Intelligent detection method for product defects on automatic production line and detection system based on machine vision

Abstract

The invention discloses an intelligent detection method for product defects on an automatic production line and a detection system based on machine vision, and relates to the technical field of industrial product quality detection. According to the method, a multi-waveband imaging technology is combined with temperature and chemical component information to generate a multi-dimensional feature data set, and an unsupervised learning algorithm is utilized to perform clustering analysis, so that the limitation of traditional single-waveband imaging is broken through, product features are comprehensively captured, the defect detection range and accuracy are remarkably improved, and a foundation is laid for subsequent analysis and classification; further utilizing an attention mechanism and a deep learning technology to accurately position and classify defects, and triggering deep scanning through a priority index to improve the detection precision and the system adaptability; and finally, combining with a random forest algorithm to analyze defect influence, and feeding back and optimizing production parameters through a dynamic adjustment mechanism, thereby realizing production optimization closed-loop management, and improving production efficiency and product quality detection.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial product quality inspection, and in particular to an intelligent product defect inspection method on an automated production line and a machine vision-based inspection system. Background Art

[0002] Product quality inspection in the manufacturing industry is a key area for promoting industrial intelligence and efficiency improvement. Its importance is self-evident and directly related to the stable operation of the production line and the market competitiveness of products. With the widespread application of automated production lines, how to achieve rapid and accurate defect identification has become a core requirement for the development of the industry. However, existing inspection methods mostly rely on traditional image processing technology or single-band imaging, which generally suffer from the large demand for defect samples, limited inspection range, and insufficient ability to identify complex defects. These limitations make it difficult for current solutions to adapt to the real dilemma of the manufacturing industry with a wide variety of products, diverse defect forms, and scarce annotated data. In particular, when it comes to internal damage or non-surface defects, traditional methods often seem inadequate.

[0003] Against this backdrop, the core challenges facing this field have gradually become prominent. First, the scarcity of defect samples makes it difficult to implement traditional machine learning methods that rely on large-scale labeled data. How to achieve efficient defect recognition based on positive sample modeling has become a bottleneck for technological breakthroughs. Second, existing visual inspection systems lack the ability to integrate multi-dimensional information and are limited to single visible light imaging. They are unable to fully capture key features such as product temperature, chemical composition or internal structure, resulting in the simplification of detection dimensions. Finally, the priority focusing mechanism for defective areas has not yet been perfected, and the system finds it difficult to locate potential problem areas as quickly as human vision, which further exacerbates the contradiction between real-time performance and accuracy. These unresolved technical factors directly lead to the insufficient adaptability of the inspection system in complex scenarios, limiting its application potential in automated production lines. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent detection method for product defects on an automated production line and a detection system based on machine vision.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] This application provides a method for intelligently detecting product defects on an automated production line, comprising the following steps:

[0007] Acquire product positive sample data, collect visible light, infrared light, and ultraviolet light images through multi-band imaging technology, and integrate temperature information capture and chemical composition identification to obtain an initial data set containing multi-dimensional features;

[0008] Based on the initial data set, a positive sample modeling method is used to cluster the multi-dimensional features using a pre-established unsupervised learning algorithm to determine the distribution range of the product's normal state characteristics. Abnormal areas are then extracted from the multi-band imaging data and preliminarily classified using a convolutional neural network to obtain the location results of potential defect areas.

[0009] Based on the positioning results, the attention mechanism is applied to perform weighted processing on the potential defect areas, and the internal structure and surface morphology are analyzed through multi-dimensional feature detection to determine the defect type and severity;

[0010] Obtain weighted regional data, integrate the results of temperature information capture and chemical composition identification, use the support vector machine algorithm to perform secondary verification of complex defect identification, and determine the final defect classification;

[0011] Through the final defect classification, the priority index of the defect area focus is calculated. When the priority index exceeds the preset threshold, the high-resolution imaging module is triggered to perform a deep scan of the area to obtain detailed defect features;

[0012] Based on the depth scan data and combined with the existing information of multi-band imaging, the random forest algorithm is used to conduct a multi-dimensional integrated analysis of the defect characteristics to determine the potential impact of the defects on the automated production line. Key indicators are extracted from the potential impact. Through a real-time dynamic adjustment mechanism of contradictions, the detection results are fed back to the production line control system to obtain an optimized production parameter adjustment plan.

[0013] Furthermore, an initial dataset containing multi-dimensional features is obtained, including:

[0014] The visible light image, infrared light image and ultraviolet light image of the positive sample data are collected by multi-band imaging technology to obtain a first image set with multi-dimensional features, and then the temperature information is superimposed by fusion processing technology to obtain a second image set containing temperature features;

[0015] Extract chemical composition data from the second image set, determine the chemical composition distribution using recognition technology, and obtain a third image set with component annotations. Based on the multidimensional features of the third image set, determine whether the difference between the features exceeds a preset threshold. If so, determine the main features using a principal component analysis algorithm to obtain a fourth image set after dimensionality reduction.

[0016] According to the feature distribution of the fourth image set, a clustering algorithm is used to judge the similarity between samples to obtain the classified fifth image set. According to the classification results of the fifth image set, the temperature information and chemical composition distribution are integrated to obtain the final initial data set of multidimensional features.

[0017] Furthermore, the characteristic distribution range of the product in normal state is determined, specifically including:

[0018] Obtain multidimensional features through the initial data set, use unsupervised clustering algorithm to analyze the multidimensional features and obtain feature distribution;

[0019] Use statistical tools to determine the distribution range of feature distribution and judge the boundary of the normal state of the product. When the feature distribution exceeds the distribution range, adjust the feature analysis through positive sample modeling to determine abnormal features;

[0020] According to the abnormal characteristics, relevant subsets are extracted from the initial data set to obtain the subset feature distribution. Then, an unsupervised clustering algorithm is used to perform a secondary analysis on the subset feature distribution to determine the normal state of the subset.

[0021] By comparing the normal state of the subset with the normal state of the product, the consistency is judged and the state deviation is obtained. Then, based on the state deviation, the data modeling parameters are adjusted to determine the optimized feature distribution range.

[0022] Furthermore, after determining the characteristic distribution range of the product's normal state, it also includes: acquiring imaging data through multi-band data, using statistical methods to determine whether the characteristic value exceeds the normal distribution range to obtain an abnormal area, then extracting characteristic values from the abnormal area, classifying the characteristic values through a convolutional neural network to obtain a preliminary classification result, analyzing potential defects based on the preliminary classification result, using regional analysis technology to determine the defect boundary to obtain a positioning result, then comparing the abnormal area with the imaging data to determine whether the defect is consistent with the characteristics of the multi-band data to obtain a verification result, extracting the change trend of the abnormal area from the verification result, analyzing the change trend through a convolutional neural network to obtain the defect evolution characteristics; based on the defect evolution characteristics, using the positioning result to update the regional analysis, determine the expansion range of the abnormal area, and obtain the final positioning, obtaining the dynamic changes of the multi-band data through the final positioning, using statistical methods to analyze the characteristic value distribution, and determining the defect classification accuracy.

[0023] Furthermore, the defect type and severity are determined, including:

[0024] The potential defect area is determined by the positioning results, and the attention mechanism is used to perform weighted processing on the area to obtain weighted feature data. Multidimensional features are obtained from the weighted feature data, and then the multidimensional features are preliminarily classified to determine whether there are significant abnormalities.

[0025] When significant anomalies exist, convolutional neural networks are used to analyze multidimensional features and detect the changing trends of the internal structure. Based on these trends, clustering algorithms are used to group potential defects and determine the distribution range of defect types.

[0026] By analyzing the characteristic data of surface morphology based on the distribution range, the severity level of the defect is determined. After obtaining the severity level, the final defect classification result is obtained by combining it with the internal structure information. Key features are extracted from the final defect classification result to determine the evolution direction of potential defects.

[0027] Furthermore, the final defect classification is determined, including:

[0028] Obtain weighted regional data, then collect temperature information and chemical composition data through sensors, fuse them into the weighted regional data set to obtain a comprehensive feature set, and use a preset feature extraction method to separate complex defect-related features from the comprehensive feature set to obtain a defect feature set;

[0029] The defect feature set is initially classified using the support vector machine algorithm to obtain a preliminary defect category. When the confidence level of the preliminary defect category is lower than the preset threshold, the defect feature set is re-analyzed through a secondary verification process to obtain an adjusted defect category.

[0030] The final defect classification result is determined based on the matching degree between the adjusted defect category and the comprehensive feature set. The accuracy of the final defect classification result is verified by comparing with historical data to obtain the optimized classification output.

[0031] Furthermore, detailed defect characteristics are obtained, including:

[0032] The priority index is calculated based on the defect classification data. The classification results are processed using the support vector machine algorithm to obtain the priority index of the defect area. When the priority index exceeds the preset threshold, the trigger condition is determined to be met based on the calculation result, and the defect area that requires deep scanning is obtained;

[0033] Divide the scanning area according to the defect area, use the imaging module to generate a high-resolution image, obtain preliminary scanning data, process the preliminary scanning data through deep scanning, use the convolutional neural network algorithm to extract feature details, and obtain the first feature set;

[0034] Analyze defect features based on the first feature set, determine whether the feature details are complete, and obtain a second feature set. Then, update the defect classification data based on the second feature set, use statistical tools to verify the classification consistency, and obtain an optimized defect classification.

[0035] Recalculate the priority index based on the optimized defect classification to determine whether to trigger a new round of deep scanning and obtain the final feature details.

[0036] Furthermore, the potential impact of defects on automated production lines is determined by: acquiring data from the production line through deep scanning, extracting raw information using multi-band imaging technology to obtain preliminary defect features, and then obtaining multi-dimensional data and integrating and analyzing it using the random forest algorithm to determine a set of feature vectors;

[0037] Extract key information from the feature vector set, classify defect types through information processing technology, determine the defect distribution pattern, obtain production line operation data based on the defect distribution pattern, and use the support vector machine algorithm to analyze the potential impact of defects on automated production to obtain an estimated impact range;

[0038] The estimated impact range is logically judged using a preset threshold. When the threshold is exceeded, information processing technology is used to generate adjustment parameters and determine an optimization solution.

[0039] Adjustment parameters are extracted from the optimization plan, real-time production line data is obtained, and the impact of the adjustment is determined through comparative analysis. Based on the impact of the adjustment, information processing technology is used to verify the results and obtain the final judgment result.

[0040] Furthermore, after determining the potential impact of defects on the automated production line, the method also includes: extracting key indicators through impact analysis to obtain an indicator set, comparing the indicator set with the test results, and triggering real-time contradiction judgment when the difference exceeds a preset threshold to obtain a contradictory state, driving a dynamic adjustment mechanism based on the contradictory state to obtain an adjustment direction; updating the feedback mechanism through the adjustment direction to determine the feedback signal, transmitting the feedback signal to the control system, determining the optimal parameter combination, adjusting the production parameters based on the optimized parameter combination, and obtaining a parameter solution.

[0041] This application provides a system for intelligently detecting product defects on an automated production line based on machine vision, which is used to implement an intelligent method for detecting product defects on an automated production line, including the following steps:

[0042] A multi-band imaging module is used to collect image data of products under visible light, infrared light, and ultraviolet light, and integrate temperature information capture and chemical composition identification to generate an initial data set containing multi-dimensional features;

[0043] The data processing module is used to analyze the initial data set, perform cluster analysis using an unsupervised learning algorithm, determine the characteristic distribution range of the product's normal state, and extract abnormal areas. It is also used to perform preliminary convolutional neural network classification on abnormal areas to obtain the location results of potential defect areas;

[0044] The attention mechanism module is used to perform weighted processing on potential defect areas, analyze the internal structure and surface morphology through multi-dimensional feature detection, and determine the defect type and severity;

[0045] Complex defect verification module, used to perform secondary verification of the support vector machine algorithm on the weighted regional data to determine the final defect classification;

[0046] High-resolution imaging module, used to perform a deep scan of the defect area to obtain detailed defect features when the priority index of the defect area exceeds a preset threshold;

[0047] The defect analysis module combines multi-band imaging data and depth scanning data to perform multi-dimensional integrated analysis using a random forest algorithm to determine the potential impact of defects on the production line and extract key indicators.

[0048] The feedback control module is used to feed back the detection results to the production line control system through a dynamic adjustment mechanism of real-time contradictions, and generate an optimized production parameter adjustment plan.

[0049] The beneficial effects of the present invention are:

[0050] This invention uses multi-band imaging technology combined with temperature and chemical composition information to generate a multidimensional feature data set. It then uses an unsupervised learning algorithm for cluster analysis to determine the distribution range of normal features, thereby extracting abnormal areas and performing preliminary classification. This process breaks through the limitations of traditional single-band imaging and can comprehensively capture product features in different dimensions, including internal structures and non-surface defects. This significantly improves the scope and accuracy of defect detection, providing a solid foundation for subsequent defect analysis and classification.

[0051] The present invention uses an attention mechanism to perform weighted processing on potential defect areas, combines multi-dimensional feature detection to analyze internal structure and surface morphology, determines defect type and severity, and further uses a support vector machine algorithm to perform secondary verification on complex defects to determine the final defect classification. In addition, by calculating the priority index of the defect area, when the priority index exceeds a preset threshold, the high-resolution imaging module is triggered to perform a deep scan of the area to obtain detailed defect features. This can not only accurately locate the defect area, but also carefully classify the defects. This effectively solves the problem of the existing visual inspection system's insufficient ability to identify complex defects, further improves the accuracy and reliability of detection, and enhances the adaptability of the inspection system in complex scenarios.

[0052] Through the final defect classification, the priority index of the defect area is calculated, and the high-resolution imaging module is triggered to perform a deep scan to obtain detailed defect characteristics. Then, combined with the existing information of multi-band imaging, the random forest algorithm is used to perform a multi-dimensional integrated analysis of the defect characteristics to determine the potential impact of the defect on the automated production line, and extract key indicators. Through a real-time dynamic adjustment mechanism of contradictions, the detection results are fed back to the production line control system to generate an optimized production parameter adjustment plan, realizing closed-loop management from defect detection to production optimization. It can predict the potential risks of defects to the production process in advance, and generate optimization plans based on this, and adjust production parameters in time, thereby reducing production interruptions and defective rates caused by defects, and improving production efficiency and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] For better understanding and implementation, the technical solution of the present application is described in detail below with reference to the accompanying drawings.

[0054] Figure 1 A schematic diagram of the process of the intelligent detection method for product defects on an automated production line provided in Example 1 of the present application;

[0055] Figure 2 A schematic diagram of a process for obtaining an initial data set containing multi-dimensional features using the intelligent product defect detection method on an automated production line provided in Example 1 of the present application;

[0056] Figure 3 Schematic diagram of the process of determining the characteristic distribution range of the normal state of the product in the method for intelligent detection of product defects on the automated production line provided in Example 1 of the present application

[0057] Figure 4 A schematic diagram of the structure of the intelligent product defect detection system on an automated production line based on machine vision provided in Example 2 of the present application. DETAILED DESCRIPTION

[0058] To further illustrate the technical means and effectiveness of the present invention in achieving its intended purpose, exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present application. Rather, they are merely examples of methods and systems consistent with certain aspects of the present application, as detailed in the appended claims.

[0059] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this application and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0060] The following describes in detail the specific implementation methods, features and effects of the present invention in conjunction with the accompanying drawings and preferred embodiments.

[0061] Example 1

[0062] See also Figure 1-Figure 3 This embodiment provides a method for intelligently detecting product defects on an automated production line, comprising the following steps:

[0063] S1. Obtain product positive sample data, collect visible light, infrared light, and ultraviolet light images through multi-band imaging technology, and integrate temperature information capture and chemical composition identification to obtain an initial data set containing multi-dimensional features;

[0064] Furthermore, in step S1, an initial data set containing multi-dimensional features is obtained, including:

[0065] S11, using a multi-band imaging technique to collect visible light images, infrared images, and ultraviolet images of positive sample data to obtain a first image set with multi-dimensional features, and then using a fusion processing technique to superimpose temperature information to obtain a second image set containing temperature features;

[0066] S12. Extract chemical composition data from the second image set, determine the chemical composition distribution using recognition technology, and obtain a third image set labeled with the components. Based on the multidimensional features of the third image set, determine whether the difference between the features exceeds a preset threshold. If so, determine the main features using a principal component analysis algorithm to obtain a fourth image set after dimensionality reduction.

[0067] S13. Based on the feature distribution of the fourth image set, a clustering algorithm is used to determine the similarity between samples to obtain a classified fifth image set. Based on the classification results of the fifth image set, the temperature information and the chemical composition distribution are integrated to obtain a final initial data set of multidimensional features.

[0068] Specifically, the visible light images, infrared images and ultraviolet images of positive sample data collected through multi-band imaging technology can be understood as using the optical properties of different bands to capture the multi-dimensional information of the sample. For example, when testing the quality of a certain chemical material, the visible light image can reflect the surface color and texture, the infrared image can reveal the thermal distribution characteristics, and the ultraviolet image may expose the fluorescence reaction characteristics. Suppose in an experiment, an image of a metal sample is collected, visible light shows that its surface is grayish white, infrared light detects a local high temperature area of 80°C, and ultraviolet light captures weak fluorescence, indicating the possible presence of oxides. These images together constitute the first image set, providing the basis for multi-dimensional features.

[0069] For the first image set, a fusion processing technique is used to superimpose temperature information to generate a second image set. Specifically, the thermal data of the infrared image can be used to map temperature values to corresponding areas in the visible and ultraviolet images. For example, the high-temperature area of a metal sample is labeled as 80°C, and the low-temperature area is labeled as 25°C. After fusion, each pixel not only has color information but also a temperature value. This fusion enhances the physical meaning of the data and facilitates subsequent analysis of the relationship between thermal distribution and material properties.

[0070] Chemical composition data is extracted from the second image set. Preferably, the material characteristics can be identified through spectral analysis technology. For example, the fluorescence reaction of the ultraviolet image is combined with the thermal characteristics of the infrared data to infer that iron oxide may exist in the high-temperature area and the low-temperature area may be in a pure metallic state. After the component distribution is marked, the third image set is obtained. For example, if the high-temperature area is marked as Fe2O3 and the low-temperature area is marked as Fe, then the multidimensional features of the third image set include color, temperature and chemical composition.

[0071] After obtaining the multidimensional features of the third image set, if the difference between the features exceeds the preset threshold, for example, the color value difference exceeds 50 and the temperature difference exceeds 30°C, the principal component analysis algorithm is used for dimensionality reduction. In one possible implementation, the analysis results show that temperature and composition are the main features, and color has little influence. Finally, the fourth image set is obtained. This dimensionality reduction reduces redundant information and improves computational efficiency. According to the feature distribution of the fourth image set, a clustering algorithm is used to judge the similarity between samples. For example, K-means clustering can divide metal samples into high-temperature oxidation group and low-temperature pure metal group to obtain the fifth image set. This classification helps to quickly distinguish the sample status and provide clear grouping for subsequent modeling. Through the classification results of the fifth image set, the temperature information and chemical composition distribution are integrated to obtain the final data set. Specifically, the high-temperature oxidation group sample may show 80°C and be rich in Fe2O3, and the low-temperature group is 25°C and Fe. This fusion strengthens the integrity of the data and reflects the intrinsic connection between multidimensional features.

[0072] For the final data set, the support vector machine algorithm is used to determine the attributes of the positive samples. For example, the positive samples are set as pure metals without oxidation. The support vector machine classifies the low-temperature group as positive samples based on temperature and composition characteristics, and the output results are clearly labeled. This method improves the classification accuracy and provides a reliable basis for quality inspection. It should be noted that the combination of multi-band imaging and fusion technology can significantly improve the depth and accuracy of sample analysis. For example, compared with single visible light detection, multi-dimensional features reveal hidden oxidation problems and help optimize the material production process. In one embodiment, this method can also be extended to real-time monitoring to detect abnormal samples in a timely manner and improve the practicality of industrial applications.

[0073] S2. Based on the initial data set, a positive sample modeling method is used to perform cluster analysis on multidimensional features using a pre-established unsupervised learning algorithm to determine the characteristic distribution range of the product's normal state. Abnormal areas are then extracted from the multi-band imaging data and preliminarily classified using a convolutional neural network to obtain the location results of potential defect areas.

[0074] Furthermore, in step S2, the characteristic distribution range of the product in a normal state is determined, specifically including:

[0075] S21. Obtain multidimensional features through the initial data set, analyze the multidimensional features using an unsupervised clustering algorithm, and obtain feature distribution;

[0076] S22. Use statistical tools to determine the distribution range of feature distribution and determine the boundary of the normal state of the product. When the feature distribution exceeds the distribution range, adjust the feature analysis through positive sample modeling to determine abnormal features;

[0077] S23. Extract relevant subsets from the initial data set based on the abnormal features to obtain subset feature distribution, and then use an unsupervised clustering algorithm to perform a secondary analysis on the subset feature distribution to determine the normal state of the subset;

[0078] S24. Compare the normal state of the subset with the normal state of the product to determine consistency and obtain state deviation. Then, adjust the data modeling parameters based on the state deviation to determine the optimized feature distribution range.

[0079] Specifically, in using the initial data set to determine the distribution range of product normal state characteristics and locate potential defect areas, the principle is to first analyze the multidimensional features in the initial data set through an unsupervised clustering algorithm to obtain the characteristic distribution. This is based on the data's own structural mining rules and does not require pre-labeling of data categories. Statistical tools are then used to determine the distribution range to define the product's normal state boundary. Once the characteristic distribution is found to be outside the range, positive sample modeling is used to adjust the characteristic analysis to determine abnormal characteristics, which can accurately distinguish between normal and abnormal situations. Then, based on the abnormal characteristics, relevant subsets are extracted from the initial data set and unsupervised cluster analysis is performed again to determine the normal state of the subset, thereby further refining the processing of abnormal data. Finally, the normal state of the subset is compared with the overall normal state of the product to obtain the state deviation and adjust the data modeling parameters accordingly to obtain the optimized characteristic distribution range. Through this series of operations, the distribution range of the product's normal state characteristics is accurately defined, and abnormal areas can be effectively extracted from multi-band imaging data. Then, with the help of convolutional neural network preliminary classification, the potential defect area is successfully accurately located, greatly improving the accuracy and reliability of product defect detection on the automated production line, and providing a solid foundation for subsequent defect type judgment and processing.

[0080] Furthermore, in step S2, after determining the characteristic distribution range of the normal state of the product, it also includes: acquiring imaging data through multi-band data, using statistical methods to determine whether the characteristic value exceeds the normal distribution range to obtain an abnormal area, then extracting characteristic values from the abnormal area, classifying the characteristic values through a convolutional neural network to obtain a preliminary classification result, analyzing potential defects based on the preliminary classification result, using regional analysis technology to determine the defect boundary to obtain a positioning result, then comparing the abnormal area with the imaging data to determine whether the defect is consistent with the characteristics of the multi-band data to obtain a verification result, extracting the change trend of the abnormal area from the verification result, analyzing the change trend through a convolutional neural network to obtain the defect evolution characteristics; based on the defect evolution characteristics, using the positioning result to update the regional analysis, determine the expansion range of the abnormal area, obtain the final positioning, obtain the dynamic changes of the multi-band data through the final positioning, use statistical methods to analyze the characteristic value distribution, and determine the defect classification accuracy.

[0081] Specifically, the system accurately analyzes and locates abnormal areas of the product and tracks the evolution of defects, thereby improving the comprehensiveness and accuracy of defect detection. Through multiple rounds of analysis and verification, the exact location of the defect is determined; the potential defect type is effectively identified, and a preliminary judgment is provided using convolutional neural network classification; the consistency of defects and multi-band data characteristics is verified to ensure the reliability of detection results; the defect evolution characteristics are mastered to provide a basis for predicting defect development; the defect classification accuracy is improved by analyzing the dynamic changes of multi-band data, and ultimately provides strong support for adjusting production parameters of automated production lines, ensuring product quality, and preventing defect deterioration.

[0082] S3. Based on the positioning results, the attention mechanism is applied to perform weighted processing on the potential defect areas. The internal structure and surface morphology are analyzed through multi-dimensional feature detection to determine the defect type and severity.

[0083] Furthermore, in step S3, the defect type and severity are determined, specifically including:

[0084] The potential defect area is determined by the positioning results, and the attention mechanism is used to perform weighted processing on the area to obtain weighted feature data. Multidimensional features are obtained from the weighted feature data, and then the multidimensional features are preliminarily classified to determine whether there are significant abnormalities.

[0085] When significant anomalies exist, convolutional neural networks are used to analyze multidimensional features and detect the changing trends of the internal structure. Based on these trends, clustering algorithms are used to group potential defects and determine the distribution range of defect types.

[0086] By analyzing the characteristic data of surface morphology based on the distribution range, the severity level of the defect is determined. After obtaining the severity level, the final defect classification result is obtained by combining it with the internal structure information. Key features are extracted from the final defect classification result to determine the evolution direction of potential defects.

[0087] Among them, the potential defect area is determined by the positioning result, and the area is weighted by using the attention mechanism to obtain weighted feature data, which also includes: obtaining the positioning coordinate information of the potential defect area, extracting the multidimensional feature data of the corresponding area according to the positioning coordinate information, and the multidimensional feature data includes structural texture features and surface morphology features; using the attention weight calculation module to dynamically weight the multidimensional feature data to obtain a weighted fusion feature vector; inputting the fusion feature vector into a preset defect classification network, and the defect classification network outputs a defect type label and a corresponding confidence score; comparing the confidence score with a preset severity threshold, if the confidence score exceeds the severity threshold, generating a high-priority defect report; matching the high-priority defect report with a historical defect database, and the historical defect database contains a mapping relationship between defect types and repair solutions; outputting a defect analysis result including repair suggestions, and the defect analysis result is associated with the positioning coordinate information and the multidimensional feature data.

[0088] Specifically, for example, the attention mechanism can be understood as a resource allocation method, allocating more computing resources to key areas in the imaging data. For example, in a multi-band imaging image, if the eigenvalue of a certain area fluctuates greatly, the attention mechanism will automatically increase the weight of the area and generate weighted feature data, such as increasing the original eigenvalue from 0.5 to 0.8. This weighted processing helps to highlight the details of the abnormal area and facilitates subsequent analysis. In a possible implementation method, multidimensional features may include color depth, texture roughness, and edge sharpness. For example, the texture roughness of a normal area may be stable at around 0.2, while the abnormal area may reach 0.6. By setting a threshold, such as 0.4, preliminary classification can quickly screen out significant abnormal areas. The advantage of this method is that it can efficiently lock the problem area and reduce the subsequent computational burden.

[0089] Convolutional neural networks are used to analyze multi-dimensional features and detect changing trends in internal structures. Specifically, convolutional neural networks can capture changes in features over time or space. For example, in continuous multi-band imaging data, if the edge sharpness of a defect area gradually decreases from 0.7 to 0.3, it suggests deterioration of the internal structure. Such trend analysis helps predict the potential development direction of defects. In one embodiment, a clustering algorithm can divide feature data into three groups based on similarity: surface cracks, internal voids, and material foreign matter. For example, the characteristics of surface cracks may be concentrated in high texture roughness, while internal voids may appear as low color depth. Such grouping can clearly show the distribution of defect types, facilitating further analysis.

[0090] By analyzing the characteristic data of the surface morphology through the distribution range, the severity level of the defect can be determined. Preferably, the evaluation can be based on the smoothness of the surface morphology. For example, an area with a smoothness below 0.1 may be rated as high severity, while an area above 0.5 is rated as low severity. This grading method is intuitive and easy to understand, and can provide a basis for subsequent decision-making. After obtaining the severity level, the final defect classification result is obtained by combining it with the internal structure information. It should be noted that the internal structure information can be extracted from the deep features of the multi-band data. For example, if the surface smoothness of a certain area is 0.4, but the internal void characteristic value is as high as 0.7, it may eventually be classified as a "hidden serious defect." This comprehensive analysis improves the accuracy of the classification.

[0091] Key features are extracted from the final defect classification results to determine the evolution direction of potential defects. It is understandable that a continuous decrease in key features such as edge sharpness may indicate that the defect will expand to a larger area. For example, if the edge sharpness of a certain area drops from 0.6 to 0.2, combined with the increasing trend of texture roughness, it can be inferred that the defect is worsening. This judgment provides an important reference for dynamic monitoring and enhances predictive capabilities.

[0092] By applying the attention mechanism to weighted processing of potential defect areas, we can focus on the key features of the defects. By combining convolutional neural networks, clustering algorithms, etc. to analyze multi-dimensional features, we can accurately determine the type and severity of defects and predict the evolution direction of potential defects. At the same time, by inputting the fused feature vector into the defect classification network, we can generate high-priority defect reports and match them with the historical database, and output defect analysis results containing repair suggestions that associate positioning coordinates and multi-dimensional feature data, providing all-round support for automated production lines to quickly repair defects, improve product quality control levels and production efficiency.

[0093] S4. Obtain weighted regional data, integrate the results of temperature information capture and chemical composition identification, use the support vector machine algorithm to perform secondary verification on complex defect identification, and determine the final defect classification;

[0094] Furthermore, in step S4, the final defect classification is determined, specifically including:

[0095] Obtain weighted regional data, then collect temperature information and chemical composition data through sensors, fuse them into the weighted regional data set to obtain a comprehensive feature set, and use a preset feature extraction method to separate complex defect-related features from the comprehensive feature set to obtain a defect feature set;

[0096] The defect feature set is initially classified using the support vector machine algorithm to obtain a preliminary defect category. When the confidence level of the preliminary defect category is lower than the preset threshold, the defect feature set is re-analyzed through a secondary verification process to obtain an adjusted defect category.

[0097] The final defect classification result is determined based on the matching degree between the adjusted defect category and the comprehensive feature set. The accuracy of the final defect classification result is verified by comparing with historical data to obtain the optimized classification output.

[0098] Specifically, weighted regional data is integrated with temperature and chemical composition data. After feature extraction, a support vector machine is used to initially classify defects. A secondary verification is performed when confidence is low, and the final classification is determined based on the degree of match. The output is optimized using historical data. This significantly improves the accuracy and stability of complex defect classification, providing key support for precise defect handling on production lines, ensuring product quality and production efficiency.

[0099] S5. Calculate the priority index of the defect area through the final defect classification. When the priority index exceeds a preset threshold, trigger the high-resolution imaging module to perform a deep scan of the area to obtain detailed defect features.

[0100] Furthermore, in step S5, detailed defect characteristics are obtained, including:

[0101] The priority index is calculated based on the defect classification data. The classification results are processed using the support vector machine algorithm to obtain the priority index of the defect area. When the priority index exceeds the preset threshold, the trigger condition is determined to be met based on the calculation result, and the defect area that requires deep scanning is obtained;

[0102] Divide the scanning area according to the defect area, use the imaging module to generate a high-resolution image, obtain preliminary scanning data, process the preliminary scanning data through deep scanning, use the convolutional neural network algorithm to extract feature details, and obtain the first feature set;

[0103] Analyze defect features based on the first feature set, determine whether the feature details are complete, and obtain a second feature set. Then, update the defect classification data based on the second feature set, use statistical tools to verify the classification consistency, and obtain an optimized defect classification.

[0104] Recalculate the priority index based on the optimized defect classification to determine whether to trigger a new round of deep scanning and obtain the final feature details.

[0105] Specifically, it achieves intelligent, precise detection and classification optimization of defect areas. The support vector machine algorithm is used to process defect classification data to obtain a priority index, which is used to intelligently determine the defect areas that need to be focused on. When the index exceeds the threshold, the high-resolution imaging module is triggered to perform a deep scan, and the convolutional neural network is used to extract feature details. After analyzing the feature integrity, the defect classification data is updated, and the classification consistency is verified through statistical tools to optimize the classification. The priority index is then recalculated to decide whether to start a new round of deep scanning, and finally detailed and accurate defect features are obtained. This process can not only focus on key defect areas and improve detection efficiency, but also continuously optimize defect classification, provide high-precision defect feature information for product quality control, and help automated production lines take more effective quality improvement measures.

[0106] S6. Based on the depth scan data and combined with the existing information from multi-band imaging, a random forest algorithm is used to conduct a multi-dimensional integrated analysis of the defect characteristics to determine the potential impact of the defect on the automated production line. Key indicators are extracted from the potential impact, and the test results are fed back to the production line control system through a real-time dynamic adjustment mechanism to obtain an optimized production parameter adjustment plan.

[0107] Furthermore, in step S6, the potential impact of the defect on the automated production line is determined, specifically including: acquiring data from the production line through deep scanning, extracting original information using multi-band imaging technology, obtaining preliminary defect features, and then acquiring multi-dimensional data, integrating and analyzing it using the random forest algorithm to determine a set of feature vectors;

[0108] Extract key information from the feature vector set, classify defect types through information processing technology, determine the defect distribution pattern, obtain production line operation data based on the defect distribution pattern, and use the support vector machine algorithm to analyze the potential impact of defects on automated production to obtain an estimated impact range;

[0109] The estimated impact range is logically judged using a preset threshold. When the threshold is exceeded, information processing technology is used to generate adjustment parameters and determine an optimization solution.

[0110] Adjustment parameters are extracted from the optimization plan, real-time production line data is obtained, and the impact of the adjustment is determined through comparative analysis. Based on the impact of the adjustment, information processing technology is used to verify the results and obtain the final judgment result.

[0111] Furthermore, after determining the potential impact of defects on the automated production line, the method also includes: extracting key indicators through impact analysis to obtain an indicator set, comparing the indicator set with the test results, and triggering real-time contradiction judgment when the difference exceeds a preset threshold to obtain a contradictory state, driving a dynamic adjustment mechanism based on the contradictory state to obtain an adjustment direction; updating the feedback mechanism through the adjustment direction to determine the feedback signal, transmitting the feedback signal to the control system, determining the optimal parameter combination, adjusting the production parameters based on the optimized parameter combination, and obtaining a parameter solution.

[0112] Specifically, the random forest algorithm is used to integrate deep scanning and multi-band imaging data to accurately determine the potential impact of defects on the automated production line. The support vector machine algorithm is combined with the production line operation data to obtain an estimated value of the impact range. When the threshold is exceeded, adjustment parameters are generated to determine the optimization plan, and the effect is verified based on real-time data. Subsequently, key indicators are extracted and compared with the test results to trigger real-time conflict judgment to drive the dynamic adjustment mechanism, update the feedback mechanism, and enable the control system to adjust the production parameters according to the optimized parameter combination, achieving adaptive adjustment of the production line, effectively improving production efficiency and product quality, reducing costs, and enhancing corporate competitiveness.

[0113] Example 2

[0114] See also Figure 4 This embodiment provides a system for intelligently detecting product defects on an automated production line based on machine vision, which is used to implement an intelligent method for detecting product defects on an automated production line, including the following steps:

[0115] A multi-band imaging module is used to collect image data of products under visible light, infrared light, and ultraviolet light, and integrate temperature information capture and chemical composition identification to generate an initial data set containing multi-dimensional features;

[0116] The data processing module is used to analyze the initial data set, perform cluster analysis using an unsupervised learning algorithm, determine the characteristic distribution range of the product's normal state, and extract abnormal areas. It is also used to perform preliminary convolutional neural network classification on abnormal areas to obtain the location results of potential defect areas;

[0117] The attention mechanism module is used to perform weighted processing on potential defect areas, analyze the internal structure and surface morphology through multi-dimensional feature detection, and determine the defect type and severity;

[0118] Complex defect verification module, used to perform secondary verification of the support vector machine algorithm on the weighted regional data to determine the final defect classification;

[0119] High-resolution imaging module, used to perform a deep scan of the defect area to obtain detailed defect features when the priority index of the defect area exceeds a preset threshold;

[0120] The defect analysis module combines multi-band imaging data and depth scanning data to perform multi-dimensional integrated analysis using a random forest algorithm to determine the potential impact of defects on the production line and extract key indicators.

[0121] The feedback control module is used to feed back the detection results to the production line control system through a dynamic adjustment mechanism of real-time contradictions, and generate an optimized production parameter adjustment plan.

[0122] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for intelligently detecting product defects on an automated production line, characterized by: The process includes the following steps: obtaining product positive sample data, collecting visible light, infrared light, and ultraviolet light images through multi-band imaging technology, integrating temperature information capture and chemical composition identification to obtain an initial data set containing multi-dimensional features; Based on the initial data set, a positive sample modeling method is used to cluster the multi-dimensional features using a pre-established unsupervised learning algorithm to determine the distribution range of the product's normal state characteristics. Abnormal areas are then extracted from the multi-band imaging data and preliminarily classified using a convolutional neural network to obtain the location results of potential defect areas. Based on the positioning results, the attention mechanism is applied to perform weighted processing on the potential defect areas, and the internal structure and surface morphology are analyzed through multi-dimensional feature detection to determine the defect type and severity; Obtain weighted regional data, integrate the results of temperature information capture and chemical composition identification, use the support vector machine algorithm to perform secondary verification of complex defect identification, and determine the final defect classification; Through the final defect classification, the priority index of the defect area focus is calculated. When the priority index exceeds the preset threshold, the high-resolution imaging module is triggered to perform a deep scan of the area to obtain detailed defect features; Based on the depth scan data and combined with the existing information of multi-band imaging, the random forest algorithm is used to conduct a multi-dimensional integrated analysis of the defect characteristics to determine the potential impact of the defects on the automated production line. Key indicators are extracted from the potential impact. Through a real-time dynamic adjustment mechanism of contradictions, the detection results are fed back to the production line control system to obtain an optimized production parameter adjustment plan.

2. The intelligent product defect detection method on an automated production line according to claim 1, characterized in that: Get an initial data set containing multi-dimensional features, including: The visible light image, infrared light image and ultraviolet light image of the positive sample data are collected by multi-band imaging technology to obtain a first image set with multi-dimensional features, and then the temperature information is superimposed by fusion processing technology to obtain a second image set containing temperature features; Extract chemical composition data from the second image set, determine the chemical composition distribution using recognition technology, and obtain a third image set with component annotations. Based on the multidimensional features of the third image set, determine whether the difference between the features exceeds a preset threshold. If so, determine the main features using a principal component analysis algorithm to obtain a fourth image set after dimensionality reduction. According to the feature distribution of the fourth image set, a clustering algorithm is used to judge the similarity between samples to obtain the classified fifth image set. According to the classification results of the fifth image set, the temperature information and chemical composition distribution are integrated to obtain the final initial data set of multidimensional features.

3. The intelligent product defect detection method on an automated production line according to claim 1, characterized in that: Determine the characteristic distribution range of the product in normal state, including: Obtain multidimensional features through the initial data set, use unsupervised clustering algorithm to analyze the multidimensional features and obtain feature distribution; Use statistical tools to determine the distribution range of feature distribution and judge the boundary of the normal state of the product. When the feature distribution exceeds the distribution range, adjust the feature analysis through positive sample modeling to determine abnormal features; According to the abnormal characteristics, relevant subsets are extracted from the initial data set to obtain the subset feature distribution. Then, an unsupervised clustering algorithm is used to perform a secondary analysis on the subset feature distribution to determine the normal state of the subset. By comparing the normal state of the subset with the normal state of the product, the consistency is judged and the state deviation is obtained. Then, based on the state deviation, the data modeling parameters are adjusted to determine the optimized feature distribution range.

4. The method for intelligently detecting product defects on an automated production line according to claim 3, wherein: After determining the characteristic distribution range of the product's normal state, it also includes: acquiring imaging data through multi-band data, using statistical methods to determine whether the characteristic value exceeds the normal distribution range to obtain the abnormal area, then extracting characteristic values from the abnormal area, classifying the characteristic values through a convolutional neural network to obtain a preliminary classification result, analyzing potential defects based on the preliminary classification results, using regional analysis technology to determine the defect boundary to obtain a positioning result, then comparing the abnormal area with the imaging data to determine whether the defect is consistent with the characteristics of the multi-band data to obtain a verification result, extracting the change trend of the abnormal area from the verification result, analyzing the change trend through a convolutional neural network to obtain the defect evolution characteristics; based on the defect evolution characteristics, using the positioning results to update the regional analysis, determine the expansion range of the abnormal area, and obtain the final positioning, obtaining the dynamic changes of the multi-band data through the final positioning, using statistical methods to analyze the characteristic value distribution, and determining the defect classification accuracy.

5. The intelligent product defect detection method on an automated production line according to claim 1, characterized in that: Determine the type and severity of the defect, including: The potential defect area is determined by the positioning results, and the attention mechanism is used to perform weighted processing on the area to obtain weighted feature data. Multidimensional features are obtained from the weighted feature data, and then the multidimensional features are preliminarily classified to determine whether there are significant abnormalities. When significant anomalies exist, convolutional neural networks are used to analyze multidimensional features and detect the changing trends of the internal structure. Based on these trends, clustering algorithms are used to group potential defects and determine the distribution range of defect types. By analyzing the characteristic data of surface morphology based on the distribution range, the severity level of the defect is determined. After obtaining the severity level, the final defect classification result is obtained by combining it with the internal structure information. The key features are extracted from the final defect classification result to determine the evolution direction of the potential defect.

6. The intelligent method for detecting product defects on an automated production line according to claim 1, characterized in that: Determine the final defect classification, including: Obtain weighted regional data, then collect temperature information and chemical composition data through sensors, fuse them into the weighted regional data set to obtain a comprehensive feature set, and use a preset feature extraction method to separate complex defect-related features from the comprehensive feature set to obtain a defect feature set; The defect feature set is initially classified using the support vector machine algorithm to obtain a preliminary defect category. When the confidence level of the preliminary defect category is lower than the preset threshold, the defect feature set is re-analyzed through a secondary verification process to obtain an adjusted defect category. The final defect classification result is determined based on the matching degree between the adjusted defect category and the comprehensive feature set. The accuracy of the final defect classification result is verified by comparing with historical data to obtain the optimized classification output.

7. The intelligent method for detecting product defects on an automated production line according to claim 1, wherein: Obtain detailed defect characteristics, including: The priority index is calculated based on the defect classification data. The classification results are processed using the support vector machine algorithm to obtain the priority index of the defect area. When the priority index exceeds the preset threshold, the trigger condition is determined to be met based on the calculation result, and the defect area that requires deep scanning is obtained; Divide the scanning area according to the defect area, use the imaging module to generate a high-resolution image, obtain preliminary scanning data, process the preliminary scanning data through deep scanning, use the convolutional neural network algorithm to extract feature details, and obtain the first feature set; Analyze defect features based on the first feature set, determine whether the feature details are complete, and obtain a second feature set. Then, update the defect classification data based on the second feature set, use statistical tools to verify the classification consistency, and obtain an optimized defect classification. Recalculate the priority index based on the optimized defect classification to determine whether to trigger a new round of deep scanning and obtain the final feature details.

8. The intelligent product defect detection method on an automated production line according to claim 1, characterized in that: Determine the potential impact of defects on automated production lines, specifically by acquiring data from the production line through deep scanning, extracting raw information using multi-band imaging technology to obtain preliminary defect features, and then acquiring multi-dimensional data and integrating and analyzing it using the random forest algorithm to determine a set of feature vectors. Extract key information from the feature vector set, classify defect types through information processing technology, determine the defect distribution pattern, obtain production line operation data based on the defect distribution pattern, and use the support vector machine algorithm to analyze the potential impact of defects on automated production to obtain an estimated impact range; The estimated impact range is logically judged using a preset threshold. When the threshold is exceeded, information processing technology is used to generate adjustment parameters and determine an optimization solution. Adjustment parameters are extracted from the optimization plan, real-time production line data is obtained, and the impact of the adjustment is determined through comparative analysis. Based on the impact of the adjustment, information processing technology is used to verify the results and obtain the final judgment result.

9. The intelligent method for detecting product defects on an automated production line according to claim 8, characterized in that: After determining the potential impact of defects on the automated production line, the process also includes: extracting key indicators through impact analysis to obtain an indicator set, comparing the indicator set with the test results, and triggering real-time conflict judgment when the difference exceeds a preset threshold to obtain a conflicting state. The dynamic adjustment mechanism is driven according to the conflicting state to obtain the adjustment direction; the feedback mechanism is updated through the adjustment direction to determine the feedback signal, which is transmitted from the feedback signal to the control system to determine the optimal parameter combination, adjust the production parameters according to the optimized parameter combination, and obtain a parameter solution.

10. A system for intelligently detecting product defects on an automated production line based on machine vision, for implementing the method for intelligently detecting product defects on an automated production line according to any one of claims 1 to 9, characterized in that: The steps include: A multi-band imaging module is used to collect image data of products under visible light, infrared light, and ultraviolet light, and integrate temperature information capture and chemical composition identification to generate an initial data set containing multi-dimensional features; The data processing module is used to analyze the initial data set, perform cluster analysis using an unsupervised learning algorithm, determine the characteristic distribution range of the product's normal state, and extract abnormal areas. It is also used to perform preliminary convolutional neural network classification on abnormal areas to obtain the location results of potential defect areas; The attention mechanism module is used to perform weighted processing on potential defect areas, analyze the internal structure and surface morphology through multi-dimensional feature detection, and determine the defect type and severity; Complex defect verification module, used to perform secondary verification of the support vector machine algorithm on the weighted regional data to determine the final defect classification; High-resolution imaging module, used to perform a deep scan of the defect area to obtain detailed defect features when the priority index of the defect area exceeds a preset threshold; The defect analysis module combines multi-band imaging data and depth scanning data to perform multi-dimensional integrated analysis using a random forest algorithm to determine the potential impact of defects on the production line and extract key indicators. The feedback control module is used to feed back the detection results to the production line control system through a dynamic adjustment mechanism of real-time contradictions, and generate an optimized production parameter adjustment plan.

Images