Precise stainless steel quality detection method based on big data analysis

Through big data analysis and multimodal detection technology, the surface and internal defects of stainless steel are identified and the supervised learning model is built, which solves the problem of insufficient quality detection accuracy of stainless steel in the existing technology, and achieves comprehensive inspection and accuracy improvement of stainless steel quality.

CN119048476AInactive Publication Date: 2024-11-29XUZHOU HUIDONG PRECISION METAL CO LTD
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411177982.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing stainless steel quality inspection technology has insufficient detection accuracy, which is difficult to meet the modern industry's demand for high-quality and high-reliability products, especially in large-scale inspections, which often have missed inspection problems.

Method used

The precision stainless steel quality detection method based on big data analysis is adopted. By obtaining the stainless steel detection historical data set, surface and internal defects are identified, correlation analysis is carried out, and a supervisory learning model is constructed to achieve comprehensive inspection of stainless steel quality.

Benefits of technology

The accuracy of stainless steel quality inspection is improved, and through multimodal detection technology and defect correlation analysis, the leakage detection rate is reduced, meeting the demand of modern industry for high-quality products.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048476A_ABST
    Figure CN119048476A_ABST
Patent Text Reader

Abstract

The invention discloses a precision stainless steel quality detection method based on big data analysis, and relates to the technical field of quality detection.The method comprises the steps that a stainless steel detection historical data set is obtained; performing surface defect identification based on the stainless steel image data to obtain a stainless steel surface defect set, and performing internal defect identification based on the stainless steel internal structure data to obtain a stainless steel internal defect set; performing correlation analysis on the stainless steel surface defect set and the stainless steel internal defect set to obtain surface-internal defect correlation information; performing supervised training based on the stainless steel surface defect set, the stainless steel internal defect set and the surface-internal defect associated information to obtain a stainless steel quality detection model; and obtaining a to-be-detected stainless steel surface defect set and a to-be-detected stainless steel internal defect set, and inputting the stainless steel quality detection model to complete quality detection. The technical problem that in the prior art, the detection precision is insufficient during stainless steel quality detection is solved, and the technical effect of improving the stainless steel quality detection precision is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of quality detection, and in particular to a precision stainless steel quality detection method based on big data analysis. Background Art

[0002] With the rapid development of industrial production and manufacturing technology, stainless steel materials have been widely used in many fields, especially in industries such as aerospace, automobile manufacturing, medical equipment and construction, where the quality requirements for stainless steel are getting higher and higher. However, with the expansion of production scale and the diversification of stainless steel application scenarios, the existing stainless steel quality inspection technology faces increasing challenges. Traditional stainless steel quality inspection methods are often limited to surface inspection or a single non-destructive testing technology, which makes it difficult to ensure comprehensive quality assessment while ensuring efficient inspection. Especially when dealing with large quantities of stainless steel materials, traditional inspection methods often have problems with insufficient inspection accuracy leading to missed inspections, which makes it difficult to meet the needs of modern industry for high-quality and high-reliability products. Summary of the invention

[0003] The present application provides a precision stainless steel quality detection method based on big data analysis, which is used to solve the technical problem of insufficient detection accuracy in the prior art when performing stainless steel quality detection.

[0004] In view of the above problems, the present application provides a precision stainless steel quality detection method based on big data analysis.

[0005] The present application provides a precision stainless steel quality detection method based on big data analysis, the method comprising: A stainless steel inspection history data set is obtained, wherein the stainless steel inspection history data set includes stainless steel image data and stainless steel internal structure data, wherein each stainless steel image data has corresponding stainless steel internal structure data; surface defect recognition is performed based on the stainless steel image data to obtain a stainless steel surface defect set, and at the same time, internal defect recognition is performed based on the stainless steel internal structure data to obtain a stainless steel internal defect set; association analysis is performed based on the stainless steel surface defect set and the stainless steel internal defect set to obtain surface-internal defect association information; supervised training is performed based on the stainless steel surface defect set, the stainless steel internal defect set and the surface-internal defect association information to obtain a stainless steel quality inspection model; a stainless steel surface defect set to be inspected and a stainless steel internal defect set to be inspected are obtained, and the stainless steel quality inspection model is input to complete quality inspection.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: The present application obtains a stainless steel detection historical data set, which includes stainless steel image data and stainless steel internal structure data, wherein each stainless steel image data has corresponding stainless steel internal structure data; performs surface defect recognition based on the stainless steel image data to obtain a stainless steel surface defect set, and at the same time, performs internal defect recognition based on the stainless steel internal structure data to obtain a stainless steel internal defect set; performs correlation analysis based on the stainless steel surface defect set and the stainless steel internal defect set to obtain surface-internal defect correlation information; performs supervised training based on the stainless steel surface defect set, the stainless steel internal defect set and the surface-internal defect correlation information to obtain a stainless steel quality detection model; obtains a stainless steel surface defect set to be detected and a stainless steel internal defect set to be detected, and inputs them into the stainless steel quality detection model to complete quality detection. The present invention solves the technical problem that the prior art has insufficient detection accuracy when performing stainless steel quality detection. Through big data analysis and multimodal detection technology, the historical detection data of stainless steel, surface and internal defect recognition, and defect correlation analysis are used to construct a supervised learning model to achieve comprehensive detection of stainless steel quality, thereby achieving the technical effect of improving the accuracy of stainless steel quality detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic flow chart of a precision stainless steel quality detection method based on big data analysis provided in an embodiment of the present application.

[0009] Figure 2 A schematic diagram of the process of obtaining surface-internal defect correlation information in the precision stainless steel quality inspection method based on big data analysis provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] This application provides a precision stainless steel quality detection method based on big data analysis, which is used to solve the technical problem of insufficient detection accuracy in the existing technology when performing stainless steel quality detection. Through big data analysis and multimodal detection technology, using historical detection data of stainless steel, surface and internal defect identification, and defect correlation analysis, a supervised learning model is constructed to achieve comprehensive detection of stainless steel quality, thereby achieving the technical effect of improving the accuracy of stainless steel quality detection.

[0011] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0012] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0013] Examples, such as Figure 1 As shown, the present application provides a precision stainless steel quality detection method based on big data analysis, the method comprising: Step S100: Acquire a stainless steel detection history data set, wherein the stainless steel detection history data set includes stainless steel image data and stainless steel internal structure data, wherein each stainless steel image data has corresponding stainless steel internal structure data.

[0014] In an embodiment of the present application, a connection is first made to the data management system of a factory or enterprise, which stores a large amount of historical data accumulated during the production and testing of stainless steel. These data include image data of stainless steel surface detection and internal structure data obtained by non-destructive testing technologies such as X-rays or ultrasound. Each record contains a stainless steel surface image and its corresponding internal structure data, which are indexed and paired in the database through data association rules. Next, the corresponding stainless steel image data and internal structure data are extracted from the database by executing a query statement or calling an API interface.

[0015] Through the above process, the stainless steel inspection historical data set is obtained.

[0016] Step S200: performing surface defect recognition based on the stainless steel image data to obtain a stainless steel surface defect set, and at the same time, performing internal defect recognition based on the stainless steel internal structure data to obtain a stainless steel internal defect set.

[0017] In an embodiment of the present application, the surface image data of stainless steel is first analyzed by a pre-trained image recognition algorithm, such as a convolutional neural network, to automatically identify and extract abnormal features in the stainless steel surface image, such as scratches, dents, cracks, etc., to generate a stainless steel surface defect set.

[0018] At the same time, the internal structure data of stainless steel is obtained through non-destructive testing technology. In order to identify internal defects, a stainless steel internal defect recognition model trained based on the internal defect annotation sample data set is used. Similar to surface defect recognition, the internal defect recognition model extracts features from the input data, finds irregular structures or abnormal signal reflection areas, and thus determines the location and type of internal defects. Finally, a stainless steel internal defect set is generated.

[0019] The stainless steel surface defect set and the stainless steel internal defect set are obtained through the above process.

[0020] Furthermore, in the method provided in the embodiment of the application, surface defect recognition is performed based on the stainless steel image data to obtain a stainless steel surface defect set, and further includes: A sample stainless steel image data set is obtained, wherein the sample stainless steel image data set includes surface defect annotations; a stainless steel surface defect recognition model is constructed based on the sample stainless steel image data set; and surface defects of the stainless steel image data are recognized by using the stainless steel surface defect recognition model to obtain a stainless steel surface defect set.

[0021] In an embodiment of the present application, a sample stainless steel image dataset is first obtained from a historical database, and the dataset includes surface defect information that has been manually or automatically annotated. The annotated content includes the type of defect (such as scratches, dents, cracks, etc.), the location, size, and other features of the defect. Next, the annotated sample dataset is used for model training, and meaningful features are automatically extracted from the image through a convolutional neural network. The model optimizes its recognition ability by repeatedly learning the surface defects and their annotations in the sample images. During the training process, image processing such as data enhancement and denoising is performed to improve the generalization ability of the model and ensure that the model can still identify defects under different image conditions.

[0022] After the model training is completed, surface defects of stainless steel image data are identified based on the trained stainless steel surface defect recognition model to obtain a stainless steel surface defect set.

[0023] Furthermore, in the method provided in the embodiment of the application, internal defects are identified based on the internal structure data of the stainless steel to obtain a set of internal defects of the stainless steel, and the method also includes: A sample stainless steel internal structure data set is obtained, wherein the sample stainless steel internal structure data set includes internal defect annotation identifiers; a stainless steel internal defect recognition model is constructed based on the sample stainless steel internal structure data set; internal defects of the stainless steel internal structure data are recognized by using the stainless steel internal defect recognition model to obtain a stainless steel internal defect set.

[0024] In an embodiment of the present application, a sample stainless steel internal structure data set is first obtained from a historical database or test record. This data set is collected through non-destructive testing techniques such as X-rays, ultrasound or CT scanning, and records the structural information inside the stainless steel. These data have been defect-annotated, and the annotation content includes the type and location of defects such as internal cracks, bubbles, and voids. Next, based on the annotated sample data set, deep learning techniques such as convolutional neural networks or traditional machine learning algorithms such as support vector machines are used for model training to build a stainless steel internal defect recognition model. During the training process, the model gradually optimizes its recognition ability by learning the internal defect features in the sample, such as density changes, abnormal structures, etc. This process usually includes data preprocessing, feature extraction, and model tuning, so that the model can adapt to different internal structural features.

[0025] After the model training is completed, the internal structure data of the stainless steel is input into the trained model for automatic recognition. The model analyzes the internal structure data and identifies abnormal features and internal defects. This process detects and classifies possible defects in the stainless steel, such as cracks, holes and material inhomogeneities, through the model's automated feature extraction function.

[0026] Finally, the identified internal defect information is summarized into a stainless steel internal defect set.

[0027] Step S300: performing correlation analysis based on the stainless steel surface defect set and the stainless steel internal defect set to obtain surface-internal defect correlation information.

[0028] Further, such as Figure 2 As shown, in the method provided in the embodiment of the application, correlation analysis is performed based on the stainless steel surface defect set and the stainless steel internal defect set to obtain surface-internal defect correlation information, and also includes: The stainless steel surface defect set and the stainless steel internal defect set are coordinate mapped to obtain a stainless steel multimodal space fusion set; feature extraction is performed on the stainless steel multimodal space fusion set to obtain a stainless steel surface feature data set and a stainless steel internal feature data set; a cross-analysis is performed based on the stainless steel surface feature data set and the stainless steel internal feature data set to generate a defect association metric matrix; combined with the defect association metric matrix, the association strength between surface defects and internal defects is evaluated to generate a multimodal defect association weight grid; based on the multimodal defect association weight grid, surface-internal defect association information is generated.

[0029] In the embodiment of the present application, the stainless steel surface defect set and the stainless steel internal defect set are firstly coordinate mapped, and the surface and internal defects are aligned in the same spatial coordinate system for subsequent multimodal fusion and analysis. By mapping the surface and internal defect points into three-dimensional space, an integrated stainless steel multimodal spatial fusion set is constructed, which contains the spatial position of each defect and its corresponding multimodal data, i.e., surface and internal information.

[0030] Next, feature extraction is performed on the stainless steel multimodal spatial fusion set. The surface and internal features of stainless steel are extracted from the multimodal data using feature extraction algorithms, such as convolutional neural networks or texture-based, morphological analysis methods. The feature extraction process identifies key geometric shapes, texture patterns, density distribution and other information, and ultimately generates stainless steel surface feature datasets and stainless steel internal feature datasets.

[0031] After feature extraction, a cross-analysis is performed based on the stainless steel surface feature dataset and the stainless steel internal feature dataset. By comparing the features of surface and internal defects, the correlation between them is found. The correlation strength between each pair of surface and internal defect features is calculated using a correlation analysis algorithm, such as the Pearson correlation coefficient, to generate a defect correlation metric matrix. Based on the generated defect correlation metric matrix, the correlation strength between surface defects and internal defects is evaluated. Combined with these correlation strengths, a multimodal defect correlation weight grid is generated using a multimodal weight assignment technique. Finally, surface-internal defect correlation information is generated based on the multimodal defect correlation weight grid.

[0032] Furthermore, in the method provided in the embodiment of the application, cross analysis is performed based on the stainless steel surface feature data set and the stainless steel internal feature data set to generate a defect correlation metric matrix, and further includes: Multidimensional feature mapping is performed on the stainless steel surface feature data set and the stainless steel internal feature data set to generate a stainless steel feature vector mapping set; pairing is performed based on the stainless steel feature vector mapping set to obtain a surface-internal feature pairing set; correlation measurement is performed based on the surface-internal feature pairing set to generate a feature correlation measurement set, and the defect correlation measurement matrix is ​​generated through matrix construction.

[0033] In the embodiment of the present application, principal component analysis is first used for multidimensional feature mapping to extract multidimensional features such as geometry, texture, and position of each defect from the stainless steel surface feature data set and the stainless steel internal feature data set. PCA is used to reduce the dimensionality of these features, map the high-dimensional data into a low-dimensional space, extract the most representative features, and generate a stainless steel feature vector mapping set through this process.

[0034] Next, based on the stainless steel feature vector mapping set, the Euclidean distance is used for feature pairing. The Euclidean distance is used to measure the straight-line distance between two feature vectors. The smaller the distance, the higher their similarity. By calculating the Euclidean distance between the surface and internal defect feature vectors, the features with smaller distances are paired to generate a surface-internal feature pairing set.

[0035] After the feature pairing is completed, the Pearson correlation coefficient is used to measure the correlation. The Pearson correlation coefficient is used to evaluate the linear relationship between two features, ranging from -1 to 1. The closer it is to 1, the stronger their positive correlation is. Based on each pair of features in the surface-internal feature pairing set, the Pearson correlation coefficient is calculated, and finally a feature correlation measurement set is generated, which provides a quantitative correlation strength for each paired feature. Finally, based on the generated feature correlation measurement set, the adjacency matrix method is used to construct the defect correlation measurement matrix.

[0036] Furthermore, in the method provided in the embodiment of the application, pairing is performed based on the stainless steel feature vector mapping set to obtain a surface-internal feature pairing set, and further includes: Based on the stainless steel feature vector mapping set, a morphologically similar feature pair set is generated through morphological similarity analysis; by performing scale similarity calculation on the morphologically similar feature pair set, a scale matching feature pair set is generated; based on the scale matching feature pair set, the spatial distribution pattern of the features is matched and screened to obtain the surface-internal feature pairing set.

[0037] In the embodiment of the present application, firstly, based on the stainless steel feature vector mapping set, morphological analysis tools such as morphological descriptors and Fourier descriptors are used to perform morphological similarity analysis on the defects on the surface and inside of the stainless steel. In this step, morphological features such as geometric shapes, edges and contours of surface and internal defects are extracted. The edges and contours of stainless steel defects are analyzed and represented by Fourier descriptors to form mathematical expressions and extract shape features. Subsequently, the morphological similarity is calculated, and the morphological similarity is measured by the Hausdorff distance or a similarity metric based on shape matching. Through this analysis, pairs of surface and internal defects with similar shapes are identified, and a set of morphologically similar feature pairs is generated.

[0038] Next, the scale similarity is calculated for the set of morphologically similar feature pairs. In this process, the scale-invariant feature transform (SIFT) technique is used to analyze the scale similarity between morphologically similar feature pairs. Specifically, the SIFT feature point extraction algorithm is first applied to each pair of morphologically similar feature pairs. In this process, a large number of key points are detected in the surface and internal defect images, and their scale information is calculated for each key point. SIFT constructs a Gaussian difference pyramid to perform multi-scale analysis on the image and find the best position of each key point at different scales. Next, the scale similarity is calculated for each pair of morphologically similar feature pairs, and their similarity is evaluated by calculating the Euclidean distance of the surface and internal defect feature points. After the calculation, a similarity threshold is set. For example, when the Euclidean distance is less than 0.8, the pair of feature points is considered to match in scale. In addition, the similarity is further verified by calculating the scale difference between the two feature points. The scale difference is used to ensure that their relative scale sizes remain consistent even if the absolute sizes are different. If the scale difference exceeds the set range, these inconsistent feature pairs will be filtered out. Finally, based on these similarity calculation results and ratio verification, the filtered feature pairs are aggregated together to generate a scale-matched feature pair set.

[0039] After obtaining the scale-matching feature pair set, the spatial pattern recognition algorithm is further used to perform spatial distribution pattern matching screening on these matching features. The purpose of this step is to ensure that the spatial distribution patterns of surface and internal defects are also consistent. By extracting the spatial position data of each pair of features, such as the position in the three-dimensional coordinate system, their relative position, arrangement order and spatial structure are analyzed. Use a grid-based spatial alignment algorithm or a point cloud-based registration technology to compare the consistency of these features in spatial distribution. If the surface and internal defects also match in spatial arrangement, these feature pairs are screened out to eventually generate a surface-internal feature pairing set.

[0040] Furthermore, in the method provided in the embodiment of the application, the correlation strength between surface defects and internal defects is evaluated by combining the defect correlation metric matrix, and a multimodal defect correlation weight grid is generated, which also includes: In combination with the defect association metric matrix, disturbance calculation is performed through topological perturbation analysis, and the association change values ​​of each defect pair are counted to generate a disturbance defect association weight matrix; multi-level grading is performed based on the disturbance defect association weight matrix to generate a multi-level defect weight matrix; based on the multi-level defect weight matrix, the multi-modal defect association weight grid is generated through grid mapping.

[0041] In an embodiment of the present application, firstly, based on the defect association metric matrix, the association relationship between defect pairs is dynamically analyzed by topological perturbation analysis. In this process, a random perturbation model and graph theory analysis method are used to simulate the influence of external physical perturbations, such as stress, temperature, material fatigue, etc., on the association strength between defect pairs. A small random perturbation is applied, and the relationship between each pair of defects is adjusted to observe how the association values ​​between these defect pairs fluctuate with the change of the disturbance. In this process, the change in the association strength of each pair of defects is counted, and a matrix containing the association change values ​​of all defect pairs after disturbance is generated by calculating these changes, that is, the disturbance defect association weight matrix. This matrix quantifies the association change characteristics of each defect pair under different physical disturbance conditions.

[0042] Next, a multi-level classification was performed based on the generated disturbance defect association weight matrix. In this step, a hierarchical clustering analysis method was used to classify and grade the association change values ​​of defect pairs. Specifically, the association change values ​​in the matrix were first sorted, and defect pairs with similar association changes were clustered together. Then, using the hierarchical clustering analysis technique, defect pairs with similar weight changes were grouped into the same level. Through this analysis, a multi-level defect weight matrix was generated, which shows the association strength and change characteristics of each defect pair at different levels.

[0043] After generating the multi-level defect weight matrix, the data is visualized using grid mapping technology. In order to more intuitively display the association relationship in the multimodal data, two-dimensional / three-dimensional rasterization processing and heat map generation technology are used. First, according to the data in the multi-level defect weight matrix, each defect pair and its corresponding association strength value are mapped to a two-dimensional or three-dimensional grid. Each grid cell corresponds to a specific defect pair and its association strength value, and the change in color or brightness is used to reflect the difference in association strength between different defect pairs. For three-dimensional data, a three-dimensional grid is used to display the spatial distribution of multi-level associations. Through these mapping processes, a multimodal defect association weight grid is generated.

[0044] Furthermore, in the method provided in the embodiment of the application, the surface-internal defect association information is generated based on the multimodal defect association weight grid, and further includes: A weight threshold is preset, the multimodal defect association weight grid is screened, defect pairs exceeding the weight threshold are extracted, and a preliminary associated defect pair set is generated; an association path analysis is performed based on the preliminary associated defect pair set to generate a defect pair association path graph; a cluster analysis is performed on the defect nodes of the defect pair association path graph to generate a defect cluster association set, and an association pattern mining is performed to generate a defect pattern association set, and the defect pattern association set is used as the surface-internal defect association information.

[0045] In the embodiment of the present application, the generated multimodal defect association weight grid is first screened by applying a preset weight threshold. The preset weight threshold is pre-set by technical experts as the minimum standard for measuring the strength of defect pair association. Each defect pair in the grid is compared with the threshold, and those defect pairs whose association weight exceeds the threshold are extracted. Through this screening step, defect pairs with low association are filtered out, and only defect pairs with high association are retained, and finally a preliminary set of associated defect pairs is generated.

[0046] Next, we perform correlation path analysis based on the extracted preliminary correlation defect pair set. To understand the deeper correlation between defect pairs, we construct a graph structure in which each node represents a defect pair, and the edges in the graph represent the correlation relationship and path between defect pairs. We use the breadth-first search algorithm to traverse and find the shortest path between all defect pairs in the graph. Breadth-first search can effectively explore the direct and indirect correlation paths between all nodes, and finally generate a defect pair correlation path graph. This path graph shows the network structure between defect pairs and reveals the multi-level correlation between defect pairs.

[0047] After generating the defect pair association path graph, cluster analysis is performed on these defect nodes. In order to accurately divide defect pairs with similar characteristics, the K-means clustering algorithm is used. Specifically, based on the characteristic information such as the association weight and path length of the defect pairs, these data are first input into the K-means clustering algorithm, the number of clusters K is set according to the analysis needs, and then the algorithm begins to assign the defect pairs to K different clusters. The K-means algorithm adjusts the center point position of each cluster through continuous iteration to minimize the distance between the sample points and the center point within the cluster. After several iterations, the clustering results are finally obtained. The defect pairs in each cluster show high similarity in association weights and path characteristics. Through this process, defect cluster association sets are generated. These clusters show structured patterns of defect pairs, which are helpful for further analysis of the correlation of defect pairs.

[0048] After completing the cluster analysis, association pattern mining is performed based on the defect cluster association set. The goal of association pattern mining is to identify frequent association patterns hidden in the data. The Apriori algorithm is used to mine the high-frequency association relationship between defect pairs. The Apriori algorithm gradually expands the frequent item set to find those association pairs that appear frequently in different clusters. Specifically, the association relationship of defect pairs is analyzed, and defect pairs with repeated association characteristics are gradually discovered, and a frequent item set is generated based on their frequency of occurrence. Through this mining process, the association rules between specific defect pairs are identified, and a defect pattern association set is generated, which contains the key association patterns between different defect pairs.

[0049] Finally, the generated defect pattern association set is used as the surface-internal defect association information.

[0050] Step S400: performing supervised training based on the stainless steel surface defect set, the stainless steel internal defect set and surface-internal defect association information to obtain a stainless steel quality detection model.

[0051] In the embodiment of the present application, the acquired stainless steel surface defect set, stainless steel internal defect set and surface-internal defect association information are first integrated into a supervised learning training data set. In this process, the stainless steel surface defect set and the stainless steel internal defect set provide the stainless steel surface and internal defect data as input feature sets, and the surface-internal defect association information is used as a label set for the training labels in the supervised learning model to guide the model to identify the association pattern between surface and internal defects.

[0052] After the data is prepared, it is trained using a supervised learning algorithm, such as a decision tree, support vector machine, logistic regression, or deep neural network. The role of these models is to automatically establish a reasoning mechanism by learning the relationship between the features in the input data (surface defects and internal defect data) and the output target (associated information). During the training phase, the model gradually captures the complex associations between surface defects and internal defects by continuously adjusting internal parameters. During the model training process, the batch gradient descent algorithm is used as the optimization mechanism. Each time the model is trained, the model weights are adjusted according to the input surface and internal defect features to minimize the error between the model's predicted output and the actual label (surface-internal defect associated information). After each training batch, the model continuously updates the weights and biases to gradually optimize its performance, enabling it to more accurately map the complex relationship between the input defect features and associated information.

[0053] Through this supervised training process, a stainless steel quality inspection model is finally generated.

[0054] Step S500: obtaining a set of surface defects of stainless steel to be inspected and a set of internal defects of stainless steel to be inspected, and inputting them into the stainless steel quality inspection model to complete quality inspection.

[0055] In the embodiment of the present application, a high-resolution camera is first used to perform a comprehensive surface scan of the stainless steel to be inspected to obtain image data of the stainless steel to be inspected. At the same time, non-destructive testing technology is used to obtain internal structure data of the stainless steel to be inspected.

[0056] Afterwards, the collected stainless steel surface image data to be detected is input into the constructed and trained stainless steel surface defect recognition model to obtain the stainless steel surface defect set to be detected. At the same time, the collected stainless steel internal structure data to be detected is input into the trained stainless steel internal defect recognition model to obtain the stainless steel internal defect set to be detected.

[0057] Finally, the surface defect set and internal defect set of the stainless steel to be tested are input into the stainless steel quality inspection model for detection. By analyzing the correlation between the surface and internal defects, the quality analysis of the stainless steel to be tested is completed, and the specific defects of the stainless steel to be tested are output, reducing the missed detection caused by unilateral detection.

[0058] Through this process, the accuracy of quality inspection of stainless steel to be inspected is improved.

[0059] In the embodiments of the present application, in summary, the embodiments of the present application have at least the following technical effects: The present application obtains a stainless steel detection historical data set, which includes stainless steel image data and stainless steel internal structure data, wherein each stainless steel image data has corresponding stainless steel internal structure data; performs surface defect recognition based on the stainless steel image data to obtain a stainless steel surface defect set, and at the same time, performs internal defect recognition based on the stainless steel internal structure data to obtain a stainless steel internal defect set; performs correlation analysis based on the stainless steel surface defect set and the stainless steel internal defect set to obtain surface-internal defect correlation information; performs supervised training based on the stainless steel surface defect set, the stainless steel internal defect set and the surface-internal defect correlation information to obtain a stainless steel quality detection model; obtains a stainless steel surface defect set to be detected and a stainless steel internal defect set to be detected, and inputs them into the stainless steel quality detection model to complete quality detection. The present invention solves the technical problem that the prior art has insufficient detection accuracy when performing stainless steel quality detection. Through big data analysis and multimodal detection technology, the historical detection data of stainless steel, surface and internal defect recognition, and defect correlation analysis are used to construct a supervised learning model to achieve comprehensive detection of stainless steel quality, thereby achieving the technical effect of improving the accuracy of stainless steel quality detection.

[0060] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0061] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0062] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A precision stainless steel quality detection method based on big data analysis, characterized in that: include: Acquire a stainless steel detection history data set, wherein the stainless steel detection history data set includes stainless steel image data and stainless steel internal structure data, wherein each stainless steel image data has corresponding stainless steel internal structure data; Performing surface defect recognition based on the stainless steel image data to obtain a stainless steel surface defect set, and performing internal defect recognition based on the stainless steel internal structure data to obtain a stainless steel internal defect set; Perform correlation analysis based on the stainless steel surface defect set and the stainless steel internal defect set to obtain surface-internal defect correlation information; Perform supervised training based on the stainless steel surface defect set, the stainless steel internal defect set and surface-internal defect correlation information to obtain a stainless steel quality detection model; Obtain a set of surface defects of the stainless steel to be inspected and a set of internal defects of the stainless steel to be inspected, and input them into the stainless steel quality inspection model to complete the quality inspection.

2. The precision stainless steel quality detection method based on big data analysis according to claim 1, characterized in that: Surface defect recognition is performed based on the stainless steel image data to obtain a stainless steel surface defect set, including: Acquire a sample stainless steel image data set, wherein the sample stainless steel image data set includes surface defect annotation marks; Building a stainless steel surface defect recognition model based on the sample stainless steel image data set; The stainless steel surface defect recognition model is used to perform surface defect recognition on the stainless steel image data to obtain a stainless steel surface defect set.

3. The precision stainless steel quality detection method based on big data analysis according to claim 1 is characterized in that: Based on the internal structure data of the stainless steel, internal defects are identified to obtain a set of internal defects of the stainless steel, including: Acquire a sample stainless steel internal structure data set, wherein the sample stainless steel internal structure data set includes internal defect marking marks; Building a stainless steel internal defect recognition model based on the sample stainless steel internal structure data set; The stainless steel internal defect recognition model is used to perform internal defect recognition on the stainless steel internal structure data to obtain a stainless steel internal defect set.

4. The precision stainless steel quality detection method based on big data analysis according to claim 1, characterized in that: Based on the stainless steel surface defect set and the stainless steel internal defect set, correlation analysis is performed to obtain surface-internal defect correlation information, including: Coordinate mapping is performed on the stainless steel surface defect set and the stainless steel internal defect set to obtain a stainless steel multimodal space fusion set; Performing feature extraction on the stainless steel multimodal spatial fusion set to obtain a stainless steel surface feature data set and a stainless steel internal feature data set; Perform cross analysis based on the stainless steel surface feature data set and the stainless steel internal feature data set to generate a defect correlation metric matrix; In combination with the defect association metric matrix, the association strength between surface defects and internal defects is evaluated to generate a multimodal defect association weight grid; Based on the multimodal defect association weight grid, surface-internal defect association information is generated.

5. The precision stainless steel quality detection method based on big data analysis according to claim 4 is characterized in that: Based on the stainless steel surface feature data set and the stainless steel internal feature data set, a cross analysis is performed to generate a defect correlation metric matrix, including: Performing multi-dimensional feature mapping on the stainless steel surface feature data set and the stainless steel internal feature data set to generate a stainless steel feature vector mapping set; Perform pairing based on the stainless steel feature vector mapping set to obtain a surface-internal feature pairing set; Based on the surface-internal feature pairing set, correlation measurement is performed to generate a feature correlation measurement set, and the defect correlation measurement matrix is ​​generated through matrix construction.

6. The precision stainless steel quality detection method based on big data analysis according to claim 5, characterized in that: Pairing is performed based on the stainless steel feature vector mapping set to obtain a surface-internal feature pairing set, including: Based on the stainless steel feature vector mapping set, generating a morphologically similar feature pair set through morphological similarity analysis; Generate a scale matching feature pair set by calculating the scale similarity of the morphologically similar feature pair set; Based on the scale matching feature pair set, the spatial distribution pattern of the features is matched and screened to obtain the surface-interior feature pair set.

7. The precision stainless steel quality detection method based on big data analysis according to claim 4 is characterized in that: Combined with the defect association metric matrix, the association strength between surface defects and internal defects is evaluated to generate a multimodal defect association weight grid, including: Combined with the defect association metric matrix, disturbance calculation is performed through topological disturbance analysis, and the association change value of each defect pair is counted to generate a disturbance defect association weight matrix; Perform multi-level grading based on the disturbance defect association weight matrix to generate a multi-level defect weight matrix; Based on the multi-level defect weight matrix, the multi-modal defect association weight grid is generated through grid mapping.

8. The method for precision stainless steel quality detection based on big data analysis according to claim 4, characterized in that: Based on the multimodal defect association weight grid, surface-internal defect association information is generated, including: Preset a weight threshold, screen the multimodal defect association weight grid, extract defect pairs exceeding the weight threshold, and generate a preliminary association defect pair set; Performing correlation path analysis based on the preliminary correlation defect pair set to generate a defect pair correlation path diagram; A cluster analysis is performed on the defect nodes of the defect pair association path graph to generate a defect cluster association set, and an association pattern mining is performed to generate a defect pattern association set, and the defect pattern association set is used as the surface-internal defect association information.

Citation Information

Cited By

  • Training method and system of diffusion plate defect detection model and medium

    CN119904455A

  • Defect detection method combining vision and X-ray detection technology

    CN120598968A

  • Nuclear power sealing gasket production quality detection method based on machine vision

    CN121767367A