Accessory identification method and system based on image classification model

Abstract

The invention relates to the technical field of computer vision and image processing, in particular to an accessory identification method and system based on an image classification model. The method comprises the following steps: acquiring industrial accessory data, and carrying out topological structure analysis according to the industrial accessory data so as to obtain industrial accessory topological structure data; performing accessory corrosion simulation according to the industrial accessory topological structure data to obtain accessory corrosion data; positioning an accessory corrosion position based on the accessory corrosion data so as to obtain accessory corrosion position data, and performing image acquisition so as to obtain an accessory corrosion image; performing corrosion stress concentration point analysis on the accessory corrosion image to obtain a corrosion stress concentration point image; and performing type division on the corrosion stress concentration point image so as to obtain a pitting corrosion stress concentration point image and a uniform corrosion stress concentration point image. According to the invention, the fault identification and life prediction accuracy of industrial accessories is improved based on computer vision and image processing technologies.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and image processing, and particularly relates to a method and system for accessory recognition based on an image classification model. Background Art

[0002] Traditional methods usually rely on manual feature extraction, which leads to low efficiency and is vulnerable to human factors, and it is difficult to adapt to different types of accessories. Due to the diversity of accessory corrosion data and accessory type images, traditional methods often cannot achieve highly accurate classification in image classification, and are prone to misidentification or missed identification, especially when there are complex corrosion morphologies on the surface of accessories. Traditional methods have a relatively single analysis of the corrosion models of different accessories in different environments, lacking accurate modeling of actual environmental factors (such as humidity, electrolyte concentration, etc.), resulting in low accuracy in identification and life prediction under extreme working conditions. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide a method and system for accessory recognition based on an image classification model to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for accessory recognition based on an image classification model includes the following steps:

[0005] Step S1: Obtain industrial accessory data, and perform topological structure analysis according to the industrial accessory data to obtain industrial accessory topological structure data; perform accessory corrosion simulation according to the industrial accessory topological structure data to obtain accessory corrosion data;

[0006] Step S2: Locate the accessory corrosion position based on the accessory corrosion data to obtain accessory corrosion position data, and perform image acquisition to obtain accessory corrosion images; analyze the corrosion stress concentration points of the accessory corrosion images to obtain corrosion stress concentration point images;

[0007] Step S3: Classify the corrosion stress concentration point images to obtain pitting corrosion stress concentration point images and uniform corrosion stress concentration point images; classify the accessory types according to the industrial accessory topological structure data to obtain accessory type data, and perform accessory type image acquisition to obtain accessory type images;

[0008] Step S4: Construct an image classification model according to the accessory type images, and perform accessory recognition on the industrial accessory data to obtain accessory recognition data; predict the accessory life according to the pitting corrosion stress concentration point images and the uniform corrosion stress concentration point images for the accessory recognition data to obtain accessory life prediction data;

[0009] Step S5: Based on the accessory life prediction data, perform accessory fault probability modeling on the accessory identification data to obtain accessory fault identification data.

[0010] Through the topological structure analysis and corrosion simulation of industrial accessory data, the present invention provides comprehensive data support for the monitoring of the corrosion state of accessories. By locating the corrosion position and collecting images, the corrosion area of the accessory can be accurately obtained, and further analysis of the corrosion stress concentration points can be carried out, providing detailed image data for the corrosion characteristics of the accessory, and avoiding the inefficiency and susceptibility to human factors of traditional manual feature extraction. In the process of image classification of corrosion stress concentration points, by dividing pitting corrosion and uniform corrosion, the accuracy of accessory corrosion identification is improved. Especially when facing complex corrosion morphologies, the situations of misidentification and missed identification can be effectively reduced. In addition, by combining the accessory type images and the accessory topological structure data, more accurate classification of different types of accessories can be carried out, solving the problem of insufficient adaptation to the diversity of accessories in traditional methods. The prediction of accessory life is not only based on the images of corrosion stress concentration points, but also combines the accessory type images, which helps to more accurately predict the service life of accessories, thus providing a more reliable data basis for subsequent fault probability modeling and fault identification. Finally, this method avoids the singularity in the corrosion model analysis and environmental factor modeling of traditional methods, and can accurately identify and predict the life according to different accessories, environmental factors and working conditions, significantly improving the efficiency and accuracy of accessory health management.

[0011] Preferably, step S1 is specifically as follows:

[0012] Step S11: Obtain industrial accessory data, and perform accessory hole feature extraction and accessory slot feature extraction according to the industrial accessory data to obtain accessory hole data and accessory slot data;

[0013] Step S12: Based on the accessory hole data, perform assembly connection point channel identification on the accessory slot data to obtain assembly connection point channel data;

[0014] Step S13: Perform connection node accessory feature extraction on the industrial accessory data to obtain connection node accessory data;

[0015] Step S14: Based on the connection node accessories and the assembly connection point channel data, perform topological construction to obtain industrial accessory topological structure data;

[0016] Step S15: Based on the industrial accessory topological structure data, perform accessory corrosion analysis to obtain accessory corrosion data.

[0017] Through the extraction of hole and groove features of industrial fittings, the present invention provides a more refined description of the geometric structure and functional characteristics of the fittings, overcoming the problem of low efficiency caused by manual feature extraction in traditional methods. Based on the fitting hole data, further identification of the assembly connection point channels is carried out, providing more explicit data support for the assembly and connection of the fittings, and enabling better analysis of the assembly process of complex fittings. Through the extraction of the features of the connection node fittings, the connection information of the fittings can be comprehensively obtained, providing detailed structural data for subsequent topology construction, thereby enhancing the accuracy of the fitting structure analysis. According to the topology structure data, corrosion analysis is carried out, which not only provides corrosion prediction based on actual geometric features, but also solves the problems of single modeling of fitting corrosion and inability to adapt to complex environmental changes in traditional methods. After integrating the operations of these steps, the corrosion modes in different environments can be more accurately identified, improving the accuracy of fitting identification, corrosion detection and life prediction. Especially in the face of extreme working conditions, this method can better adapt to the changes of environmental factors, avoiding misidentification and missed identification in traditional methods, and significantly improving the health management ability of the fittings.

[0018] Preferably, step S15 is specifically as follows:

[0019] Step S151: Obtain industrial environment humidity data by using industrial environment sensors;

[0020] Step S152: Divide the industrial fitting topology structure data according to the industrial environment humidity data to obtain humidity-affected industrial fitting data;

[0021] Step S153: Collect industrial fittings according to the humidity-affected industrial fitting data to obtain humidity-affected industrial fittings;

[0022] Step S154: Conduct electrochemical corrosion simulation on the humidity-affected industrial fittings according to the industrial environment humidity data to obtain electrochemical corrosion industrial fitting data;

[0023] Step S156: Analyze the formation of the electrolyte film on the electrochemical corrosion industrial fitting data according to the industrial environment humidity data to obtain electrolyte film data;

[0024] Step S157: Draw a current density distribution map according to the electrolyte film data to obtain a current density distribution map;

[0025] Step S158: Identify the high-density area according to the current density distribution map to obtain high current density area data;

[0026] Step S159: Conduct corrosion analysis on the humidity-affected industrial fittings according to the high current density area data to obtain fitting corrosion data.

[0027] By using industrial environment sensors to obtain humidity data, the present invention can more accurately reflect the changes in the actual environment, avoiding the analysis deviation caused by ignoring environmental factors in traditional methods. Based on the humidity data, the accessories affected by humidity can be accurately identified by partitioning the accessories, so as to conduct targeted corrosion analysis, avoiding the misidentification problem caused by the complexity of accessory types and environmental conditions in traditional methods. Electrochemical corrosion simulation of the accessories affected by humidity can more realistically simulate the corrosion behavior of the accessories in a humid environment, thus obtaining more scientific corrosion data and providing more accurate data support for subsequent analysis. The electrolyte film formation analysis provides a basis for simulating the role of the electrolyte in the corrosion process. By processing the electrolyte film data, the important influencing factors in the corrosion process can be revealed, improving the accuracy of corrosion prediction. By drawing and identifying the high-density area according to the current density distribution map, the area with the most serious corrosion can be accurately determined, avoiding the problem of unclear identification of the corrosion area in traditional methods and ensuring the reliability of the corrosion analysis results. Finally, through the analysis of the data in the high current density area, a more accurate assessment of the corrosion situation of the accessories affected by humidity can be provided, improving the comprehensiveness and accuracy of corrosion analysis. Especially under extreme working conditions, this method can better adapt to the influence of humidity changes and reduce the errors caused by insufficient models in traditional methods.

[0028] Preferably, step S2 is specifically as follows:

[0029] Step S21: Locate the corrosion position of the accessory based on the accessory corrosion data, so as to obtain the accessory corrosion position data, and perform image acquisition to obtain the accessory corrosion image;

[0030] Step S22: Analyze the pore defects of the accessory corrosion image to obtain the pore defect data;

[0031] Step S23: Analyze the crack defects of the accessory corrosion image to obtain the crack defect data;

[0032] Step S24: Identify the position of the accessory corrosion image according to the pore defect data to obtain the pore defect position data;

[0033] Step S25: Identify the position of the accessory corrosion image according to the crack defect data to obtain the crack defect position data;

[0034] Step S26: Perform an intersection operation on the corrosion stress concentration points according to the pore defect position data and the crack defect position data to obtain the corrosion stress concentration point data, and perform image acquisition to obtain the corrosion stress concentration point image.

[0035] The present invention can accurately identify the corrosion positions on the surface of fittings by locating the corrosion positions of fittings based on fitting corrosion data, ensuring the efficiency and accuracy of data collection, and avoiding the error and efficiency problems caused by manual feature extraction in traditional methods. The image acquisition step ensures that clear and real fitting corrosion images are obtained during the analysis process, providing a reliable basis for subsequent analysis. Analyzing the pore defects in the fitting corrosion image can accurately identify the pore defect data in the image, thus avoiding the defect of incomplete defect identification in traditional methods. The crack defect analysis enables in-depth exploration of the crack problems in the fitting corrosion image, providing a scientific basis for subsequent repair or replacement decisions. Identifying the positions of the pore defect data and the crack defect data can accurately calibrate the specific positions of the defects, effectively reducing the uncertainty brought by manual determination in traditional methods. Through the combined operation of the pore defect position data and the crack defect position data, the corrosion stress concentration points can be identified, thereby improving the accuracy of corrosion analysis and avoiding the problem of inability to accurately locate under complex corrosion morphologies. Finally, the acquisition of the corrosion stress concentration point image provides a comprehensive and intuitive presentation for the entire corrosion analysis, effectively improving the monitoring and early warning capabilities for the corrosion stress concentration area, and ensuring the efficiency and accuracy of fitting life prediction.

[0036] Preferably, step S22 is specifically as follows:

[0037] Step S221: Denoise the fitting corrosion image to obtain a denoised fitting corrosion image;

[0038] Step S222: Perform binary conversion according to the denoised fitting corrosion image to obtain a binary fitting corrosion image;

[0039] Step S223: Obtain historical pore edge data and perform pore edge detection on the binary fitting corrosion image to obtain a pore edge image;

[0040] Step S224: Perform roundness detection on the pore edge image to obtain roundness data, and perform statistics on the roundness data to obtain high-roundness data;

[0041] Step S225: Locate the pore centroid of the binary fitting corrosion image according to the high-roundness data to obtain pore centroid data;

[0042] Step S226: Generate a heat map according to the pore centroid data to obtain a pore heat map;

[0043] Step S227: Evaluate the number of pores in the pore heat map to obtain pore number data;

[0044] Step S228: Perform pore depth evaluation on the binarized fitting corrosion image based on the pore centroid data, so as to obtain pore depth data;

[0045] Step S229: Perform pore defect integration based on the pore quantity data and the pore depth data, so as to obtain pore defect data.

[0046] Through denoising the fitting corrosion image, the present invention can effectively eliminate the noise interference in the image, enhance the clarity of the image, and provide a more accurate basis for subsequent analysis. The binarization conversion further simplifies the complexity of the image and converts it into a binary image that is easy to analyze, which helps to quickly identify the corrosion areas on the fitting surface and improves the efficiency of image analysis. The pore edge detection can accurately identify the edges of the pores, thereby accurately positioning the morphology and distribution of the pores, effectively improving the accuracy of pore defect detection. The roundness detection can evaluate the uniformity of the pores by carefully analyzing the pore morphology, further enhancing the accurate evaluation of corrosion defects and avoiding the problem that pores with irregular morphologies are ignored or misclassified in traditional methods. The statistics of high roundness data helps to identify pores with high roundness, further refining the defect classification and contributing to optimizing subsequent defect repair schemes. The pore centroid positioning accurately locates the core positions of the pores, facilitating subsequent depth and quantity analysis. The heat map generation visually presents the pore distribution information, which helps to perform visual analysis on the pore distribution pattern, thereby providing support for subsequent quality control and maintenance decisions. The pore quantity evaluation provides data support for the comprehensive analysis of pore defects and helps to evaluate the severity of corrosion. The pore depth evaluation further analyzes the extent of pore expansion and provides detailed information on the impact of corrosion on the fitting. Through the integration of the pore quantity data and the pore depth data, a comprehensive evaluation of pore defects can be carried out, further enhancing the accuracy and comprehensiveness of fitting corrosion detection.

[0047] Preferably, step S228 is specifically as follows:

[0048] Perform region positioning on the binarized fitting corrosion image based on the pore centroid data, so as to obtain fitting pore corrosion region data;

[0049] Perform three-dimensional reconstruction of the fitting based on the fitting pore corrosion region data, so as to obtain a three-dimensional pore corrosion fitting;

[0050] Perform light mapping simulation on the three-dimensional pore corrosion fitting, so as to obtain light mapping data;

[0051] Perform light gradient calculation on the light mapping data, so as to obtain light gradient data;

[0052] Perform pore depth calculation based on the light gradient data, so as to obtain pore depth data.

[0053] Through region localization of the binary fitting corrosion image based on stomatal centroid data, the present invention can accurately identify the stomatal corrosion region on the fitting surface, providing a more accurate data basis for subsequent analysis. The three-dimensional reconstruction of the fitting uses the data of the stomatal corrosion region to generate the spatial model of the fitting, providing a more intuitive and realistic display of the corrosion morphology, which helps to more deeply understand the influence of the corrosion process on the fitting structure. The light mapping simulation can further analyze the influence of corrosion on light propagation by simulating the surface reflection characteristics of the fitting under different lighting conditions, providing more physical basis for defect analysis. The light gradient calculation can more accurately reflect the depth change of the corrosion region by calculating the gradient of the lighting change, which helps to reveal the depth characteristics of the stomata. Finally, the stomatal depth can be estimated based on the light gradient data, enabling an accurate assessment of the stomatal expansion degree and providing reliable support for the remaining life prediction and corrosion control of the fitting.

[0054] Preferably, step S23 is specifically as follows:

[0055] Step S231: Convert the fitting corrosion image to grayscale to obtain the grayscale fitting corrosion image;

[0056] Step S232: Calculate the pixel gradient based on the grayscale fitting corrosion image to obtain the pixel gradient data;

[0057] Step S233: Draw a gradient magnitude map based on the pixel gradient data to obtain the gradient magnitude map;

[0058] Step S234: Set the high crack threshold and the low crack threshold based on the grayscale fitting corrosion image to obtain the high crack threshold data and the low crack threshold data;

[0059] Step S235: Perform edge detection on the gradient magnitude map based on the high crack threshold data and the low crack threshold data. If the pixel gradient is greater than the high crack threshold data, obtain the first pixel edge point data; if the pixel gradient is between the high crack threshold data and the low crack threshold data, further judge. If the pixel gradient is connected to the first pixel edge point data, obtain the second pixel edge point data;

[0060] Step S236: Integrate the first pixel edge point data and the second pixel edge point data to obtain the crack pixel edge point data;

[0061] Step S237: Calculate the crack length based on the crack pixel edge point data for the grayscale fitting corrosion image to obtain the crack length data;

[0062] Step S238: Calculate the crack width based on the crack pixel edge point data for the grayscale fitting corrosion image to obtain the crack width data;

[0063] Step S239: Evaluate the crack area based on the crack length data and crack width data to obtain crack area data, and visualize the crack defects on the gray-scale fitting corrosion image according to the crack area data to obtain crack defect data.

[0064] Through gray-scale conversion of the fitting corrosion image, the present invention can simplify the image processing process and make subsequent corrosion feature analysis more efficient. Pixel gradient calculation can extract important edge information in the image, which helps to capture the features of cracks and other subtle defects. The drawing of the gradient magnitude map can effectively display the intensity changes in the image, facilitating subsequent detailed edge detection. Setting high and low thresholds for cracks helps to accurately identify the crack edges, thereby improving the accuracy of crack detection. By performing edge detection on the gradient magnitude map and combining the judgment of the first pixel edge point and the second pixel edge point, the edges of the cracks can be accurately extracted, reducing the possibility of false detection and missed detection. The integration of crack pixel edge point data makes the acquisition of the crack contour more accurate, thus providing a high-quality data basis for the calculation of the crack length, width, and area. Calculating the length and width of the cracks can not only quantify the scale of the cracks but also provide a reference for the remaining life assessment of the fittings. The crack area evaluation combined with crack defect visualization can not only show the actual impact of the cracks but also help users intuitively understand the damage degree of corrosion to the fittings, thereby providing support for further corrosion control and maintenance decisions.

[0065] Preferably, step S3 is specifically as follows:

[0066] Step S31: Identify the corrosion morphology of the corrosion stress concentration point image to obtain corrosion morphology data;

[0067] Step S32: Classify the corrosion stress concentration point image according to the corrosion morphology data to obtain pitting corrosion stress concentration point images and uniform corrosion stress concentration point images;

[0068] Step S33: Classify the fitting types according to the industrial fitting topology structure data to obtain control fittings and transmission fittings;

[0069] Step S34: Collect high-resolution images of the control fittings to obtain high-resolution control fitting images;

[0070] Step S35: Perform infrared imaging acquisition on the transmission fittings to obtain infrared imaging of the transmission fittings;

[0071] Step S36: Integrate the fitting type images according to the high-resolution control fitting images and the infrared imaging of the transmission fittings to obtain fitting type images.

[0072] Through the corrosion morphology recognition of the corrosion stress concentration point images, the present invention can accurately identify different corrosion morphologies on the surface of the fittings, thereby providing detailed corrosion characteristic data for subsequent analysis. Classifying according to the corrosion morphology data helps to divide the corrosion morphology into two types: pitting corrosion and uniform corrosion, which can more meticulously analyze the influence of different types of corrosion on the performance of the fittings, and then take corresponding maintenance measures. By classifying the fitting types according to the industrial fitting topological structure data, the control fittings and transmission fittings can be effectively distinguished. Different types of fittings require different processing methods and models in corrosion analysis, making the analysis more targeted. High-resolution image acquisition of the control fittings can obtain more accurate images of the fitting surface, which is crucial for further analyzing its corrosion condition and predicting its remaining life. Infrared imaging acquisition of the transmission fittings can obtain important information on temperature distribution and thermal characteristics, providing supplementary data for further evaluating its corrosion condition. By integrating the high-resolution images of the control fittings with the infrared imaging images of the transmission fittings, a comprehensive fitting type image can be formed, further enhancing the overall analysis ability of the fittings, and thus helping to make more scientific and accurate decisions when maintaining and replacing the fittings.

[0073] Preferably, step S4 is specifically as follows:

[0074] Step S41: Construct an image classification model based on the fitting type image, input the industrial fitting data into the image classification model, and perform fitting image classification to obtain a fitting classification image;

[0075] Step S42: Identify the fittings according to the fitting classification image to obtain fitting identification data;

[0076] Step S43: Predict the local fatigue life of the fitting identification data based on the pitting corrosion stress concentration point image to obtain local fatigue life prediction data;

[0077] Step S44: Predict the global fatigue cumulative life of the fitting identification data based on the uniform corrosion stress concentration point image to obtain global fatigue cumulative life prediction data;

[0078] Step S45: Integrate the local fatigue life prediction data and the global fatigue cumulative life prediction data to obtain fitting life prediction data.

[0079] Through constructing an image classification model and inputting and classifying industrial accessory data, the present invention can achieve accurate identification and classification of different types of accessories, greatly improving the analysis efficiency and reducing human errors. The generation of accessory classification images enables clear division of different features of accessories, thus contributing to subsequent fatigue life prediction. By conducting local fatigue life prediction on accessory identification data, the service life of accessories in a local area can be evaluated in detail, especially in areas affected by pitting stress concentration points, which can provide early warnings for the maintenance of high-risk areas. Through global fatigue cumulative life prediction on images of uniform corrosion stress concentration points, the fatigue cumulative effect suffered by accessories during the entire usage process can be evaluated, which is crucial for long-term reliability assessment. Integrating local fatigue life prediction data and global fatigue cumulative life prediction data can not only provide a more comprehensive accessory life assessment but also improve the accuracy of life prediction, providing a more scientific basis for the maintenance, replacement of accessories, and optimization of the working environment.

[0080] Preferably, this specification also provides an accessory identification system based on an image classification model for implementing an accessory identification method based on an image classification model as described above. The accessory identification system based on an image classification model includes:

[0081] An accessory corrosion simulation module, configured to obtain industrial accessory data and conduct topological structure analysis based on the industrial accessory data to obtain industrial accessory topological structure data; conduct accessory corrosion simulation based on the industrial accessory topological structure data to obtain accessory corrosion data;

[0082] A corrosion stress concentration point analysis module, configured to locate the accessory corrosion position based on the accessory corrosion data to obtain accessory corrosion position data, and conduct image acquisition to obtain an accessory corrosion image; conduct corrosion stress concentration point analysis on the accessory corrosion image to obtain a corrosion stress concentration point image;

[0083] An accessory type division module, configured to divide the type of the corrosion stress concentration point image to obtain a pitting stress concentration point image and a uniform corrosion stress concentration point image; divide the type of accessories based on the industrial accessory topological structure data to obtain accessory type data, and conduct accessory type image acquisition to obtain an accessory type image;

[0084] An accessory life prediction module, configured to construct an image classification model based on the accessory type image and conduct accessory identification on the industrial accessory data to obtain accessory identification data; conduct accessory life prediction on the accessory identification data based on the pitting stress concentration point image and the uniform corrosion stress concentration point image to obtain accessory life prediction data;

[0085] The accessory failure identification module is used to perform accessory failure probability modeling on accessory identification data according to accessory life prediction data, so as to obtain accessory failure identification data.

[0086] The accessory identification system based on the image classification model of the present invention can implement any accessory identification method based on the image classification model of the present invention. It is a medium for coordinating the operations and signal transmissions between various modules to complete the accessory identification method based on the image classification model. The internal modules of the system cooperate with each other, improving the accuracy of industrial accessory failure identification and life prediction. Brief Description of the Drawings

[0087] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, objectives, and advantages of the present invention will become more apparent:

[0088] Figure 1 It is a schematic flowchart of the steps of the accessory identification method based on the image classification model of the present invention;

[0089] Figure 2 It is a detailed schematic flowchart of step S1 in the present invention;

[0090] Figure 3 It is a detailed schematic flowchart of step S4 in the present invention;

[0091] The implementation, functional characteristics, and advantages of the objectives of the present invention will be further described in conjunction with the embodiments and with reference to the drawings. Detailed Embodiments

[0092] The technical method of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0093] In addition, the drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus their repeated descriptions will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0094] It should be understood that although terms such as "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly, the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0095] To achieve the above object, please refer to Figures 1 to 3 , the present invention provides a method for identifying accessories based on an image classification model, and the method includes the following steps:

[0096] Step S1: Obtain industrial accessory data, and perform topological structure analysis based on the industrial accessory data to obtain industrial accessory topological structure data; perform accessory corrosion simulation based on the industrial accessory topological structure data to obtain accessory corrosion data;

[0097] In this embodiment, industrial accessory data is obtained. These data can be collected through physical scanning, 3D scanners or other sensors, such as using laser scanning or CT scanning equipment to obtain the three-dimensional shape and its details of the accessory. These data include information such as the size, shape, material properties, and surface state of the accessory. Using these data for topological structure analysis, methods such as finite element analysis (FEA) are used to accurately model the geometric shape and internal and external structures of the accessory, generating the topological structure data of the accessory. During the analysis process, factors such as the load distribution and stress concentration of the accessory need to be considered in particular. The accessory corrosion simulation simulates the corrosion effect of different environmental factors (such as humidity, electrolyte concentration, etc.) on the accessory through a finite element model. The key parameters of the corrosion simulation include environmental temperature, humidity, electrolyte concentration, corrosion rate, etc. According to the simulation results, the accessory corrosion data obtained should include information such as the corrosion point distribution, corrosion depth, and corrosion rate. The finally formed corrosion data can reflect the corrosion area of the accessory in the image.

[0098] Step S2: Locate the accessory corrosion position based on the accessory corrosion data to obtain accessory corrosion position data, and perform image acquisition to obtain an accessory corrosion image; analyze the corrosion stress concentration points of the accessory corrosion image to obtain a corrosion stress concentration point image;

[0099] In this embodiment, the corrosion position is located. Using an image processing method, the image is threshold segmented based on the corrosion data of the fittings to distinguish the corrosion area from the non-corrosion area. At this time, a global threshold or an adaptive threshold algorithm can be used, and common methods such as the Otsu algorithm are used to determine the optimal threshold. After image segmentation processing, the corrosion position data of the fittings is obtained. This data describes the pixel positions of the corrosion area in the image. Next, image acquisition is performed to obtain a high-resolution corrosion image of the fittings for subsequent analysis. During this process, the analysis of the corrosion stress concentration points is completed through image edge detection. Using the Sobel operator or the Canny edge detection algorithm, the gradient of the image is calculated to extract the edge data of the corrosion stress concentration points in the image, and an image of the corrosion stress concentration points is obtained. The detection result reflects the stress concentration area of the corrosion area, usually manifested as an area with a large brightness difference.

[0100] Step S3: Classify the image of the corrosion stress concentration points to obtain an image of pitting corrosion stress concentration points and an image of uniform corrosion stress concentration points; classify the fitting types according to the topological structure data of industrial fittings to obtain fitting type data, and perform image acquisition of the fitting types to obtain fitting type images;

[0101] In this embodiment, the classification criteria need to be determined. For the image of pitting corrosion stress concentration points, it can be judged whether it is pitting corrosion by the corrosion depth, area, and morphological characteristics of the local area. Specifically, a corrosion depth threshold is set (for example, a depth greater than 1 mm can be regarded as pitting), and a morphological characteristic threshold is set (for example, the area of the corrosion area exceeds 0.1 cm 2 That is, it is classified as the pitting type). The image of uniform corrosion stress concentration points is judged by the difference from pitting corrosion. The surface corrosion depth of the area of uniform corrosion is relatively uniform, usually without obvious protrusions or depressions. The fitting type classification is judged according to the function, structure, or material characteristics of the fittings. For example, control fittings (such as motors and transmission components) can be distinguished from transmission fittings (such as pipes and transmission devices). The classification of fitting types depends on the geometric shape, application scenario, and functional requirements of the fittings. Through image acquisition of the fittings and combining the physical characteristic data of the fittings, fitting type images are obtained.

[0102] Step S4: Construct an image classification model based on the fitting type images and perform fitting identification on the industrial fitting data to obtain fitting identification data; predict the fitting life based on the image of pitting corrosion stress concentration points and the image of uniform corrosion stress concentration points for the fitting identification data to obtain fitting life prediction data;

[0103] In this embodiment, a feature extraction method based on traditional computer vision, such as HOG (Histogram of Oriented Gradients) or SIFT (Scale-Invariant Feature Transform), is used to extract the key feature points of the image. Then, classification algorithms such as K-means clustering or support vector machine (SVM) are used to classify the extracted features, so as to identify the type of the accessory. After being input into the image classification model, the accessory is recognized according to the image features of the accessory, and finally the accessory recognition data is obtained. For the prediction of the accessory life of the pitting stress concentration point image and the uniform corrosion stress concentration point image, it is necessary to combine the corrosion model and the fatigue analysis method. Specifically, based on the stress distribution, corrosion depth, and load conditions in the corrosion area, local and global fatigue life predictions are carried out through fatigue life prediction algorithms (such as Miner's linear cumulative damage theory). The local fatigue life prediction considers the influence of stress concentration in the corrosion area, calculates the fatigue damage degree using a formula, and obtains the local life prediction data. The global fatigue cumulative life prediction is based on the overall load and corrosion conditions of the accessory for analysis, calculates the cumulative fatigue damage of the accessory during long-term use, and thus obtains the global fatigue cumulative life prediction data.

[0104] Step S5: Based on the accessory life prediction data, perform accessory failure probability modeling on the accessory recognition data, so as to obtain accessory failure recognition data.

[0105] In this embodiment, a large amount of historical failure data of accessories in different corrosion and fatigue states needs to be collected to establish a failure prediction model. Using statistical methods, such as logistic regression or Bayesian network, failure probability modeling is carried out according to variables such as the corrosion degree of the accessory, life prediction data, and stress concentration situation. The specific parameters include the material properties of the accessory, working environment (such as temperature, humidity), stress distribution, etc. Through the regression analysis of historical data, the failure probability model of each type of accessory is obtained. Finally, through the calculation and parameter optimization of the model, accessory failure recognition data is obtained, which is specifically the probability of each accessory failing under the current working state, and further provides a basis for the maintenance and replacement decisions of the accessory.

[0106] Preferably, step S1 is specifically:

[0107] Step S11: Obtain industrial accessory data, and perform accessory hole feature extraction and accessory groove feature extraction according to the industrial accessory data, so as to obtain accessory hole data and accessory groove data;

[0108] In this embodiment, industrial accessory data is obtained. This data is collected by a 3D scanner, a laser scanner, or a CT scanning device to ensure accurate recording of the geometric shape, dimensions, and surface details of the accessory. Next, based on the geometric data of the industrial accessory, hole features of the accessory are extracted. The extraction of hole features can be achieved through geometric shape-based calculation methods, such as calculating the diameter, depth, position, and shape of the holes, and combining with CAD software to model the accessory surface to obtain the specific features of each hole. The position of the hole is usually represented by a point in a 3D coordinate system, and the dimensions of the hole (e.g., diameter and depth) are calculated through geometric analysis. The extraction of slot features is performed by analyzing the cutting lines on the accessory surface and using edge detection algorithms (such as Canny edge detection) to identify the shape, length, width, and depth of the slots. All the extracted hole data and slot data are represented in a three-dimensional coordinate system. The accessory hole data includes the number, position, and dimensions of the holes, and the accessory slot data includes the geometric parameters and position of the slots.

[0109] Step S12: Identify the assembly connection point channels based on the accessory hole data for the accessory slot data, so as to obtain the assembly connection point channel data;

[0110] In this embodiment, through geometric analysis algorithms (such as point cloud registration or spatial relationship analysis), based on the geometric positions of the accessory holes and slots, the spatial connection relationship between the holes and slots is identified. A threshold is set, such as a distance threshold of 0.5 mm. If the edge of the hole is less than this threshold from the edge of the slot, it is considered that there is a connection relationship between the hole and the slot. Then, based on these connection points, the assembly connection point channels of the accessory are identified through a spatial construction algorithm (such as Voronoi diagram construction). Each connection point represents the connection channel between different parts of the accessory, and the geometric information of these connection point channels includes the shape, length, width, etc. of the channel. Finally, the assembly connection point channel data identified by this method will be used for subsequent topology construction.

[0111] Step S13: Extract the connection node accessory features from the industrial accessory data, so as to obtain the connection node accessory data;

[0112] In this embodiment, it is necessary to extract the features of each connection node from the geometric data of the fittings. The connection nodes are usually the contact or connection points between multiple fittings, such as bolt holes, welding points, or other types of connection interfaces. By analyzing the geometric shape and connection method of the fittings, morphological operations (such as dilation, erosion, opening, and closing operations) are used to identify the connection nodes. For each connection node, calculate its position in three-dimensional space (represented by a spatial coordinate system), size (such as diameter, height, etc.), and functional attributes (such as whether it is a threaded hole, welding point, etc.). In addition, the detailed feature information of the connection nodes can be further extracted by combining the design drawings of the fittings and the mechanical connection requirements (such as the standard size of bolts or the strength requirements of welding points). Finally, the obtained connection node fitting data contains information such as the spatial position, type, and connection method of the nodes.

[0113] Step S14: Perform topology construction based on the connection node fittings and the assembled connection point channel data to obtain the industrial fitting topology structure data;

[0114] In this embodiment, it is necessary to clarify the construction criteria of the topology: the connection nodes and channels of each fitting should be modeled according to the relative position relationship between the fittings. By calculating the connection relationship between the nodes, the shortest path algorithm or Dijkstra algorithm in graph theory is used to determine the optimal connection path between the fittings. The topology structure data of the fittings will be presented in the form of a connection graph, where each node represents a fitting or a connection point, and each edge represents a connection channel. During topology construction, it is necessary to analyze the geometric parameters of each connection channel, such as the length, diameter, wall thickness, etc. of the connection channel, and consider its influence in combination with the working environment conditions (such as temperature, pressure, etc.). In this process, the necessary parameters include the distance between the nodes, the capacity of the channel, the mechanical properties of the material, etc. The constructed topology structure data should not only present the connection relationship between the fittings but also have the ability to describe the interaction state of the fittings.

[0115] Step S15: Perform fitting corrosion analysis based on the industrial fitting topology structure data to obtain the fitting corrosion data.

[0116] In this embodiment, the object and criteria of corrosion analysis are clarified. Corrosion analysis is usually based on the material properties of the fittings, the use environment (such as humidity, temperature, chemical environment, etc.), and the geometric morphology of the fittings. Through material corrosion models, such as the Nernst equation or the Grotian corrosion model, combined with the topological structure data of the fittings, the corrosion behavior of the fittings under different working environments is simulated. During the analysis process, parameters of the corrosion rate need to be set, and these parameters can be obtained based on experimental data. For example, the corrosion rate is set to 0.05 mm / year. The corrosion areas of each fitting are identified, and the stress distribution in the corrosion areas is simulated through finite element analysis (FEA). The data of the corrosion areas include corrosion depth, morphology, location, etc. The results of the corrosion analysis will form detailed fitting corrosion data. Finally, by analyzing the influence of the corrosion state and topological structure of the fittings, the corrosion damage degree of each fitting is obtained, providing a basis for subsequent maintenance, replacement, and life prediction.

[0117] Preferably, step S15 is specifically as follows:

[0118] Step S151: Obtain industrial environment humidity data using industrial environment sensors;

[0119] In this embodiment, the acquisition of industrial environment humidity data is carried out through environmental sensors installed around factories or production lines. The sensor selection should be based on environmental requirements, and high-precision humidity sensors, such as the HIH series of Honeywell or the SHT series of SENSIRION, are used. The measured humidity range is 0%RH to 100%RH, and the accuracy is ±2%RH. The sensors need to be installed in different areas, including high-humidity areas, low-humidity areas, etc., to ensure that the data covers the entire industrial environment. The data acquisition frequency is set to sample once per minute and is stored in a central database for subsequent analysis. The standardization process of the humidity data should be corrected based on the humidity value under normal temperature and normal humidity conditions to eliminate the influence of temperature changes on the humidity readings and ensure the accuracy of the data.

[0120] Step S152: Divide the industrial fitting topological structure data according to the industrial environment humidity data to obtain humidity-influenced industrial fitting data;

[0121] In this embodiment, the basis for classifying humidity-affected components is factors such as the material of the component, surface treatment, and working position. For example, for components vulnerable to corrosion, such as those made of metal materials like aluminum alloy, steel, or copper alloy, electrochemical corrosion is likely to occur in a high-humidity environment. The topological structure data of the components plays a crucial role in this step. By comparing the humidity exposure of each component, the humidity-affected components are distinguished from other components. Humidity-affected components are usually those located in areas with higher moisture or with more surface contact with moisture. The classification criterion is that when the humidity is greater than 70%, the component is classified as a humidity-affected component. The humidity exposure time of the component is also one of the classification criteria. The longer the humidity exposure time, the greater the likelihood of the component being affected by humidity.

[0122] Step S153: Collect industrial components based on the humidity-affected industrial component data to obtain humidity-affected industrial components.

[0123] In this embodiment, by conducting on-site sampling of the components, representative components are ensured to be selected from the humidity-affected areas. The collection tools are high-precision measuring instruments, such as portable temperature and humidity meters, laser scanners, and digital cameras. The selection of components is based on the intensity and time of humidity exposure. Components that have been in a high-humidity environment for a long time are preferentially collected. For example, components stored in a factory building near a water source or in an area with more steam gas release can be selected. The collected component data includes the geometric shape, surface condition, material, and installation position of the component. The data records are stored digitally for convenient subsequent analysis.

[0124] Step S154: Conduct an electrochemical corrosion simulation on the humidity-affected industrial components based on the industrial environment humidity data to obtain electrochemical corrosion industrial component data.

[0125] In this embodiment, an electrochemical corrosion model is used for analysis. The model is based on parameters such as the corrosion potential, corrosion rate, and concentration of the electrolyte of the material. First, corresponding environmental parameters need to be set for each humidity-affected component, such as temperature, humidity, and oxygen concentration in the air. In the electrochemical corrosion simulation, the corrosion potential is a key parameter, usually obtained through laboratory corrosion cell tests. The corrosion rate is calculated using the Nernst equation. Assuming that the corrosion rate of components with greater humidity influence is 0.02 mm / year and the electrolyte concentration is 0.1 mol / L. On this basis, the finite element analysis (FEA) method is used to simulate the distribution and process of the electrochemical corrosion process, and finally, electrochemical corrosion data, including the thickness and distribution of the corrosion layer, is obtained.

[0126] Step S156: Conduct an analysis of the formation of the electrolyte film on the electrochemical corrosion industrial component data based on the industrial environment humidity data to obtain electrolyte film data.

[0127] In this embodiment, the formation analysis of the electrolyte membrane mainly depends on environmental parameters such as solution concentration and temperature in the corrosion model. According to the given parameters in the corrosion model, such as electrolyte concentration, temperature, etc., the formation situation of the electrolyte membrane is deduced. Assuming that the formation thickness of the electrolyte membrane is affected by factors such as humidity and temperature, when the humidity is 75%, the formation thickness of the electrolyte membrane can be set to 0.1 mm. The conductivity, hardness and other characteristics of the electrolyte membrane are measured through experiments, and the properties of the membrane layer are determined using the conductivity calculation model. In addition, a scanning electron microscope (SEM) is used to observe the structure of the membrane layer, further verifying the distribution and characteristics of the membrane layer, and finally obtaining the formation data of the electrolyte membrane.

[0128] Step S157: Draw a current density distribution map based on the electrolyte membrane data to obtain a current density distribution map.

[0129] In this embodiment, during the electrochemical corrosion process, the distribution of the current density is crucial for the corrosion process. The current density can be calculated by Ohm's law in combination with the conductivity of the electrolyte membrane. Assuming that at a certain specific humidity, the current density is 0.01 A / cm 2 , in this current density, the area where corrosion occurs is the high current density area of the membrane layer. According to the change of the current density in different regions, a numerical simulation tool, such as COMSOL Multiphysics, is used to draw a two-dimensional distribution map of the current density. The coordinate axes of the current density distribution map represent the spatial position of the fitting, and each point represents the current density value at that position. The area with a higher current density is usually the place where corrosion is most severe.

[0130] Step S158: Identify the high density area based on the current density distribution map to obtain high current density area data.

[0131] In this embodiment, by setting a current density threshold, such as 0.02 A / cm 2 , the areas where the current density is greater than this threshold are identified. These areas are the high density areas of the current density, usually corresponding to more severe corrosion on the surface of the fitting. In this step, an image processing algorithm, such as the region growing algorithm, is used to identify the high density current areas in the figure. Based on the image processing results, the area with the highest current density is marked, and the coordinates and size of each high density area are recorded. This data provides an important basis for subsequent corrosion analysis, indicating that these areas are the places where the corrosion of the fitting is most concentrated and significant.

[0132] Step S159: Conduct a corrosion analysis of the humidity-affected industrial fitting based on the high current density area data to obtain fitting corrosion data.

[0133] In this embodiment, by combining the data of the high current density region, a corrosion rate calculation model is used to predict the influence range of corrosion. The corrosion rate is relatively large in the high current density region. Assume that the corrosion rate in the high current density region is 0.05 mm / year. Using the finite element analysis method and combining the actual current density data, the evolution of the corrosion process is simulated to obtain corrosion data. The corrosion data includes corrosion depth, expansion range, corrosion morphology, etc. Finally, the corrosion state and degree of the fitting under the influence of humidity are obtained. These data provide a reference basis for subsequent fitting maintenance and replacement.

[0134] Preferably, step S2 is specifically as follows:

[0135] Step S21: Locate the corrosion position of the fitting based on the fitting corrosion data, so as to obtain the fitting corrosion position data, and perform image acquisition to obtain the fitting corrosion image;

[0136] In this embodiment, based on the previously obtained corrosion data, the surface of the fitting is accurately located. By using a high-precision laser scanner or digital image measurement device (such as the LS-9000 series laser scanner of KEYENCE or the high-resolution digital camera of Canon), the surface of the fitting is scanned to obtain the three-dimensional point cloud data of the fitting. On this basis, combining the corrosion area indicated in the corrosion data, CAD software such as AutoCAD or SolidWorks is used to model the surface of the fitting. By fusing the corrosion data with the point cloud data, the corrosion position is determined, and the corrosion position data is output. Then, the corrosion area of the fitting is imaged by a high-resolution camera or an endoscope. The resolution of the image acquisition is set to at least 20 million pixels, and the exposure time and light source should be adjusted according to the environmental conditions to ensure clear corrosion images. These images will be used for further defect analysis and position identification.

[0137] Step S22: Analyze the pore defects in the fitting corrosion image to obtain the pore defect data;

[0138] In this embodiment, the collected corrosion images are preprocessed, including denoising and enhancing the contrast. The image preprocessing is performed using image processing software such as the Image Processing Toolbox in MATLAB. Median filtering (window size 3x3) is used to remove noise, and histogram equalization is used to enhance the contrast of the images. The processed images are then subjected to an edge detection-based algorithm (such as Canny edge detection) to extract the stomatal defect regions. The recognition threshold for stomatal defects is set such that regions with a contrast change amplitude exceeding a certain standard (such as 20%) are regarded as stomatal defects. The positions of the stomata are determined using the pixel coordinates of the images, and parameters such as the area, depth, and shape of the stomata are recorded to form stomatal defect data. For the stomatal defect regions, the region growing method can be used to further refine the recognition edges to ensure the accuracy of the defect regions.

[0139] Step S23: Analyze the crack defects in the accessory corrosion image to obtain crack defect data;

[0140] In this embodiment, the same preprocessing operations as in step S22 are performed on the collected corrosion images. Then, an image analysis algorithm is used to extract the cracks in the images. Crack defects usually exhibit slender linear features and show different brightness and contrast in the images. The Sobel filter is used for image edge detection, and the threshold for edge detection is set such that regions with an intensity change greater than 15% are considered crack regions. Then, morphological operations such as dilation and erosion are used to further clarify the crack edges. Finally, the length, width, depth, etc. of the cracks are extracted through region property analysis. The recording of crack data includes the starting point, ending point, and total length of the crack, and the crack position is determined through coordinate positioning to form crack defect data.

[0141] Step S24: Identify the positions of the accessory corrosion images based on the stomatal defect data to obtain stomatal defect position data;

[0142] In this embodiment, based on the stomatal defect data in step S22, the positions of the stomata in the corrosion images are located through an image processing algorithm. The matching method based on image contours is used for the identification of stomatal positions. The recognition criteria for stomata are set as having a diameter greater than 0.5 mm and an approximately circular shape. The Hough transform method is used for circular detection. According to the detection results, the specific positions of each stomata are marked by coordinates and stored as stomatal defect position data. During the stomatal position recognition process, the threshold parameters (such as diameter and shape tolerance) can be adjusted through experiments to ensure the accurate recognition of stomatal positions. The output result of this step is the stomatal position data, including the coordinate information and related geometric features of each stomata.

[0143] Step S25: Identify the position of the crack defect on the corrosion image of the fitting based on the crack defect data, so as to obtain the crack defect position data;

[0144] In this embodiment, based on the crack defect data obtained in step S23, the precise positioning of the crack position is carried out. The crack identification uses a crack tracking algorithm based on morphological operations. In the corrosion image, the crack presents a linear or curved characteristic. Therefore, a crack tracking method based on edge information is adopted, combined with the geometric features of the crack, such as the crack width is less than 0.3 mm and has continuity. The crack path is smoothed by a curve fitting algorithm (such as B-spline curve fitting), and through threshold screening, the maximum length of the crack is set to 5 mm and the maximum width is 0.3 mm to locate the accurate position of the crack. Finally, the starting and ending points of the crack are recorded, and the crack defect position data, including the position coordinates and morphological information of the crack, are output.

[0145] Step S26: Perform an intersection operation on the corrosion stress concentration points based on the pore defect position data and the crack defect position data, so as to obtain the corrosion stress concentration point data, and perform image acquisition to obtain the corrosion stress concentration point image.

[0146] In this embodiment, the purpose of the intersection operation is to identify the regions where pores and cracks overlap or are close, and these regions are often the key regions of stress concentration. The identification of the intersection region is realized by an image processing algorithm. The image superposition technology is adopted to superimpose the coordinates of the pore position and the crack position, and calculate the area and coordinate position of the overlapping region. The set criterion is that when the area of the overlapping region is greater than 0.1 mm 2 , and the distance between the two defects is less than 0.5 mm, it is regarded as a stress concentration point. After the position calibration of the intersection region, the region is imaged by a high-resolution camera. When imaging, a special light source, such as fiber optic illumination, is used to ensure a clear local image. Finally, the obtained corrosion stress concentration point image can be used for subsequent structural analysis and performance evaluation.

[0147] Preferably, step S22 is specifically:

[0148] Step S221: Denoise the corrosion image of the fitting to obtain a denoised corrosion image of the fitting;

[0149] In this embodiment, an image processing software is used to denoise the collected corrosion images. The noise in the images usually appears as randomly distributed gray value deviations, and the median filtering method can be used to reduce the noise. The window size of the median filtering is set to 3x3 pixels. In the neighborhood of each pixel, the median value is selected as the new value of this pixel, so as to eliminate the noise influence in a small range. In addition, according to the characteristics of the image, the noise threshold can be set to ±10% of the gray value range of the image (for example, between the gray value range of 0 to 255, the threshold is set to ±25), to avoid over-denoising affecting the details of the image. The denoised fitting corrosion image provides a clearer data input for the next step of processing.

[0150] Step S222: Perform binarization conversion on the denoised fitting corrosion image to obtain a binarized fitting corrosion image;

[0151] In this embodiment, a suitable threshold is selected for binarization processing of the image. The threshold selection can use the Otsu algorithm to automatically calculate the global optimal threshold of this image. The Otsu algorithm selects an optimal threshold by maximizing the between-class variance. During the implementation process, first divide the gray value range of the denoised image (for example, 0 to 255) into two categories, calculate the variance of each category, and then select a threshold to maximize the between-class variance. After setting this threshold, all pixel values in the image greater than the threshold are set to white (255), and the part less than the threshold is set to black (0). In this way, a binarized fitting corrosion image is obtained, ready to enter the next step of edge detection.

[0152] Step S223: Obtain historical pore edge data, and perform pore edge detection on the binarized fitting corrosion image to obtain a pore edge image;

[0153] In this embodiment, historical pore edge data is obtained. These data can be from previous detection images, obtained through manual annotation or automated detection. The detection of the pore edge is realized by the Canny edge detection algorithm. First, use a Gaussian filter to smooth the binarized image to eliminate low-frequency noise. The standard deviation of the Gaussian filter is set to 1.5 to balance the smoothing effect and the retention of edge information. Then, use the Canny algorithm for edge detection, set the low threshold to 50 and the high threshold to 150 to ensure that only strong edges are retained. Through this method, the edge information of the pores in the image is extracted to generate a pore edge image for subsequent roundness detection.

[0154] Step S224: Perform roundness detection on the pore edge image to obtain roundness data, and perform statistics on the roundness data to obtain high-roundness data;

[0155] In this embodiment, each connected region in the stoma edge image is extracted through the contour analysis algorithm, and the boundary of each region is calculated. To accurately calculate the roundness of the stoma, the perimeter and area of each stoma need to be calculated. The roundness calculation formula is: Roundness = (4π × Area) / Perimeter 2 . According to this formula, the closer the roundness value is to 1, the closer the region is to a circle. The roundness threshold is set to 0.8, that is, when the roundness is greater than 0.8, the stoma is considered to be close to a circle. Then, the roundness of all detected stomata is statistically analyzed, the roundness value of each stoma is recorded, and high-roundness data is calculated according to the set standard to ensure the accurate identification and statistics of circular stomata.

[0156] Step S225: Locate the stoma centroid of the binarized fitting corrosion image according to the high-roundness data, so as to obtain stoma centroid data;

[0157] In this embodiment, the centroid calculation is based on stomata with a roundness greater than 0.8. The centroid is calculated using the second moment method. Within the contour area of each stoma, the centroid coordinates are calculated using the pixel values of the image. The formula is: Centroid X = Σ(xi × Ii) / ΣIi, Centroid Y = Σ(yi × Ii) / ΣIi, where xi and yi are the coordinates of each pixel point, and Ii is the gray value of the corresponding pixel point. Through this method, the centroid position of each stoma can be accurately determined. Finally, the centroid data of all stomata is recorded as stoma centroid data for subsequent heat map generation and depth evaluation.

[0158] Step S226: Generate a heat map according to the stoma centroid data, so as to obtain a stoma heat map;

[0159] In this embodiment, the centroid coordinates need to be mapped into the two-dimensional space of the image. The generation of the heat map is based on the density distribution of the stoma centroid. The Gaussian kernel density estimation method is used to calculate the density value around each point. The kernel function adopts the standard normal distribution, and its width (standard deviation) is set to 0.5 mm, which can smoothly estimate the density. Then, according to the calculated density value, a heat value is assigned to each position in the image, and the heat value range is from 0 (low density area) to 255 (high density area). Finally, the heat map is presented as a color image, and the color depth is proportional to the density, so as to intuitively display the distribution of stomata.

[0160] Step S227: Evaluate the number of stomata in the stoma heat map, so as to obtain stoma number data;

[0161] In this embodiment, it is necessary to perform threshold segmentation on the generated pore heat map, set a pore density threshold, for example, when the thermal image pixel value exceeds 150, it is defined as the pore-dense area. Through this threshold, the number of pores in the heat map is determined. In specific operations, the region labeling algorithm (such as the connected component labeling method) can be used to label the high-density areas in the heat map, regard each pore as an independent region, and count the number of these regions. Finally, the number data of the pores is recorded for subsequent defect integration and depth evaluation.

[0162] Step S228: Perform pore depth evaluation on the binarized fitting corrosion image according to the pore centroid data, so as to obtain pore depth data;

[0163] In this embodiment, the pore centroid data is used for the preliminary evaluation of the pore depth. According to the position of the pores, the three-dimensional scanning data or the microscope depth detection device is used to measure the depth of the pores. The depth is measured using an optical microscope or a laser depth detector, and the depth measurement accuracy is set to 0.01 mm. According to the position of the centroid, using the three-dimensional reconstruction method and combining the geometric information of the fitting surface for depth estimation, and record the depth data of each pore. These data provide a basis for the next step of pore defect integration.

[0164] Step S229: Perform pore defect integration according to the pore number data and the pore depth data, so as to obtain pore defect data.

[0165] In this embodiment, the integration process includes combining the number and depth information of each pore, and calculating the comprehensive defect degree of each pore by the weighted average method. For each pore, the depth weight is set to 0.5 and the number weight is set to 0.5. According to these weights, the defect data of each pore is comprehensively calculated. Finally, a comprehensive defect data set including the pore number and depth is obtained, and this data set can be used for subsequent fitting quality evaluation and defect analysis.

[0166] Preferably, step S228 is specifically:

[0167] Perform region positioning on the binarized fitting corrosion image according to the pore centroid data, so as to obtain fitting pore corrosion region data;

[0168] In this embodiment, based on the coordinate information of the stomatal centroid, the area of each stomatal is located through an image processing algorithm. The area location can be achieved by calculating the boundary of the circular area around the centroid. A fixed radius value (such as 5 mm) is set, and a circular area is drawn around the centroid of each stomatal. This radius value can be adjusted according to the size of the accessory and the size of the stomatal. For the pixels within each circular area, check whether they are white (indicating the corroded area). If so, classify them as the corroded area of the corresponding stomatal. In this way, the corroded area of each stomatal in the accessory is determined, and the corresponding accessory stomatal corroded area data is generated, which will be used for subsequent 3D reconstruction.

[0169] Based on the accessory stomatal corroded area data, 3D reconstruction of the accessory is carried out to obtain a 3D stomatal corroded accessory;

[0170] In this embodiment, the stomatal corroded area data is converted into point cloud data in a 3D coordinate system. This process uses laser scanning technology or optical scanning technology to obtain the 3D point cloud data of the accessory surface, and the accuracy of the point cloud data is set to 0.01 mm. By matching the stomatal corroded area data with the point cloud data, the specific position and corrosion morphology of each stomatal are determined. Next, a 3D reconstruction algorithm (such as voxelization method or meshing method) is used to process the point cloud data to generate a complete 3D model. For the reconstruction of the corroded area, a certain offset can be given during the reconstruction process to ensure that the morphology and depth of the corroded area can be accurately reflected in the 3D model, and finally a 3D stomatal corroded accessory is generated.

[0171] Lighting mapping simulation is performed on the 3D stomatal corroded accessory to obtain lighting mapping data;

[0172] In this embodiment, based on the 3D model of the accessory, a lighting model is established. The model includes the geometric information of the accessory surface and the parameters of the environmental light source. The setting of the environmental light source includes the intensity, direction, and angle of the incident light source. Assume that the intensity of the light source is set to 500 lux and the direction is 45 degrees. Then, using the radiative transfer equation in the lighting model and combining the relationship between the surface normal vector of the accessory and the light source direction, the lighting intensity of each point is calculated. The calculation formula for the lighting intensity is: I = I_0 * cos(θ), where I_0 is the incident light intensity and θ is the angle between the light source and the surface normal. Through this process, the lighting intensity data of each surface point is generated, and finally the complete lighting mapping data is obtained.

[0173] Lighting gradient calculation is performed on the lighting mapping data to obtain lighting gradient data;

[0174] In this embodiment, the illumination gradient is determined by calculating the change of illumination intensity in space. The illumination gradient refers to the rate of change of illumination intensity within a local area, which is usually achieved by calculating the derivatives of illumination intensity in the x, y, and z directions. The calculation process can be processed using a gradient operator, and the formula is: where I is the illumination intensity, and x, y, and z are spatial coordinates. In actual operation, by performing smoothing processing on the illumination mapping data, setting the smoothing window size to 5x5 pixels to remove noise interference, and then calculating the gradient value. The illumination gradient value can reflect the details of illumination changes, especially the changes in the stoma edge area, which is crucial for subsequent stoma depth calculation.

[0175] Based on the illumination gradient data, the stoma depth is calculated to obtain the stoma depth data.

[0176] In this embodiment, the illumination gradient data is compared with a known illumination intensity model. The stoma depth calculation is based on the change law of illumination intensity in the stoma area. Since the stoma forms a depression on the surface, the illumination intensity changes significantly at the stoma edge. According to this change, by comparing with the known depth-illumination intensity relationship, the initial value of depth calculation is set. Specifically, by selecting a set of standard depths and corresponding illumination intensity change values, the depth of each stoma area is calculated using the interpolation method. The interpolation method can adopt linear interpolation or Lagrange interpolation. The accuracy of the depth value is set to 0.01 mm. The calculated stoma depth data reflects the position and depth characteristics of the stoma on the fitting surface, and is used for subsequent quality inspection and corrosion analysis.

[0177] Preferably, step S23 is specifically as follows:

[0178] Step S231: Convert the fitting corrosion image to grayscale to obtain a grayscale fitting corrosion image;

[0179] In this embodiment, the original color fitting corrosion image is separated into channels, and the red, green, and blue color channels are extracted respectively. Then, the weighted average method is used to perform grayscale processing on these three color channels. The specific weighted formula is: grayscale value = 0.2989 × red channel + 0.5870 × green channel + 0.1140 × blue channel. Through this formula, the grayscale value of each pixel is obtained. Finally, a grayscale fitting corrosion image is generated, and the grayscale value range of each pixel in this image is between 0 and 255, representing the grayscale level from black to white. This image will be used as the basis for subsequent processing.

[0180] Step S232: Calculate the pixel gradient based on the grayscale fitting corrosion image to obtain the pixel gradient data;

[0181] In this embodiment, the Sobel operator is applied to calculate the gradient of the grayscale image. The Sobel operator calculates the gradient values of each pixel point in the horizontal and vertical directions by performing convolution operations on the grayscale image in the horizontal and vertical directions. Specifically, the horizontal filter of the Sobel operator is [1, 0, -1], and the vertical filter is [1, 2, 1]. By convolving the grayscale image with these two filters, the gradient values of each pixel point in the x and y directions are obtained. Next, calculate the gradient magnitude using the formula: Gradient Magnitude = √(Gx 2 + Gy 2 ), where Gx and Gy are the gradient values in the horizontal and vertical directions respectively. The calculation result will generate pixel gradient data, which contains the gradient information of each pixel point in the image.

[0182] Step S233: Draw a gradient magnitude map based on the pixel gradient data to obtain the gradient magnitude map;

[0183] In this embodiment, based on the pixel gradient data obtained in step S232, image processing software or programming languages (such as MATLAB or Python) are used to visualize the gradient magnitude of each pixel. By mapping the gradient magnitude value of each pixel to the pixel grayscale value of the image, a new image is generated, representing the gradient magnitude at each position. To enhance the image contrast, the histogram equalization algorithm can be applied to normalize the pixel gradient magnitude values between 0 and 255. The gradient magnitude map generated by this process can clearly show the regions with gradient changes in the image, especially the edges and cracks.

[0184] Step S234: Set a high crack threshold and a low crack threshold based on the grayscale fitting corrosion image to obtain high crack threshold data and low crack threshold data;

[0185] In this embodiment, by statistically analyzing the gradient magnitude distribution of the known corrosion regions in the image, the high threshold and the low threshold are determined. The high threshold is set to the 95th percentile of the gradient magnitude in the image, and the low threshold is set to 60% of the high threshold. The specific numerical setting is as follows: assuming that the gradient magnitude distribution in the grayscale image ranges from 0 to 255, if the 95th percentile is 200, then the high threshold is 200 and the low threshold is 120. By setting these thresholds, cracks can be effectively distinguished from other corrosion regions in subsequent edge detection. The high threshold is used to initially detect significant crack edges, while the low threshold is used to further refine the detection boundary.

[0186] Step S235: Perform edge detection on the gradient magnitude map according to the crack high threshold data and the crack low threshold data. If the pixel gradient is greater than the crack high threshold data, obtain the first pixel edge point data; if the pixel gradient is between the crack high threshold data and the crack low threshold data, further judge. If the pixel gradient is connected to the first pixel edge point data, obtain the second pixel edge point data;

[0187] In this embodiment, the Canny edge detection algorithm is used. By analyzing the pixel gradients of the gradient magnitude map, the pixels with gradient magnitudes greater than the crack high threshold are identified and marked as the first pixel edge point data. For the pixels with pixel gradient values between the crack high threshold and the crack low threshold, by further analyzing the gradient values of their neighboring pixels, it is judged whether they are connected to the first pixel edge point data. If so, the pixel is marked as the second pixel edge point data. At this time, a connection threshold (for example, set to an adjacent connection range of 5 pixels) is set to judge whether the pixels are connected. The finally obtained first pixel edge points and second pixel edge point data are used to identify the crack edge region in the image.

[0188] Step S236: Integrate the first pixel edge point data and the second pixel edge point data to obtain the crack pixel edge point data;

[0189] In this embodiment, the first pixel edge points and the second pixel edge point data obtained in step S235 are merged to generate a complete set of crack edge points. To ensure the coherence of the edges, a connection algorithm (such as an edge point clustering algorithm) is used to connect the discrete edge points to form a complete crack edge contour. This process analyzes the distances between adjacent pixel points, sets the connection threshold to 1 pixel, and ensures that adjacent pixel points can be effectively connected into a continuous crack edge. Finally, the obtained crack pixel edge point data contains complete crack edge information and can be used for subsequent measurements such as length and width.

[0190] Step S237: Calculate the crack length of the gray-scale fitting corrosion image according to the crack pixel edge point data to obtain the crack length data;

[0191] In this embodiment, by analyzing the crack edge point data, the Hough transform or directly calculating the Euclidean distance between the edge points is used to measure the length of the crack. Assuming that the crack edge point data forms a continuous line segment, by calculating the distances between every two adjacent points and adding these distances together, the total length of the crack is obtained. The specific calculation formula is: crack length = Σ(√((x_i + 1 - x_i) 2 +(y_i + 1 - y_i) 2), where (x_i, y_i) and (x_i + 1, y_i + 1) are the coordinates of adjacent edge points respectively. This process can accurately calculate the total length of the crack.

[0192] Step S238: Calculate the crack width based on the crack pixel edge point data for the gray-scale fitting corrosion image, so as to obtain the crack width data;

[0193] In this embodiment, two parallel boundaries at the crack edge are selected, and the crack width is calculated by measuring the pixel distance between these two boundaries. The specific steps are as follows: Select a crack boundary line, sample along the normal direction at multiple positions of the crack, and measure the distance to the other boundary line. Set the interval of the measurement points to 2 mm, calculate the width at each position, and finally calculate the average width of the crack. At this time, the result of the width calculation is in pixels. When converting to the actual size, it needs to be calculated in combination with the resolution of the image.

[0194] Step S239: Evaluate the crack area based on the crack length data and the crack width data, so as to obtain the crack area data, and perform crack defect visualization on the gray-scale fitting corrosion image according to the crack area data, so as to obtain the crack defect data.

[0195] In this embodiment, the crack area is obtained by multiplying the crack length by the width. Suppose the crack length is 50 mm and the width is 5 mm, then the crack area is 250 mm 2 . According to the crack area data, perform visualization processing on the crack area, and use a heat map or a shadow map to display the crack distribution. Specifically, by mapping the pixel values of the crack area to the color value range of the image, a crack defect data map is generated, thus realizing the visual display of the crack defect.

[0196] Preferably, step S3 is specifically as follows:

[0197] Step S31: Identify the corrosion morphology of the corrosion stress concentration point image, so as to obtain the corrosion morphology data;

[0198] In this embodiment, the binarization method is used to convert the image into a black-and-white image. The set threshold is 128 of the gray value. The pixel points with gray values lower than 128 are converted to black, and the pixel points with gray values higher than 128 are converted to white. Then, using the erosion operation in morphological analysis, the image is eroded. The erosion structuring element is selected as a rectangular kernel of 3x3 pixels, aiming to remove the fine noise in the image and highlight the erosion morphology. Next, the morphological reconstruction technique is used to repair the eroded image to ensure that the erosion morphology remains consistent. At this time, an edge detection algorithm (such as the Canny operator) is applied to identify the erosion morphology, and the identified area represents the corrosion stress concentration point. Finally, through the extraction of the geometric features of the erosion morphology, the obtained erosion morphology data describes the shape, size, and distribution characteristics of the corrosion points.

[0199] Step S32: Classify the corrosion stress concentration point image according to the corrosion morphology data, so as to obtain the pitting corrosion stress concentration point image and the uniform corrosion stress concentration point image;

[0200] In this embodiment, for the corrosion morphology data, by analyzing the geometric features of the corrosion morphology (such as area, roundness, aspect ratio, etc.), the corrosion morphology is divided into two categories using the set classification criteria. The criterion for setting the pitting corrosion stress concentration point is: when the area of the corrosion morphology is less than 5 mm 2 and the roundness is greater than 0.8, it is classified as the pitting corrosion type. The criterion for setting the uniform corrosion stress concentration point is: when the area of the corrosion morphology is greater than 5 mm 2 and the aspect ratio is close to 1, it is classified as the uniform corrosion type. For each corrosion stress concentration point, calculate its geometric features and compare them with the above criteria. According to the classification results, the corrosion point image is respectively marked as the pitting corrosion stress concentration point image or the uniform corrosion stress concentration point image.

[0201] Step S33: Classify the industrial accessory types according to the industrial accessory topology structure data, so as to obtain the control accessories and the transmission accessories;

[0202] In this embodiment, the topology structure data of industrial accessories is collected, and this data includes information such as the geometric shape, size, and functional positioning of the accessories. According to the functional characteristics of the accessories, the criteria for classifying the accessory types are set. The criterion for the control accessories is: the accessories have functions of adjustment, restriction, or support, and are usually the core parts that determine the system performance and stability, such as control valves, sensors, etc. The criterion for the transmission accessories is: the accessories are mainly used to transmit power, fluid, or signals, have a large surface area and a small structural complexity, such as pipelines, transmission cables, etc. By analyzing the topology structure data of the accessories and classifying the accessories according to the above criteria, the accessories are finally classified into control accessories and transmission accessories.

[0203] Step S34: Collect high-resolution images of the control components to obtain high-resolution control component images;

[0204] In this embodiment, select the device for high-resolution image collection, such as an industrial camera with a resolution of 50 million pixels. Set the focal length of the camera to 50 mm to ensure clear image focus, and mount the camera on a fixed bracket to avoid vibration or deviation during image collection. The control components are placed in a standardized lighting environment, and the light source is a uniform white LED lamp to ensure sufficient light on the surface of the components without shadows. During image collection, select an exposure time of 0.02 seconds and set the ISO value to 200 to ensure clear detailed images. During the shooting process, use a tripod to stabilize the camera to ensure that the shooting angle and position of each image are consistent, avoiding affecting the image quality due to changes in the viewing angle. The finally obtained high-resolution images can clearly show each detail and the corrosion area of the control components.

[0205] Step S35: Perform infrared imaging collection on the transmission components to obtain infrared imaging of the transmission components;

[0206] In this embodiment, use an infrared thermal imager (such as FLIR T640 with a resolution of 640x480) for imaging, set the temperature range of the thermal imager to -20°C to 100°C to adapt to the operating environment temperature of most transmission components. Align the thermal imager with the transmission components and ensure no external interference during imaging, such as fans or high-temperature objects. During infrared imaging, select appropriate focal length and field of view angle to ensure obtaining the temperature distribution map of the surface of the transmission components. During imaging, set the temperature threshold to 40°C to highlight the abnormal areas (such as hot spots caused by overheating, friction, etc.) due to thermal effects. During the imaging process, keep the imaging distance constant to avoid affecting the image quality due to changes in the viewing angle. Finally, the temperature distribution map of the surface of the components can be obtained through the infrared imaging data, revealing the temperature changes and thermal load of the transmission components.

[0207] Step S36: Integrate the component type images based on the high-resolution control component images and the infrared imaging of the transmission components to obtain the component type images.

[0208] In this embodiment, the high-resolution control component image is registered with the infrared imaging image of the transmission component. During the registration process, an algorithm based on feature point matching (such as the SIFT algorithm) is used to ensure accurate matching of corresponding points in the two images. After registration, image fusion is performed, and the feature information of the two images is superimposed. The specific method is as follows: The color information of the control component image is combined with the temperature distribution information of the transmission component infrared image, and the weighted average method is used to fuse the images to ensure that the two types of information can be clearly displayed on the same image. The finally generated component type image contains both the detailed information in the high-resolution image and the temperature distribution information in the infrared imaging, and can comprehensively reflect the state and working characteristics of the component.

[0209] Preferably, step S4 is specifically as follows:

[0210] Step S41: Construct an image classification model based on the component type image, input the industrial component data into the image classification model, and perform component image classification to obtain a component classification image;

[0211] In this embodiment, a labeled component type image dataset is collected, including the high-resolution control component images and infrared imaging images integrated in step S36. Each image has been calibrated with its component type, such as a control component or a transmission component. Then, a convolutional neural network (CNN) is used as the basic structure of the image classification model. The number of layers of the CNN model is designed to be 8 layers. The first four layers are convolutional layers, using a 3x3 convolutional kernel, a stride of 1, and a ReLU activation function is connected after the convolutional operation. The pooling layer uses max pooling, with a pooling window size of 2x2 and a stride of 2. The last fully connected layer has two layers, with 512 nodes in the first layer and 2 nodes in the second layer, and the classification result of the component is output. The data preprocessing used during model training includes image normalization, scaling the pixel values of each image to between 0 and 1. During the training process, the Adam optimizer is used, the learning rate is set to 0.001, and the cross-entropy loss is selected as the loss function. The input industrial component data includes component images and their labeled tags (such as "control component", "transmission component"). A batch size of 32 is used during the training process, and 50 iterations are performed. Finally, the component images are classified into different categories through the model output results to generate component classification images.

[0212] Step S42: Identify the component based on the component classification image to obtain component identification data;

[0213] In this embodiment, edge detection is performed on the classified accessory images. The Canny operator is used to process the accessory images, and the edge detection thresholds are set to 100 and 200 to ensure that important edges in the images can be accurately extracted. Then, morphological operations (dilation and erosion) are used to process the extracted edges to remove noise and discontinuous edges in the images. Next, a contour detection algorithm is used to extract the contours of the accessories in the processed images, and the minimum area threshold for contour detection is set to 200 pixels to eliminate noise contours with too small areas. The extracted contours identify the accessories as different types, specifically including control accessories and transmission accessories. The accessory identification data will include the type information of each accessory and its spatial position and size in the image. During this process, the recognition results of the accessory images are labeled, and the classification and identification data of each accessory are output.

[0214] Step S43: Perform local fatigue life prediction on the accessory identification data based on the pitting stress concentration point image, so as to obtain local fatigue life prediction data;

[0215] In this embodiment, the position and size data of the corrosion points are extracted from the pitting stress concentration point image. The extraction of the corrosion points is performed through thresholding of the image, and the threshold is set to 100 to ensure that all corrosion regions in the image are identified. For the extracted corrosion point data, the local stress of each point is further calculated through geometric features (such as area, shape, etc.) in the image. Then, a fatigue life prediction model based on the stress-strain relationship is applied, and the Palmgren-Miner method is used for local fatigue life prediction. At this time, the input of the model is the local stress data and its morphological features of each point, and the output is the local fatigue life prediction value. During the prediction process, the stress threshold is set to 10 MPa to ensure that the fatigue life assessment is only performed on the corrosion regions exceeding this threshold. According to the different corrosion stress regions, the local fatigue life data are calculated to obtain the fatigue life prediction corresponding to each corrosion point.

[0216] Step S44: Perform global fatigue cumulative life prediction on the accessory identification data based on the uniform corrosion stress concentration point image, so as to obtain global fatigue cumulative life prediction data;

[0217] In this embodiment, the global stress distribution map of the corrosion region is extracted from the uniform corrosion stress concentration point image, and morphological operations (such as erosion and dilation) are used to further process the image to remove areas smaller than the specified area (the set area threshold is 5 mm 2Noise points. Then, using the global fatigue accumulation model of the material, cumulative fatigue calculation is performed based on the stress data and corrosion morphology of each corrosion area. The global fatigue cumulative life prediction adopts Miner's rule, which weights and sums the local fatigue data of each corrosion point according to the relationship between stress and fatigue cycles to obtain the global fatigue cumulative life of the fitting. In the global fatigue cumulative life prediction, the damage coefficient of each fatigue cycle is set to 0.1, and the number of cycles is set to 10 6 times based on the working environment of the industrial fitting. Finally, the global fatigue cumulative life data is output, representing the fatigue life of the fitting under the overall corrosion state.

[0218] Step S45: Integrate the local fatigue life prediction data and the global fatigue cumulative life prediction data to obtain the fitting life prediction data.

[0219] In this embodiment, the local fatigue life data and the global fatigue cumulative life data obtained in steps S43 and S44 are collected. The local fatigue life data includes the local fatigue life prediction values of each corrosion point, and the global fatigue cumulative life data represents the fatigue state of the entire fitting. When combining the two, first, the local fatigue life prediction data is weighted, and the weight value is set to 0.7 to emphasize the fatigue influence of the local area; the weight of the global fatigue cumulative life prediction data is set to 0.3 to highlight the influence of the overall corrosion state. The total fatigue life prediction value of the fitting is calculated by the weighted average method. This integration process uses weighting coefficients to ensure that the combination of local and global fatigue lives can accurately reflect the overall fatigue condition of the fitting. Finally, the fitting life prediction data obtained by calculation will be output for evaluating the overall service life of the fitting.

[0220] Preferably, this specification also provides a fitting recognition system based on an image classification model for performing a fitting recognition method based on an image classification model as described above. The fitting recognition system based on an image classification model includes:

[0221] A fitting corrosion simulation module, configured to obtain industrial fitting data, perform topological structure analysis based on the industrial fitting data to obtain industrial fitting topological structure data; perform fitting corrosion simulation based on the industrial fitting topological structure data to obtain fitting corrosion data;

[0222] A corrosion stress concentration point analysis module, configured to locate the fitting corrosion position based on the fitting corrosion data to obtain fitting corrosion position data, and perform image acquisition to obtain a fitting corrosion image; perform corrosion stress concentration point analysis on the fitting corrosion image to obtain a corrosion stress concentration point image;

[0223] The accessory type classification module is used to classify the images of corrosion stress concentration points, so as to obtain pitting corrosion stress concentration point images and uniform corrosion stress concentration point images; classify the accessory types according to the industrial accessory topological structure data, so as to obtain accessory type data, and collect accessory type images, so as to obtain accessory type images;

[0224] The accessory life prediction module is used to construct an image classification model based on the accessory type images, and identify the industrial accessories according to the accessory data, so as to obtain accessory identification data; predict the accessory life according to the pitting corrosion stress concentration point images and the uniform corrosion stress concentration point images for the accessory identification data, so as to obtain accessory life prediction data;

[0225] The accessory fault identification module is used to model the accessory fault probability for the accessory identification data according to the accessory life prediction data, so as to obtain accessory fault identification data.

[0226] The accessory identification system based on the image classification model of the present invention can implement any accessory identification method based on the image classification model of the present invention, and is used as a medium for coordinating the operations and signal transmissions between various modules to complete the accessory identification method based on the image classification model. The internal modules of the system cooperate with each other, improving the accuracy of industrial accessory fault identification and life prediction.

[0227] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be included in the present invention.

[0228] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for identifying accessories based on an image classification model, characterized in that: The following steps are involved: Step S1: acquiring industrial accessory data, and performing topological structure analysis based on the industrial accessory data, thereby obtaining industrial accessory topological structure data; performing accessory corrosion simulation based on the industrial accessory topological structure data, thereby obtaining accessory corrosion data; Step S2: locating the corrosion position of the accessory based on the accessory corrosion data, thereby obtaining the accessory corrosion position data, and performing image acquisition, thereby obtaining the accessory corrosion image; Perform corrosion stress concentration point analysis on the corrosion image of the accessories to obtain a corrosion stress concentration point image; Step S3: classifying the corrosion stress concentration point images into types, thereby obtaining pitting stress concentration point images and uniform corrosion stress concentration point images; classifying the accessory types according to the industrial accessory topological structure data, thereby obtaining accessory type data, and collecting accessory type images, thereby obtaining accessory type images; Step S4: constructing an image classification model according to the accessory type image, and performing accessory recognition on the industrial accessory data, thereby obtaining accessory recognition data; The life of the accessory is predicted based on the pitting stress concentration point image and the uniform corrosion stress concentration point image, thereby obtaining the life prediction data of the accessory; Step S5: performing accessory failure probability modeling on the accessory identification data according to the accessory life prediction data, thereby obtaining accessory failure identification data.

2. The accessory recognition method based on the image classification model according to claim 1, characterized in that: Step S1 is specifically as follows: Step S11: acquiring industrial accessory data, and performing accessory hole feature extraction and accessory slot feature extraction according to the industrial accessory data, thereby obtaining accessory hole data and accessory slot data; Step S12: performing assembly connection point channel identification on the accessory slot data according to the accessory hole data, thereby obtaining assembly connection point channel data; Step S13: extracting connection node accessory features from the industrial accessory data, thereby obtaining connection node accessory data; Step S14: constructing topology according to the connection node accessories and assembly connection point channel data, thereby obtaining industrial accessories topology structure data; Step S15: performing accessory corrosion analysis based on the industrial accessory topology data to obtain accessory corrosion data.

3. The accessory recognition method based on the image classification model according to claim 2 is characterized in that: Step S15 is specifically as follows: Step S151: using industrial environment sensors to obtain industrial environment humidity data; Step S152: dividing the industrial accessories topology data into accessories affected by humidity according to the industrial environment humidity data, thereby obtaining humidity-affected industrial accessories data; Step S153: collecting industrial accessories according to the data of industrial accessories affected by humidity, thereby obtaining industrial accessories affected by humidity; Step S154: performing electrochemical corrosion simulation on industrial accessories affected by humidity according to the industrial environment humidity data, thereby obtaining electrochemical corrosion industrial accessories data; Step S156: performing electrolyte membrane formation analysis on electrochemical corrosion industrial accessories data according to industrial environment humidity data, thereby obtaining electrolyte membrane data; Step S157: drawing a current density distribution diagram according to the electrolyte membrane data, thereby obtaining a current density distribution diagram; Step S158: identifying high-density areas according to the current density distribution map, thereby obtaining high current density area data; Step S159: Conduct corrosion analysis on industrial accessories affected by humidity based on the high current density area data, thereby obtaining accessory corrosion data.

4. The accessory recognition method based on the image classification model according to claim 1, characterized in that: Step S2 is specifically as follows: Step S21: locating the corrosion position of the accessory based on the accessory corrosion data, thereby obtaining the accessory corrosion position data, and performing image acquisition, thereby obtaining an accessory corrosion image; Step S22: performing pore defect analysis on the accessory corrosion image to obtain pore defect data; Step S23: performing crack defect analysis on the accessory corrosion image to obtain crack defect data; Step S24: performing position recognition on the accessory corrosion image according to the pore defect data, thereby obtaining pore defect position data; Step S25: performing position recognition on the accessory corrosion image according to the crack defect data, thereby obtaining crack defect position data; Step S26: performing corrosion stress concentration point intersection operation according to the pore defect position data and the crack defect position data to obtain corrosion stress concentration point data, and performing image acquisition to obtain a corrosion stress concentration point image.

5. The accessory recognition method based on the image classification model according to claim 4 is characterized in that: Step S22 is specifically as follows: Step S221: performing denoising processing on the accessory corrosion image to obtain a denoised accessory corrosion image; Step S222: performing a binarization conversion on the denoised accessory corrosion image to obtain a binarized accessory corrosion image; Step S223: acquiring historical pore edge data, and performing pore edge detection on the binary accessory corrosion image, thereby obtaining a pore edge image; Step S224: performing roundness detection on the pore edge image to obtain roundness data, and performing statistics on the roundness data to obtain high roundness data; Step S225: locating the pore centroid of the binary accessory corrosion image according to the high roundness data, thereby obtaining pore centroid data; Step S226: generating a heat map according to the stomatal centroid data, thereby obtaining a stomatal heat map; Step S227: evaluating the number of pores on the pore heat map to obtain pore number data; Step S228: evaluating the pore depth of the binary accessory corrosion image according to the pore centroid data, thereby obtaining pore depth data; Step S229: integrating pore defects according to the pore quantity data and the pore depth data, thereby obtaining pore defect data.

6. The accessory recognition method based on the image classification model according to claim 5, characterized in that: Step S228 is specifically as follows: The binary accessory corrosion image is regionally located according to the pore centroid data, thereby obtaining the accessory pore corrosion area data; Perform three-dimensional reconstruction of accessories based on the data of the accessories' pore corrosion area, thereby obtaining three-dimensional pore corrosion accessories; Perform illumination mapping simulation on three-dimensional pore corrosion accessories to obtain illumination mapping data; Performing illumination gradient calculation on illumination mapping data to obtain illumination gradient data; The stomatal depth is estimated based on the light gradient data to obtain the stomatal depth data.

7. The accessory recognition method based on the image classification model according to claim 4, characterized in that: Step S23 is specifically as follows: Step S231: performing grayscale conversion on the accessory corrosion image to obtain a grayscale accessory corrosion image; Step S232: Calculate pixel gradients according to the grayscale accessory erosion image to obtain pixel gradient data; Step S233: drawing a gradient amplitude map according to the pixel gradient data, thereby obtaining a gradient amplitude map; Step S234: setting a crack high threshold and a crack low threshold according to the grayscale accessory corrosion image, thereby obtaining crack high threshold data and crack low threshold data; Step S235: performing edge detection on the gradient amplitude map according to the crack high threshold data and the crack low threshold data, and obtaining first pixel edge point data if the pixel gradient is greater than the crack high threshold data; If the pixel gradient is between the crack high threshold data and the crack low threshold data, further determining that if the pixel gradient is connected to the first pixel edge point data, the second pixel edge point data is obtained; Step S236: integrating the first pixel edge point data and the second pixel edge point data to obtain crack pixel edge point data; Step S237: Calculating the crack length of the grayscale accessory corrosion image according to the crack pixel edge point data, thereby obtaining crack length data; Step S238: Calculating the crack width of the grayscale accessory corrosion image according to the crack pixel edge point data, thereby obtaining crack width data; Step S239: Perform crack area evaluation based on the crack length data and the crack width data to obtain crack area data, and visualize the crack defects of the grayscale accessory corrosion image based on the crack area data to obtain crack defect data.

8. The accessory recognition method based on the image classification model according to claim 1, characterized in that: Step S3 is specifically as follows: Step S31: performing corrosion morphology recognition on the corrosion stress concentration point image to obtain corrosion morphology data; Step S32: classifying the corrosion stress concentration point images according to the corrosion morphology data, thereby obtaining pitting stress concentration point images and uniform corrosion stress concentration point images; Step S33: classifying the types of accessories according to the industrial accessories topology data, thereby obtaining control accessories and transmission accessories; Step S34: performing high-resolution image acquisition on the controllable accessory, thereby obtaining a high-resolution controllable accessory image; Step S35: performing infrared imaging acquisition on the transmission accessory, thereby obtaining infrared imaging of the transmission accessory; Step S36: integrating the accessory type image according to the high-resolution control accessory image and the transmission accessory infrared imaging, thereby obtaining the accessory type image.

9. The accessory recognition method based on the image classification model according to claim 1, characterized in that: Step S4 is specifically as follows: Step S41: constructing an image classification model according to the accessory type image, inputting the industrial accessory data into the image classification model, and performing accessory image classification to obtain an accessory classification image; Step S42: performing accessory recognition according to the accessory classification image, thereby obtaining accessory recognition data; Step S43: performing local fatigue life prediction on the accessory identification data according to the pitting stress concentration point image, thereby obtaining local fatigue life prediction data; Step S44: performing global fatigue cumulative life prediction on the accessory identification data according to the uniform corrosion stress concentration point image, thereby obtaining global fatigue cumulative life prediction data; Step S45: Integrate the local fatigue life prediction data and the global fatigue cumulative life prediction data to obtain the accessory life prediction data.

10. An accessory recognition system based on an image classification model, characterized in that: Used to execute the accessory recognition method based on the image classification model as claimed in claim 1, the accessory recognition system based on the image classification model comprises: The accessory corrosion simulation module is used to obtain industrial accessory data and perform topological structure analysis based on the industrial accessory data to obtain industrial accessory topological structure data; perform accessory corrosion simulation based on the industrial accessory topological structure data to obtain accessory corrosion data; The corrosion stress concentration point analysis module is used to locate the corrosion position of the accessory based on the corrosion data of the accessory, thereby obtaining the corrosion position data of the accessory, and to collect images, thereby obtaining the corrosion image of the accessory; the corrosion stress concentration point analysis is performed on the corrosion image of the accessory, thereby obtaining the corrosion stress concentration point image; The accessory type classification module is used to classify the corrosion stress concentration point image into types, thereby obtaining the pitting stress concentration point image and the uniform corrosion stress concentration point image; classify the accessory type according to the industrial accessory topological structure data, thereby obtaining the accessory type data, and collect the accessory type image, thereby obtaining the accessory type image; The accessory life prediction module is used to build an image classification model based on the accessory type image, and perform accessory recognition on the industrial accessory data to obtain accessory recognition data; perform accessory life prediction on the accessory recognition data based on the pitting stress concentration point image and the uniform corrosion stress concentration point image to obtain accessory life prediction data; The accessory failure identification module is used to perform accessory failure probability modeling on the accessory identification data according to the accessory life prediction data, so as to obtain the accessory failure identification data.

Images