arrangement, and eliminates the interference of ambient light through a ring LED light source to ensure the consistency of the collected image. The image frame rate is set to 60 frames / second to achieve fast and continuous acquisition to meet the real-time requirements of online detection.

[0077] Use laser scanning equipment to obtain three-dimensional topographic data of the ceramic surface. The resolution of laser scanning reaches the micron level to capture small surface changes such as tiny pores or deep cracks. Use a multi-angle scanning strategy, combined with a rotating table and a laser scanning head, to achieve three-dimensional topographic data coverage of the entire surface of the ceramic product. The data format is in the form of point cloud, and is output as a standard PLY file or STL file for subsequent processing.

[0078] Furthermore, an adaptive filtering algorithm (such as Gaussian filtering) is used to denoise the 2D image data, and the filter kernel size is dynamically adjusted to balance detail preservation and noise removal:

[0079] ;

[0080] in, is the Gaussian weight function, is the Gaussian kernel (weight function) at the offset The value at the position; The filtered image at position The pixel value of For the input image at position The pixel value is the original data before filtering; The image position coordinates of the current filter value (the pixel point in the output image);

[0081] Compared with fixed filters, adaptive filtering can be adjusted according to the noise characteristics of different areas in the image. It can not only effectively remove background noise, but also retain the key features of tiny defects such as cracks.

[0082] At the same time, contrast enhancement techniques (such as Laplace filtering) are applied to highlight edge features in the image and improve the recognizability of texture features.

[0083] The surface fitting process of 3D morphology data is to use bicubic spline interpolation method to perform surface fitting on point cloud data to smooth the surface shape and remove acquisition errors:

[0084] ;

[0085] in, For a two-dimensional surface at position The height value (or other relevant eigenvalue) at is obtained by using the cubic spline basis function and the weight coefficient The results of the calculations; is the weight coefficient or the weight of the control point, each weight The local control of an area on the corresponding surface determines the shape of the surface fitting; and Represents the cubic spline basis function, which is defined in the direction and The expression of the cubic spline basis function is:

[0086] ;

[0087] in, is a parameter whose value range is [0,1].

[0088] Surface fitting reduces measurement errors in laser scanning by smoothing 3D topography data while maintaining the integrity of surface geometric features. This technology is particularly effective when processing ceramic surfaces with complex shapes and can significantly improve the accuracy of geometric feature extraction.

[0089] The smoothed data retains the concave and convex features while reducing noise points and discontinuities, facilitating subsequent geometric feature extraction.

[0090] The 2D image data and 3D shape data are standardized, including resolution adjustment and coordinate system alignment; the 2D image resolution is adjusted to a unified standard (such as 4096×4096 pixels); the point cloud coordinates of the 3D shape data are mapped to a fixed reference system, using the normalization formula:

[0091] ;

[0092] in, and are the minimum and maximum values ​​of the point cloud data, respectively. is the normalized point cloud data value, whose range is compressed to [0, 1], is a numerical value of the original point cloud data.

[0093] Use the gray level co-occurrence matrix (GLCM) to extract image texture features, including contrast, uniformity, and energy:

[0094] ;

[0095] in, represents the gray value of two pixels in the gray level co-occurrence matrix, Represents the gray value in the gray-level co-occurrence matrix and grayscale value The co-occurrence probability of Represents the square of grayscale difference, emphasizing the contribution of larger grayscale differences to contrast.

[0096] Convert the image to HSV color space and calculate the hue mean and standard deviation of each pixel to distinguish different color areas.

[0097] Calculate the surface normal and curvature features of 3D point cloud data to identify abnormal surface areas;

[0098] The curvature calculation formula is:

[0099] ;

[0100] in, is the curvature, is the normal vector, is the surface arc length, is the normal vector About Arc Length The rate of change.

[0101] The extraction of texture features can accurately capture the directionality of cracks, color features can distinguish different types of glaze defects, and the extraction of geometric features can quantify the depth of pores or the degree of cracking. Through the comprehensive extraction of these features, the identifiability and classification accuracy of defects are greatly improved.

[0102] S2: For the preprocessed data, image analysis method is used to identify the defective areas on the ceramic surface and mark their locations and ranges.

[0103] Perform edge detection on 2D image data, use Canny algorithm to identify edge features of ceramic surface, and extract edge information related to defects, specifically:

[0104] The Canny edge detection algorithm is used to identify the edge features of the ceramic surface. The Canny algorithm includes the following steps:

[0105] Compute the gradient magnitude and direction of an image:

[0106] ;

[0107] in, For images at position The gradient amplitude of and Respectively represent the horizontal and vertical gradients of the image;

[0108] Non-maximum suppression: Directionally refine the gradient amplitude and only retain possible edge pixels;

[0109] Double threshold processing: By setting a high threshold and low threshold , filter out strong edges and connected edge points.

[0110] It should be noted that the application of Canny algorithm in ceramic surface defect detection has the following advantages: it can effectively extract the boundary information of cracks or glaze defects and reduce background noise interference; through dual threshold adjustment, it can balance between strong edges (crack boundaries) and weak edges (light-colored defects).

[0111] The 3D morphology data is segmented and processed, and the abnormal geometric morphology area on the ceramic surface is segmented by a segmentation algorithm based on region growing. Specifically, the 3D morphology data is segmented by an algorithm based on region growing. The specific steps are as follows:

[0112] Select initial seed point , usually an abnormal point of surface height value;

[0113] Determine whether adjacent pixels meet the growth conditions:

[0114] ;

[0115] in, and The adjacent points and The height value of is the preset growth threshold;

[0116] Pixels that meet the conditions are added to the segmentation area and gradually expanded until the growth of all seed points is completed.

[0117] It should be noted that the region growing algorithm can accurately segment abnormal surface morphology areas, such as pores or concave areas. , which can adapt to the geometric characteristics of different types of ceramic surfaces.

[0118] Compared with the simple threshold segmentation method, the region growing algorithm can accurately segment continuous and locally changing surface shapes (such as depressions and protrusions), effectively improving the defect recognition ability of 3D data.

[0119] The edge information of the 2D image data is fused with the segmentation result of the 3D morphology data to generate a multimodal feature map containing the defect area, specifically:

[0120] The edge information of 2D image data is fused with the segmentation result of 3D shape data. The fusion process adopts weighted superposition method:

[0121] ;

[0122] in, is a multimodal feature map, is the edge detection result of the 2D image, is the segmentation result of 3D shape data, is the weighting coefficient, satisfying .

[0123] It should be noted that the key to multimodal fusion is to balance the weights of 2D and 3D data and dynamically adjust the fusion results through weighted coefficients. Traditional defect recognition often relies solely on 2D images or 3D data, while the present invention significantly improves the ability to identify complex surface defects through multimodal fusion. For example, the combination of edge information and height information makes the boundaries of cracks and pores clearer.

[0124] By setting the morphological parameter threshold, the suspected defect area in the multimodal feature map is screened to determine the boundary, location and range of the defect area, specifically:

[0125] For each suspected defect area, calculate its area , that is, the total number of pixels in the area. The area calculation is used to eliminate areas that are too small or too large, such as isolated noise points (too small area) or pseudo defects covering multiple areas (too large area). The screening conditions can be set as:

[0126] ;

[0127] in, and They represent the lower and upper limits of the area respectively, and the area meeting the conditions will be retained as the candidate defect area;

[0128] Further calculate the shape factor for the candidate area , which is defined as:

[0129] ;

[0130] in, is the area of ​​the region, is the perimeter of the region. The shape factor reflects the regularity of the region. When it is close to 1, the shape of the region is closer to a circle; for slender regions such as cracks, the shape factor value is lower; by setting a reasonable range of the shape factor, pseudo regions that do not meet the defect characteristics can be eliminated.

[0131] It should be noted that in step S2, the shape factor The calculation is used for morphological parameter screening, which can reduce the number of false detection areas by preliminarily judging whether the area has defect characteristics.

[0132] For the screened and retained areas, their boundary information is further extracted, and the precise boundary of the defective area is determined by contour detection technology (such as edge tracking based on gradient or morphological operations).

[0133] The central position (e.g., centroid coordinates) and coverage (e.g., the size of the minimum bounding rectangle) of each defect area in the feature map are calculated to provide spatial positioning information for subsequent feature extraction and classification analysis.

[0134] Considering the diversity of surface defect characteristics of ceramic products (such as cracks, pores, depressions, etc.), the threshold of morphological parameters can be dynamically adjusted according to specific detection requirements. For example, when detecting tiny cracks, the lower limit of the area can be appropriately lowered. and the lower limit of the range of shape factors.

[0135] S3: Build a classification model and use the identified defect area data as input to classify the defect types and extract the features of each defect.

[0136] Construct a convolutional neural network classification model, which consists of an input layer, a feature extraction layer, a fully connected classification layer, and an output layer. The input layer receives 2D image data and 3D shape data, and extracts color, texture, and geometric features respectively.

[0137] The feature extraction layer extracts local and global features through convolution and pooling operations; the fully connected classification layer integrates the extracted features to classify the defective areas;

[0138] The output layer outputs the classification results and feature parameters, including the type, location, size, shape parameters and depth information of the defect;

[0139] In the feature extraction layer, a convolution operation is performed, and its calculation formula is:

[0140] ;

[0141] in, After the convolution operation, the and Channel The output eigenvalues ​​of For input data at location Pixel value or point cloud height value; is the convolution kernel at position and Channel The weight matrix of ; For Channel The convolution operation can effectively extract the texture and geometric features of the local area, such as the edge features of the crack or the shape features of the concave area.

[0142] The eigenvalues ​​after the convolution operation are transformed nonlinearly through the ReLU activation function, and the calculation formula is:

[0143] ;

[0144] ReLU reduces the gradient vanishing problem and improves the training efficiency of deep networks.

[0145] The convolution feature values ​​are pooled to reduce the data dimension. The pooling calculation formula is:

[0146] ;

[0147] in, is the output feature value after pooling, are the row and column indices of the input feature matrix, is the input data value within the pooling window. The pooling operation further improves the spatial invariance of the model, making it robust to defect areas of different sizes.

[0148] The output of the feature extraction layer is normalized, and the normalization formula is:

[0149] ;

[0150] in, is the normalized eigenvalue, is the mean of the eigenvalues, is the variance of the eigenvalue, To prevent small positive numbers from dividing by zero;

[0151] In the fully connected classification layer, the classification probability of each defect type is calculated. The classification probability calculation formula is:

[0152] ;

[0153] in, For input data Belongs to category The probability of is the actual category label of the input data, For the classification network, The score, is the total number of categories. The Softmax classifier is able to effectively distinguish different types of defects, such as cracks, pores, or glaze flaws.

[0154] In the output layer, the shape parameters of the defect area are calculated, and the shape factor calculation formula is:

[0155] ;

[0156] in, is the shape factor, is the area of ​​the defect region, is the perimeter of the defect area.

[0157] It should be noted that in step S3, the shape factor The calculation of is used as one of the output features of the classification model, which is further used to describe the specific morphological characteristics of the defect (such as the relationship between area and perimeter), providing support for defect classification and feature analysis.

[0158] S4: Compare the classified defect characteristics with the preset production process parameter rule library to analyze the possible causes of the defects.

[0159] Features for each defect area are extracted from the output of the S3 classification model, including:

[0160] Defect type: such as cracks, pores, glaze flaws, etc. Defect location: coordinates of the center of mass in a two-dimensional plane or three-dimensional morphology Defect size: length, width and depth of the defect Defect shape factor. Describes the morphological regularity of the defect area.

[0161] The extracted defect features are compared with the production process parameter rule base, which includes: the range of firing temperature and time, the ratio and particle size distribution of ceramic raw materials, the glaze thickness and uniformity parameters, and the pressure range and uniformity during the pressing process; specifically:

[0162] Firing temperature and time: range to , and the actual temperature deviation ;

[0163] Raw material ratio and particle size distribution: used to determine the impact of raw material particle distribution on defects;

[0164] Glaze thickness and uniformity: thickness range to and deviation ;

[0165] Pressing pressure and uniformity: pressure range to and inhomogeneity .

[0166] By comparing the defect characteristic parameters with the process parameter range in the rule library, the possible causes of defects can be evaluated. For example, firing temperature or pressure unevenness outside the set range may be the cause of cracks or pores.

[0167] During the comparison process, the possible causes of the defect formation are evaluated based on the following matching rules:

[0168] When the depth value of the defect feature Exceeding the set threshold and the firing temperature Deviation When there is a linear correlation, it is inferred that the defect may be caused by abnormal firing temperature;

[0169] When the shape factor of the defect feature Less than threshold , and the glaze thickness Deviation When it exceeds the specified range, it is inferred that the defect may be caused by uneven glazing;

[0170] When the position distribution of defect features and the pressing pressure The unevenness When matching, it is inferred that the defect may be caused by an abnormality in the pressing process.

[0171] Use the causal reasoning model to correlate the comparison results with the preset rules and calculate the possibility of defect formation using the following formula:

[0172] ;

[0173] in, Cause Defect characteristics Cause The probability of causing Cause Defect characteristics In the cause The probability of the following occurring; Cause Cause The prior probability of is the number of possible causes; Defect characteristics In the cause The probability of the following occurring; Cause Cause The prior probability of .

[0174] Output possible defect causes based on the analysis results, sort them by likelihood, and generate suggestions for optimizing the production process, such as:

[0175] For defects caused by abnormal firing temperature, it is recommended to adjust the control range of firing temperature and time, etc.

[0176] For the defect of uneven glaze thickness, it is recommended to optimize the glaze coating equipment or adjust the glaze coating process parameters;

[0177] For defects caused by abnormal pressing pressure, it is recommended to improve the pressing die design or uniformity control strategy, etc.

[0178] Through targeted suggestions, problem links can be quickly located and improvements can be implemented, thereby reducing defect rates and improving ceramic product quality.

[0179] S5: Generate a test report based on the classification and analysis results.

[0180] Summarize the defect classification results, characteristic parameters and cause analysis results.

[0181] Generate a test report based on the summary information. The report includes defect type, location, size, cause and corresponding production process optimization suggestions; specifically:

[0182] Basic information: test product number, test time, and test equipment information;

[0183] Defect information: defect type and quantity; defect location (center of mass coordinates or area range); defect size (such as length, width, depth); defect shape characteristics (such as shape factor);

[0184] Cause analysis: the cause of each defect and its probability ranking; the process parameter deviation information of the cause (such as firing temperature, glaze thickness, etc.);

[0185] Optimization suggestions: Production process improvement measures targeting the main causes of defects, such as adjusting the firing temperature range, optimizing the uniformity of pressing pressure, etc.

[0186] The test report is displayed in a visual form and output in the form of electronic document.

[0187] Upload the generated test report to the cloud management platform to support real-time viewing and sharing; the cloud platform features include:

[0188] Permission management: supports access rights control for different user roles (such as operators, engineers, and management);

[0189] Historical records: Save the inspection history and support the retrieval and trend analysis of historical data;

[0190] Real-time reminder: When the inspection results trigger an abnormality (such as the number of defects exceeds the standard or the size is too large), an automatic warning is sent to the relevant person in charge.

[0191] Use cloud storage services (such as AWS S3, Google Cloud Storage) or private clouds within the enterprise, combined with web front-end technology to achieve real-time viewing and interaction.

[0192] Embodiment 2 is the second embodiment of the present invention, which is different from the previous embodiment in that:

[0193] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art or partly in the current technical solution. The current computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0194] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0195] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0196] Embodiment 3 is an embodiment of the present invention, which provides a ceramic product defect detection and analysis system, including a data acquisition and preprocessing module, an image analysis module, a classification and feature extraction module, a defect cause analysis module and a detection report generation module.

[0197] Data acquisition and preprocessing module: collect multimodal data of ceramic products and perform data preprocessing;

[0198] Image analysis module: for the pre-processed data, the image analysis method is used to identify the defective areas on the ceramic surface and mark their locations and ranges;

[0199] Classification and feature extraction module: builds a classification model and uses the identified defect area data as input to classify defect types and extract features of each defect;

[0200] Defect cause analysis module: compares the classified defect characteristics with the preset production process parameter rule library to analyze the possible causes of the defects;

[0201] Test report generation module: Generates test reports based on classification and analysis results.

[0202] Example 4 is an embodiment of the present invention, which provides a ceramic product defect detection and analysis method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation / comparative experiments.

[0203] The purpose of this experiment is to verify the effectiveness of the ceramic product defect detection and cause analysis method, and to prove its innovation and practical value through experimental data. The experiment selected 6 different batches of ceramic samples (numbered Ceramic_01 to Ceramic_06), and completed the inference of defect causes and generation of optimization suggestions through multimodal data collection, analysis and classification combined with the production process parameter rule library.

[0204] Experimental preparation:

[0205] Equipment: industrial camera (resolution 4096x4096, frame rate 60fps), laser scanner (resolution 10µm), server (GPU acceleration support).

[0206] Samples: Randomly selected ceramic products with different surface textures and geometrical properties.

[0207] Multimodal data acquisition:

[0208] Use industrial cameras to collect 2D image data of samples to ensure that the surface texture area of ​​each sample is covered;

[0209] A laser scanner is used to collect 3D topographic data, record the height distribution of each sample surface, and generate point cloud data.

[0210] Store acquired data in standardized formats such as PLY and PNG.

[0211] Data preprocessing:

[0212] Apply Gaussian filtering to denoise 2D image data and combine it with histogram equalization to enhance surface details;

[0213] Perform surface fitting and smoothing on 3D point cloud data to eliminate random errors in data collection;

[0214] Standardize data resolution and coordinate systems to enable seamless fusion of 2D and 3D data.

[0215] Defective area identification:

[0216] Apply Canny edge detection to 2D image data to mark the boundaries of surface cracks or flaws;

[0217] Apply region growing algorithm to 3D topography data to identify surface defects such as depressions and pores;

[0218] Combine 2D and 3D data to generate multimodal defect feature maps and screen suspected defect areas.

[0219] Classification and feature extraction:

[0220] The defect areas are classified by convolutional neural network (CNN) to extract the type, area, depth and shape factor of each defect;

[0221] The model was optimized using 1,000 training data sets, and the classification accuracy reached over 95%.

[0222] Cause Analysis:

[0223] The extracted defect features are compared with the production process rule library, and the cause of the defect is evaluated using the Bayesian reasoning model.

[0224] The causes include abnormal firing temperature, uneven glaze thickness, and deviation in pressing pressure.

[0225] Test report generation:

[0226] Generate a test report containing defect type, cause analysis and optimization suggestions based on the classification and analysis results.

[0227] Export reports in PDF format and upload them to the cloud management platform to support real-time access and analysis.

[0228] For specific experimental data, please refer to Table 1.

[0229] Table 1 Experimental data reference table

[0230]

[0231] It can be seen from the experimental data that the multimodal defect detection method of the present invention significantly improves the accuracy of defect classification and cause analysis. The specific analysis is as follows:

[0232] Defect characterization and classification: Data showing the area and shape factor of the defect region It plays an important role in the classification of defect types, for example, the crack area The values ​​are generally low (e.g. 0.49 for Ceramic_01 and 0.53 for Ceramic_04), while the values ​​for pores and glaze defects are The values ​​are relatively high (e.g., 0.73 for Ceramic_02 and 0.68 for Ceramic_03), which reflects the high efficiency of the model classification.

[0233] Scientific nature of cause analysis: Cause analysis combined with the production process parameter rule base can accurately infer the causes of defects, such as:

[0234] The cracked samples (ceramic_01, ceramic_04) showed high temperature deviations (42%, 51%), indicating that abnormal firing temperature was the main cause;

[0235] The porous samples (ceramic_02, ceramic_05) were highly correlated with the pressure deviation (32%, 28%), further verifying the effect of uneven pressing;

[0236] The samples with glaze defects (ceramic_03, ceramic_06) showed a high probability of uneven glaze thickness (78%, 69%), and its cause was clearly identified.

[0237] The present invention improves the accuracy of defect detection by combining 2D and 3D data for multimodal fusion; the cause analysis adopts the Bayesian reasoning model to achieve a comprehensive association from features to processes. Compared with the traditional single detection method, the present invention has significant advantages in classification accuracy, cause inference ability and process optimization guidance.

[0238] The inspection report generated in the experiment contains detailed defect information and optimization suggestions. According to the report, the company can adjust the firing temperature, improve the pressing process or optimize the glazing process to reduce the defect rate and improve product quality.

[0239] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A ceramic product defect detection and analysis method, characterized in that: include: Collect multimodal data of ceramic products and perform data preprocessing; For the pre-processed data, image analysis method is used to identify defective areas on the ceramic surface and mark their locations and ranges; Build a classification model and use the identified defect area data as input to classify the defect types and extract the characteristics of each defect; Compare the classified defect features with the preset production process parameter rule library to analyze the possible causes of the defects; Generate a test report based on the classification and analysis results; The construction of the classification model includes: constructing a convolutional neural network classification model, which is composed of an input layer, a feature extraction layer, a fully connected classification layer and an output layer, wherein: the input layer receives 2D image data and 3D shape data, and extracts color, texture and geometric features respectively; The feature extraction layer extracts local and global features through convolution and pooling operations; the fully connected classification layer integrates the extracted features to classify the defective areas; The output layer outputs the classification results and feature parameters, including the type, location, size, shape parameters and depth information of the defect; In the feature extraction layer, a convolution operation is performed, and its calculation formula is: ； in, After the convolution operation, the and Channel The output eigenvalues ​​of For input data at location Pixel value or point cloud height value; is the convolution kernel at position and Channel The weight matrix of ; For Channel The bias term of The eigenvalues ​​after the convolution operation are transformed nonlinearly through the ReLU activation function, and the calculation formula is: ； The convolution feature values ​​are pooled to reduce the data dimension. The pooling calculation formula is: ； in, is the output feature value after pooling, are the row and column indices of the input feature matrix, is the input data value within the pooling window; The output of the feature extraction layer is normalized, and the normalization formula is: ； in, is the normalized eigenvalue, is the input feature value, is the mean of the eigenvalues, is the variance of the eigenvalue, To prevent small positive numbers from dividing by zero; In the fully connected classification layer, the classification probability of each defect type is calculated. The classification probability calculation formula is: ； in, For input data Belongs to category The probability of is the actual category label of the input data, For the classification network, The score, is the total number of categories; In the output layer, the shape parameters of the defect area are calculated, and the shape factor calculation formula is: ； in, is the shape factor, is the area of ​​the defect region, is the perimeter of the defect area; The analyzing possible causes of defect formation includes extracting main features of the defect, including defect type, location, size, depth and shape factor, based on the defect type and feature parameters output by the classification model; Compare the extracted defect features with the production process parameter rule base, which includes: the range of firing temperature and time, the ratio and particle size distribution of ceramic raw materials, the glaze thickness and uniformity parameters, and the pressure range and uniformity during the pressing process; During the comparison process, the possible causes of the defect formation are evaluated based on the following matching rules: When the depth value of the defect feature Exceeding the set threshold and the firing temperature Deviation When there is a linear correlation, it is inferred that the defect may be caused by abnormal firing temperature; When the shape factor of the defect feature Less than threshold , and the glaze thickness Deviation When it exceeds the specified range, it is inferred that the defect may be caused by uneven glazing; When the position distribution of defect features and the pressing pressure The unevenness When matching, it is inferred that the defect may be caused by abnormalities in the pressing process; Use the causal reasoning model to correlate the comparison results with the preset rules and calculate the possibility of defect formation using the following formula: ； in, Cause Defect characteristics Cause The probability of causing Cause Defect characteristics In the cause The probability of the following occurring; Cause Cause The prior probability of is the number of possible causes; Defect characteristics In the cause The probability of the following occurring; Cause Cause The prior probability of Based on the analysis results, possible causes of defects are output and ranked according to their likelihood, generating recommendations for optimizing the production process.

2. The ceramic product defect detection and analysis method according to claim 1, characterized in that: The collecting of multimodal data of the ceramic product includes collecting 2D image data and 3D morphology data of the surface of the ceramic product, wherein: Collect 2D image data of the ceramic surface through industrial cameras to obtain the texture, color and gloss characteristics of the ceramic surface; The 3D topographic data of ceramic products is collected by laser scanning equipment to obtain the geometric shape and height information of the ceramic surface.

3. The ceramic product defect detection and analysis method according to claim 2, characterized in that: The data preprocessing comprises: De-noising the collected 2D image data, using an adaptive filtering algorithm to eliminate noise interference; Perform surface fitting processing on the collected 3D morphological data to smooth the morphological data of the ceramic surface to remove the collection error; Standardize 2D image data and 3D topography data to have a uniform resolution and coordinate reference system; Extract preliminary features of the ceramic surface, including texture features, geometric features and color features, as input data for subsequent defect analysis.

4. The ceramic product defect detection and analysis method according to claim 3, characterized in that: The identification of defective areas on the ceramic surface comprises: Perform edge detection on 2D image data, use Canny algorithm to identify edge features of ceramic surface, and extract edge information related to defects; Perform regional segmentation processing on the 3D morphology data, and use a segmentation algorithm based on region growth to segment the abnormal geometric morphology area on the ceramic surface; The edge information of the 2D image data is fused with the segmentation result of the 3D morphology data to generate a multimodal feature map containing the defect area; By setting the morphological parameter threshold, the suspected defect area in the multimodal feature map is screened to determine the boundary, location and range of the defect area.

5. The ceramic product defect detection and analysis method according to claim 4, characterized in that: Generating a test report based on the classification and analysis results includes summarizing the classification results, characteristic parameters and cause analysis results of the defects; Generate a test report based on the summary information, which includes defect type, location, size, cause and corresponding production process optimization suggestions; Display the test report in a visual form and output it in the form of electronic documents; Upload the generated test report to the cloud management platform to support real-time viewing and sharing.

6. A ceramic product defect detection and analysis system, used to implement the ceramic product defect detection and analysis method according to any one of claims 1 to 5, characterized in that: include: Data acquisition and preprocessing module, used to collect multimodal data of ceramic products and perform data preprocessing; An image analysis module is used to identify defective areas on the ceramic surface using image analysis methods on the preprocessed data and mark their locations and ranges; The classification and feature extraction module is used to build a classification model and use the identified defect area data as input to classify the defect types and extract the features of each defect; The defect cause analysis module is used to compare the classified defect characteristics with the preset production process parameter rule library to analyze the possible causes of the defect; The test report generation module is used to generate a test report based on the classification and analysis results.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the ceramic product defect detection and analysis method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the ceramic product defect detection and analysis method according to any one of claims 1 to 5 are implemented.

Images