Welding quality detection method and platform for LNG (Liquefied Natural Gas) marine pipe fitting

Abstract

The invention provides a welding quality detection method and platform for an LNG ship pipe fitting, and relates to the technical field of intelligent detection, and the method comprises the steps: determining the welding damage strength based on pipe fitting defect feature space analysis, calling a bearing performance evaluation channel to carry out performance influence analysis, and outputting a first potential performance influence coefficient; taking the standard pipe fitting welding image as a reference, performing similar traversal comparison on the pipe fitting surface image, and determining a second potential performance influence coefficient according to a deviation area proportion; evaluating and determining the damage strength of the pipe fitting according to the first potential performance influence coefficient and the second potential performance influence coefficient, and if the damage strength is greater than a strength threshold value, performing maintenance identification on the pipe fitting. According to the method, the problem that the real damage condition of the pipe fitting cannot be accurately judged due to the fact that targeted detection cannot be performed on the welding defects of the pipe fitting in combination with an application scene in an existing method can be solved, the defects on the surface and in the pipe fitting can be efficiently and accurately recognized, and the accuracy and reliability of pipe fitting damage evaluation are remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent detection technology, and in particular to a welding quality detection method and platform for LNG marine pipe fittings. Background Art

[0002] In the manufacturing and maintenance process of pipelines and pipe fittings, welding quality is a key factor to ensure the reliability and safety of pipe fittings. Pipe fittings, especially those used in high pressure, high temperature and corrosive environments, such as LNG ship fittings, have a welding quality that is directly related to the bearing capacity and service life of the pipeline. Welding defects, such as cracks, pores, slag inclusions, and lack of fusion, often occur in welded joints and their heat-affected zones. If these defects are not discovered and repaired in time, they may lead to failure of the pipe fittings and even cause accidents.

[0003] At present, there is a technical problem in the existing pipe fitting welding quality inspection method that it is unable to conduct targeted inspection of pipe fitting welding defects in combination with pipe fitting application scenarios, resulting in the inability to accurately judge the actual damage condition of the pipe fitting. Summary of the invention

[0004] The purpose of this application is to provide a welding quality inspection method and platform for LNG marine pipe fittings, so as to solve the technical problem that the existing pipe fitting welding quality inspection method cannot accurately judge the actual damage condition of the pipe fittings due to the inability to conduct targeted inspection of pipe fitting welding defects in combination with the pipe fitting application scenarios.

[0005] In view of the above problems, the present application provides a welding quality detection method and platform for LNG marine pipe fittings.

[0006] In the first aspect, the present application provides a welding quality inspection method for LNG ship pipe fittings, which is implemented through a welding quality inspection platform for LNG ship pipe fittings, including: performing ray scanning and surface image acquisition on a target Y-type three-way pipe fitting, performing internal defect identification and surface defect identification according to the pipe fitting ray image and the pipe fitting surface image, respectively, to obtain an internal defect feature distribution and a surface defect feature distribution; performing spatial fitting and maximum value retention processing on the internal defect feature distribution and the surface defect feature distribution, and modeling to generate a pipe fitting defect feature space; determining the welding damage intensity based on the pipe fitting defect feature space analysis, and calling a bearing performance evaluation channel to perform a performance impact analysis based on the welding damage intensity, and outputting a first potential performance impact coefficient; taking a standard pipe welding image of the target Y-type three-way pipe fitting as a reference, performing a similarity traversal comparison on the pipe fitting surface image, and determining a second potential performance impact coefficient based on the deviation area ratio; evaluating and determining the pipe fitting damage intensity based on the first potential performance impact coefficient and the second potential performance impact coefficient, if the pipe fitting damage intensity is greater than a risk intensity threshold, marking the target Y-type three-way pipe fitting for inspection.

[0007] In a second aspect, the present application also provides a welding quality inspection platform for LNG marine pipe fittings, which is used to execute a welding quality inspection method for LNG marine pipe fittings as described in the first aspect, including: a defect feature distribution acquisition module, which is used to perform ray scanning and surface image acquisition on a target Y-shaped tee pipe fitting, respectively perform internal defect identification and surface defect identification based on the pipe fitting ray image and the pipe fitting surface image, and obtain the internal defect feature distribution and the surface defect feature distribution; a defect feature space generation module, which is used to perform spatial fitting and maximum value retention processing on the internal defect feature distribution and the surface defect feature distribution, and model and generate a pipe fitting defect feature space; a first performance influence coefficient obtaining module, which is used to analyze and determine the welding damage strength based on the pipe fitting defect feature space, and call a bearing performance evaluation channel to perform performance influence analysis according to the welding damage strength, and output a first potential performance influence coefficient; a second performance influence coefficient obtaining module, which is used to perform similar traversal comparison on the pipe fitting surface image with the standard pipe fitting welding image of the target Y-shaped tee pipe fitting as a reference, and determine a second potential performance influence coefficient according to the deviation area ratio; a pipe fitting damage strength determination module, which is used to evaluate and determine the pipe fitting damage strength according to the first potential performance influence coefficient and the second potential performance influence coefficient, and if the pipe fitting damage strength is greater than the risk strength threshold, perform a maintenance identification on the target Y-shaped tee pipe fitting.

[0008] One or more technical solutions provided in the present application have at least the following technical effects or advantages: By performing ray scanning and surface image acquisition on a target Y-shaped tee pipe fitting, respectively performing internal defect identification and surface defect identification based on the pipe fitting ray image and the pipe fitting surface image, and obtaining the internal defect feature distribution and the surface defect feature distribution; then performing spatial fitting and maximum value retention processing on the internal defect feature distribution and the surface defect feature distribution, and modeling and generating a pipe fitting defect feature space; then analyzing and determining the welding damage strength based on the pipe fitting defect feature space, and calling a bearing performance evaluation channel to perform performance influence analysis according to the welding damage strength, and outputting a first potential performance influence coefficient; on the other hand, performing similar traversal comparison on the pipe fitting surface image with the standard pipe fitting welding image of the target Y-shaped tee pipe fitting as a reference, and determining a second potential performance influence coefficient according to the deviation area ratio; finally evaluating and determining the pipe fitting damage strength according to the first potential performance influence coefficient and the second potential performance influence coefficient, and if the pipe fitting damage strength is greater than the risk strength threshold, performing a maintenance identification on the target Y-shaped tee pipe fitting; it can efficiently and accurately identify the defects on the surface and inside of the pipe fitting, significantly improve the accuracy and reliability of pipe fitting damage assessment, so as to discover potential risks in advance and ensure the use safety and reliability of the pipe fitting.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0011] Figure 1 A schematic diagram of a welding quality inspection method for LNG marine pipe fittings for this application; Figure 2 A schematic diagram of a flow chart for determining welding damage strength in a welding quality inspection method for LNG marine pipe fittings of the present application; Figure 3 The present invention is a structural schematic diagram of a welding quality inspection platform for LNG marine pipe fittings.

[0012] Description of reference numerals: Defect feature distribution acquisition module 11, defect feature space generation module 12, first performance influence coefficient acquisition module 13, second performance influence coefficient acquisition module 14, pipe damage strength determination module 15. DETAILED DESCRIPTION

[0013] This application provides a welding quality detection method and platform for LNG marine pipe fittings, which solves the technical problem that the existing pipe fitting welding quality detection methods cannot accurately judge the actual damage status of pipe fittings due to the inability to conduct targeted detection of pipe fitting welding defects in combination with pipe fitting application scenarios. It can efficiently and accurately identify defects on the surface and inside of pipe fittings, significantly improve the accuracy and reliability of pipe fitting damage assessment, so that potential risks can be discovered in advance and the safety and reliability of pipe fittings can be ensured.

[0014] Next, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the example embodiments described herein. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts belong to the scope of protection of the present application. Additionally, it should be noted that for the sake of description, only parts related to the present application are shown in the accompanying drawings rather than all of them.

[0015] Example 1. Please refer to the attached Figure 1 . The present application provides a method for detecting the welding quality of LNG marine pipe fittings, which is applied to a platform for detecting the welding quality of LNG marine pipe fittings, and specifically includes the following steps: S100: Perform ray scanning and surface image acquisition on the target Y-shaped tee pipe fitting, respectively perform internal defect recognition and surface defect recognition based on the pipe fitting ray image and the pipe fitting surface image, and obtain the internal defect feature distribution and the surface defect feature distribution.

[0016] Furthermore, step S100 of the present application further includes: S110: Based on a multi-scale convolutional neural network, for the purpose of identifying predetermined defect features, construct an internal defect recognition model and a surface defect recognition model, wherein the predetermined defect features include defect types and size information, and the defect types at least include cracks, pores, and slag inclusions; S120: Use the internal defect recognition model to perform internal defect recognition on the pipe fitting ray image and output the internal defect feature distribution; S130: Use the surface defect recognition model to perform surface defect recognition on the pipe fitting surface image and output the surface defect feature distribution.

[0017] Specifically, the Y-shaped tee pipe fitting plays an important role in LNG marine pipe fittings. Especially in the liquefied natural gas (LNG) transportation, storage, and distribution systems, LNG marine pipe fittings are pipeline system components designed specifically for liquefied natural gas ships, with special requirements for low temperature resistance, corrosion resistance, high strength, and safety. As a part of this system, the Y-shaped tee pipe fitting undertakes the function of diverting or converging liquefied natural gas in the pipeline system, and its welding quality directly affects the operation safety of the entire pipeline system.

[0018] First, perform ray scanning and surface image acquisition on the target Y-shaped tee fitting. Among them, ray scanning is a commonly used method in non-destructive testing, mainly used to detect potential defects inside the fitting. For Y-shaped tee fittings, ray scanning can effectively reveal internal defects in the welded joints, such as pores, slag inclusions, lack of fusion, and cracks, etc.; according to the material and thickness of the fitting, select an appropriate ray device, usually an X-ray machine, and the energy of the ray needs to be selected according to the thickness and material of the fitting to ensure that it can penetrate and form a clear image; obtain the ray image of the fitting. Surface image acquisition is used to check for surface defects of the fitting, such as cracks, weld beads, pores, etc. Surface defects may have a direct impact on the sealing performance and load-bearing capacity of the fitting, so surface detection is equally crucial; by using a high-definition industrial camera, take multi-angle photos of the welded joints and pipe surfaces of the target Y-shaped tee fitting to obtain the surface image of the fitting.

[0019] The multi-scale convolutional neural network uses convolutional operations to extract features at different scales, thereby effectively capturing defects of different sizes. By using multiple convolutional kernels and pooling layers, the network can gradually extract features from simple to complex, effectively distinguish and locate different types of defects. Then, based on the multi-scale convolutional neural network, an internal defect recognition model and a surface defect recognition model are constructed for the purpose of identifying predetermined defect features. Among them, internal defect recognition mainly focuses on defects inside the fitting. This model needs to identify internal defects such as cracks and pores, including an input layer, a multi-scale convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input data of its input layer is the ray image of the fitting, and the output data is the internal defect type and defect size information; surface defect recognition mainly focuses on possible defects on the surface of the fitting, such as cracks, pores, weld beads, etc. It also includes an input layer, a multi-scale convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input data of its input layer is the surface image of the fitting, and the output data of the output layer is the surface defect type and defect size information.

[0020] Further, according to the pipe fitting inspection log, with the target Y-shaped tee pipe fitting as a constraint, a sample pipe fitting ray image set and a sample internal defect feature set (internal defect type and defect size information) are collected as the internal defect training data set; the internal defect recognition model is supervised and trained using the internal defect training data set. First, the preprocessed ray image is used as the input, and the defect type and size are used as the labels. The input data propagates forward through the network to generate a predicted output; then, the defect type and size output by the network are compared with the actual labels, and the loss value is calculated through the loss function; then, according to the calculated loss, the backpropagation algorithm is used to update the weights in the network, and an optimization algorithm (such as Adam or SGD) is used to minimize the loss function, thereby optimizing the performance of the network; iterative training is repeated until the loss function converges, and the trained internal defect recognition model is obtained; on the other hand, a sample pipe fitting surface image set and a sample surface defect feature set (surface defect type and defect size information) are collected as the surface defect training data set, and the surface defect recognition model is supervised and trained using the surface defect training data set until convergence, and the trained surface defect recognition model is obtained. By constructing an internal defect recognition model and a surface defect recognition model based on a multi-scale convolutional neural network, the welding defects in the pipe fittings can be efficiently and accurately identified and classified, significantly improving the accuracy and speed of detection.

[0021] Next, the pipe fitting ray image is input into the internal defect recognition model for internal defect recognition, and the internal defect feature distribution (including the position, type, size information, etc. of each internal defect) is output; the pipe fitting surface image is input into the surface defect recognition model for surface defect recognition, and the surface defect feature distribution (including the position, type, size, etc. of each surface defect) is output.

[0022] S200: Perform spatial fitting and maximum retention processing on the internal defect feature distribution and the surface defect feature distribution, and model and generate the pipe fitting defect feature space.

[0023] Specifically, spatial fitting and maximum retention processing are performed on the internal defect feature distribution and the surface defect feature distribution. The purpose of spatial fitting is to model the defect feature distribution in the image space, create a unified spatial model to describe the distribution of internal and external defects of the pipe fitting. Through this process, the defects can be spatially located, and the severity and influence range of the defects can be determined. First, the spatial coordinates of all defects (such as pixel coordinates in the image) are converted into a physical coordinate system, and the coordinate range is normalized to ensure the consistency of spatial data. Then, based on the defect position, size information, and defect type, interpolation methods (such as bilinear interpolation, spline interpolation, etc.) are used to perform spatial fitting on the defect distribution. Through spatial fitting, a smooth defect distribution model can be generated, showing the relative influence area of each defect. The purpose of maximum retention processing is to retain the most serious and significant defects in the pipe fitting. These defects usually have an important impact on the safety and performance of the pipe fitting. This processing step screens the spatial features of the defects and only retains those key defects that may cause performance degradation. First, the maximum value regions in the defect distribution map after spatial fitting are identified through algorithms (such as local maximum detection, non-maximum suppression, etc.). These regions usually correspond to the positions of the most serious defects in the pipe fitting. The maximum value can represent the position and intensity of the defect, that is, the position where the defect has the greatest impact on the overall structure. Then, screening is carried out within the maximum value regions, and only the most serious defects (such as crack tips, pore regions, etc.) are retained. By setting thresholds, minor defects with less impact on performance are removed. By highlighting the maximum value regions, the visualization effect of the impact of defects on the overall performance is enhanced, providing a basis for subsequent performance evaluation and decision-making.

[0024] Then, combining the results of spatial fitting and maximum value processing, a three-dimensional or two-dimensional defect feature space is created. This space reflects the defect distribution, defect intensity, and defect influence range of the pipe fitting, obtaining the defect feature space of the pipe fitting. This spatial model effectively shows the distribution, type, and influence degree of various defects in the pipe fitting, providing strong support for subsequent pipe fitting health assessment, maintenance decision-making, and performance prediction.

[0025] S300: Determine the welding damage intensity based on the analysis of the pipe fitting defect feature space, and call the bearing performance evaluation channel according to the welding damage intensity to perform performance impact analysis, and output the first potential performance impact coefficient.

[0026] Furthermore, as Figure 2 shown, step S300 of this application further includes: S310: Based on the pipe fitting defect feature space, conduct horizontal defect feature analysis and vertical defect feature analysis respectively to determine the proportion of the horizontal defect area, the dispersion degree of the horizontal defect distribution, and the depth of the vertical defect; S320: Obtain the material properties and working environment characteristics of the target Y-shaped tee pipe fitting, and conduct corrosion intensity simulation analysis according to the material properties and working environment characteristics to determine the corrosion intensity coefficient; S330: Evaluate and determine the welding damage intensity according to the proportion of the horizontal defect area, the dispersion degree of the horizontal defect distribution, the depth of the vertical defect, and the corrosion intensity coefficient.

[0027] Specifically, based on the pipe fitting defect feature space, conduct horizontal defect feature analysis and vertical defect feature analysis respectively. The horizontal defect feature analysis mainly focuses on the distribution of defects in the horizontal direction of the pipe fitting (usually the cross-sectional direction of the pipe), including the proportion of the defect area and the dispersion degree of the distribution. Among them, the proportion of the horizontal defect area refers to the ratio of the area of the region occupied by the defect to the area of the entire cross-section in the cross-section of the pipe fitting; the dispersion degree of the horizontal defect distribution is used to evaluate whether the distribution of defects on the cross-section of the pipe fitting is uniform. The smaller the dispersion degree, the more concentrated the defects are in a certain part of the pipe fitting. For example, the standard deviation or variance is used to measure the distribution degree of the horizontal defects on the cross-section of the pipe fitting, and the proportion of the horizontal defect area and the dispersion degree of the horizontal defect distribution are obtained. The vertical defect feature analysis focuses on the distribution of defects in the vertical direction of the pipe fitting (along the axis direction of the pipe), especially the depth of the defect (the vertical depth from the surface of the pipe fitting to the defect). The depth of the vertical defect is the vertical distance from the surface of the pipe fitting to the defect position. By calculating the average value of the defect depths of multiple internal defects, the average defect depth is set as the depth of the vertical defect.

[0028] Obtain the material properties and working environment characteristics of the target Y-shaped tee pipe fitting. The commonly used materials for Y-shaped tee pipe fittings include stainless steel, carbon steel, alloy steel, etc. The corrosion resistance of these materials varies greatly. Therefore, the material properties are an important factor determining the corrosion intensity of the pipe fitting, and the material properties include the corrosion rate; the corrosion intensity of the pipe fitting in the actual working environment is affected by various environmental factors. The working environment characteristics include temperature, humidity, liquid components (such as water, acid, salt), etc. The average temperature, average humidity, and average component concentration can be set as the working environment characteristics. Then, conduct corrosion intensity simulation analysis according to the material properties and working environment characteristics. For example, conduct simulation analysis by constructing a linear corrosion model, and calculate the corrosion intensity coefficient through the relationship between the corrosion rate and the environment and materials. This coefficient measures the influence degree of the corrosion environment on the pipe fitting. Among them, the larger the corrosion intensity coefficient, the more serious the corrosion situation is.

[0029] Finally, a comprehensive evaluation of the welding damage strength is carried out according to the proportion of the lateral defect area, the dispersion of the lateral defect distribution, the longitudinal defect depth, and the corrosion strength coefficient. Among them, the welding damage strength is positively correlated with the proportion of the lateral defect area, the longitudinal defect depth, and the corrosion strength coefficient, and negatively correlated with the dispersion of the lateral defect distribution. First, dimensionless processing is performed on the proportion of the lateral defect area, the dispersion of the lateral defect distribution, the longitudinal defect depth, and the corrosion strength coefficient to ensure that features of different scales and units have the same influence. Dimensionless processing usually adopts normalization or standardization methods to make each feature on the same scale and avoid disproportionate influence of certain features on the results due to their large numerical ranges. Then, weight allocation is carried out according to the influence degrees of the proportion of the lateral defect area, the dispersion of the lateral defect distribution, the longitudinal defect depth, and the corrosion strength coefficient on the welding damage strength. The greater the influence degree, the greater the corresponding weight. Then, weighted calculation is performed on the influence degrees of the proportion of the lateral defect area, the dispersion of the lateral defect distribution, the longitudinal defect depth, and the corrosion strength coefficient on the welding damage strength according to the weight allocation to obtain the welding damage strength. The welding damage strength is used to evaluate the overall health status of the pipe fittings. The higher the damage strength, the greater the potential risk during the use of the pipe fittings.

[0030] Further, step S300 of the present application further includes: S340: Pre-train the load-bearing performance evaluation channel, where the load-bearing performance evaluation channel includes P base evaluators.

[0031] Further, step S340 of the present application further includes: S341: According to the historical maintenance logs of the target Y-shaped tee pipe fittings, collect the three-dimensional feature distribution sets of the sample pipe fitting defects, and count the proportion of the pipe fitting bearing failures of the three-dimensional feature distributions of different sample pipe fitting defects within a predetermined time window to obtain the sample pipe fitting bearing failure ratio, which is set as the sample performance influence coefficient, and obtain the sample performance influence coefficient set; S342: Integrate the three-dimensional feature distribution sets of the sample pipe fitting defects and the sample performance influence coefficient set to obtain a sample training set, and equally divide it into P parts to obtain P training arrays; S343: Use the P training arrays to train the generator and discriminator of the generative adversarial network until convergence to obtain P base evaluators, and integrate them to obtain the load-bearing performance evaluation channel.

[0032] Specifically, according to the historical maintenance logs of the target Y-shaped tee fittings, defect information of the sample fittings is extracted from the maintenance logs of the target Y-shaped tee fittings. These defects include internal and surface defects of the fittings. The three-dimensional characteristics of the defects may include defects such as cracks, pores, and slag inclusions, as well as their positions, sizes, and depths in three-dimensional space. Then, based on the collected defect data, a three-dimensional feature distribution set of the sample fitting defects is constructed, and the defect features of each fitting sample will be represented as data points in a three-dimensional space. Further, the proportion of bearing failures of the fittings in the three-dimensional feature distribution of the sample fitting defects within a predetermined time window (such as within a week) is statistically analyzed. A bearing failure refers to the loss of bearing capacity of the fitting during its operation due to defects or damages, usually manifested as crack propagation, pore enlargement, strength reduction caused by slag inclusions, etc. The number of failures occurring within the predetermined time window (such as within a week) is counted, and the proportion of bearing failures of the sample fittings is obtained and set as the sample performance impact coefficient. This coefficient reflects the potential impact of defects on the bearing capacity of each fitting during its service life, and a sample performance impact coefficient set is obtained.

[0033] Further integrate the three-dimensional feature distribution set of the sample fitting defects and the sample performance impact coefficient set, that is, use the three-dimensional feature distribution of the sample fitting defects and the corresponding sample performance impact coefficient as a set of training data to obtain a sample training set. Then divide the sample training set into P equal parts, where P is an integer greater than 10, and the value of P can be set according to actual needs, such as set to 20, to obtain P training arrays. Then randomly select the first training array without replacement from the P training arrays, and conduct supervised training on the generative adversarial network. The generative adversarial network is a neural network framework composed of a generator and a discriminator. The generator generates fake samples, and the discriminator provides feedback to the generator by distinguishing between real and fake samples. Through this adversarial training, the generator can learn how to generate realistic samples. In this case, the generator will generate bearing performance data of the fittings based on the defect features of the sample fittings (such as three-dimensional defect distribution, performance impact coefficient, etc.), and the discriminator will evaluate the bearing performance of the samples, that is, the bearing capacity of the fittings under specific defect conditions. Among them, the task of the generator is to generate bearing performance data of the fittings according to the three-dimensional defect features of the fittings, and these data should be able to reflect the impact of defects on the bearing capacity of the fittings; the discriminator receives the real bearing performance data and the fake performance data generated by the generator, and outputs a probability value representing the real and fake data. The goal of the discriminator is to distinguish between real data and generated data as accurately as possible and provide feedback to the generator. Use the first training array to conduct supervised training on the generative adversarial network. In each round of training, the generator tries to generate more and more realistic bearing performance data, and the discriminator is trained according to the generated data and the real data to improve its ability to identify generated data and real data. The training continues until the network converges, the generator can generate sufficiently realistic bearing performance data, and the ability of the discriminator is also strengthened; output the first basic evaluator after the training is completed.

[0034] Continue to use other training data to perform supervised training on the generator and discriminator of the generative adversarial network, obtain P basic evaluators, and combine the P basic evaluators to obtain the load-bearing performance evaluation channel.

[0035] S350: Obtain the maximum welding damage intensity within a predetermined historical time zone, and determine the number of basic evaluators of the load-bearing performance evaluation channel to be called according to the welding damage intensity and the maximum welding damage intensity, denoted as Q; S360: Randomly activate Q selected basic evaluators among the P basic evaluators, and use the Q selected basic evaluators to perform a performance impact analysis on the pipe fitting defect feature space, output Q performance impact coefficients, and obtain the first potential performance impact coefficient after calculating the mean value.

[0036] Specifically, obtain the maximum welding damage intensity within a predetermined historical time zone (such as the most recent month); and multiply the ratio of the welding damage intensity to the maximum welding damage intensity by P and round down to obtain the number of basic evaluators to be called, denoted as Q. For example, assume the welding damage intensity is 0.2, the maximum welding damage intensity is 0.5, and P is 20, then Q is 8. Then randomly activate Q selected basic evaluators among the P basic evaluators, input the pipe fitting defect feature space into the Q selected basic evaluators for performance impact analysis, output Q performance impact coefficients, and calculate the mean value of the Q performance impact coefficients to obtain the first potential performance impact coefficient.

[0037] By selecting an appropriate number of basic evaluators according to the welding damage intensity for load-bearing performance analysis, the number of basic evaluators required can be flexibly adjusted dynamically according to the actual state of the pipe fitting, so that more basic evaluators can be used for more accurate load-bearing performance evaluation in the case of more serious damage, while in the case of less damage, the calculation amount can be reduced, the evaluation efficiency can be improved, and the reasonable utilization rate of computing power resources can be improved.

[0038] S400: Based on the standard pipe fitting welding image of the target Y-shaped tee, perform a similarity traversal comparison on the pipe fitting surface image, and determine the second potential performance impact coefficient according to the deviation area ratio.

[0039] Further, step S400 of the present application further includes: S410: Crop the welding areas of the standard pipe fitting welding image and the pipe fitting surface image respectively to obtain a standard welding area image and a pipe fitting welding area image; S420: Perform grayscale processing on the standard welding area image and the pipe fitting welding area image to obtain a standard welding area grayscale image and a pipe fitting welding area grayscale image; S430: Randomly select a first grayscale area within the pipe fitting welding area grayscale image according to a predetermined size, where the predetermined size is a pixel area of 5 by 5; S440: Perform a traversal similarity comparison between the first grayscale area and the area of the predetermined size within the standard welding area grayscale image, and set the maximum similarity as the first area similarity of the first grayscale area; S450: Continue to randomly select areas and perform traversal similarity comparisons within the pipe fitting welding area grayscale image to obtain multiple grayscale areas and multiple area similarities; S460: Set the grayscale areas with area similarities less than the similarity threshold as height difference areas, calculate the ratio of the area of the height difference area to the area of the pipe fitting welding area grayscale image to obtain the deviation area ratio, and determine the second potential performance impact coefficient according to the deviation area ratio.

[0040] Specifically, obtain the standard pipe fitting welding image of the target Y-shaped tee (usually generated from process standards or empirical data); then crop the welding areas of the standard pipe fitting welding image and the pipe fitting surface image respectively. The welding area usually refers to the specific part of the pipe fitting welding, such as the weld area. Use an edge detection algorithm (such as Canny edge detection) to determine the position of the welding area, and crop the image according to the detected edge information to obtain an image containing only the welding area, thus obtaining a standard welding area image and a pipe fitting welding area image. Further perform grayscale processing on the standard welding area image and the pipe fitting welding area image, that is, convert the image from color to grayscale. The RGB to grayscale formula can be used for grayscale processing to obtain a standard welding area grayscale image and a pipe fitting welding area grayscale image. Then obtain the predetermined size, where the predetermined size is a pixel area of 5 by 5, that is, within the pipe fitting welding area grayscale image, select a sub-area with a size of 5 by 5 pixels and randomly select a first grayscale area within the pipe fitting welding area grayscale image according to the predetermined size.

[0041] Next, the first grayscale region is traversed and compared with the region of the predetermined size within the grayscale image of the standard welding region, that is, the first grayscale region (a randomly selected 5 by 5 region from the grayscale image of the pipe fitting welding region) is compared with each region of the predetermined size within the grayscale image of the standard welding region. For example, the structural similarity (SSIM) is used to evaluate the similarity between the first grayscale region and each traversed region. By traversing the entire grayscale image of the standard welding region, the similarity between the first grayscale region and each 5 by 5 region is calculated, and the maximum similarity is selected as the similarity of the first region. Then, multiple 5 by 5 pixel regions are randomly selected from the grayscale image of the pipe fitting welding region, where regions that do not overlap with the previous regions are selected to ensure the diversity of the traversal process. For each newly selected grayscale region, the traversal and similarity comparison steps described in S440 are repeated. The similarity between each randomly selected region and the regions in the grayscale image of the standard welding region is compared, and the similarity is calculated to obtain multiple grayscale regions and multiple region similarities.

[0042] A similarity threshold is set, usually based on experimental or empirical values. This threshold is used to judge the similarity between the grayscale region and the standard welding region. Then, the grayscale regions with region similarities less than the similarity threshold are set as high-difference regions. That is, if the region similarity is less than the similarity threshold, the region is considered a high-difference region, indicating that there is a large difference between this region and the standard welding region, which may be due to welding defects, errors, or non-compliant regions. Further, the ratio of the sum of the areas of the high-difference regions (the sum of the areas of multiple high-difference regions) to the area of the grayscale image of the pipe fitting welding region is calculated to obtain the deviation area ratio. Finally, the second potential performance impact coefficient is determined based on the deviation area ratio, where the deviation area ratio and the second potential performance impact coefficient are positively correlated. That is, the larger the deviation area ratio, the larger the corresponding second potential performance impact coefficient, indicating that the bearing performance of the pipe fitting is more affected.

[0043] By performing a forward similarity comparison between the surface image of the pipe fitting and the standard pipe fitting welding image, the second potential performance impact coefficient is obtained, which can evaluate the bearing performance of the pipe fitting from another aspect, thereby further improving the accuracy and reliability of the pipe fitting damage strength analysis.

[0044] S500: Evaluate and determine the pipe fitting damage strength based on the first potential performance impact coefficient and the second potential performance impact coefficient. If the pipe fitting damage strength is greater than the risk strength threshold, a maintenance label is applied to the target Y-shaped tee pipe fitting.

[0045] Furthermore, step S500 of the present application further includes: S510: Configure a first influence weight and a second influence weight, where the first influence weight is 0.85 and the second influence weight is 0.15; S520: Based on the first influence weight and the second influence weight, perform a weighted calculation on the first potential performance influence coefficient and the second potential performance influence coefficient to obtain the pipe fitting damage strength.

[0046] Specifically, first, configure a first influence weight and a second influence weight, where the first influence weight is 0.85 and the second influence weight is 0.15; then, based on the first influence weight and the second influence weight, perform a weighted calculation on the first potential performance influence coefficient and the second potential performance influence coefficient, and set the weighted calculation result as the pipe fitting damage strength. Through the weighted calculation, by comprehensively considering the first potential performance influence coefficient and the second potential performance influence coefficient, the pipe fitting damage strength can be comprehensively evaluated from multiple aspects, thereby effectively improving the accuracy and reliability of the pipe fitting damage condition assessment.

[0047] Obtain a risk intensity threshold, which is a preset standard value used to distinguish whether the pipe fitting is within the acceptable damage range, and is usually set according to the working environment, design standard and safety requirements of the pipe fitting; judge the pipe fitting damage strength according to the risk intensity threshold. If the pipe fitting damage strength is greater than the risk intensity threshold, it means that the damage degree of the pipe fitting has exceeded the predetermined safety standard and may affect the bearing capacity or stability of the pipe fitting, then perform a maintenance label on the target Y-shaped tee pipe fitting; if the pipe fitting damage strength is less than or equal to the risk intensity threshold, it means that the damage degree of the pipe fitting is acceptable and no maintenance is required.

[0048] In summary, the welding quality detection method for LNG marine pipe fittings provided by this application has the following technical effects: By performing ray scanning and surface image acquisition on the target Y-shaped tee pipe fitting, internal defect recognition and surface defect recognition are respectively carried out according to the pipe fitting ray image and the pipe fitting surface image to obtain the internal defect feature distribution and the surface defect feature distribution; then, spatial fitting and maximum value retention processing are performed on the internal defect feature distribution and the surface defect feature distribution to model and generate the pipe fitting defect feature space; then, based on the pipe fitting defect feature space analysis, the welding damage strength is determined, and according to the welding damage strength, the bearing performance evaluation channel is called to perform performance impact analysis, and the first potential performance impact coefficient is output; on the other hand, taking the standard pipe fitting welding image of the target Y-shaped tee pipe fitting as a reference, similar traversal comparison is performed on the pipe fitting surface image, and the second potential performance impact coefficient is determined according to the deviation area ratio; finally, the pipe fitting damage strength is evaluated and determined according to the first potential performance impact coefficient and the second potential performance impact coefficient. If the pipe fitting damage strength is greater than the risk strength threshold, the target Y-shaped tee pipe fitting is marked for maintenance; it can efficiently and accurately identify the defects on the surface and inside of the pipe fitting, significantly improve the accuracy and reliability of pipe fitting damage assessment, so as to discover potential risks in advance and ensure the use safety and reliability of the pipe fitting.

[0049] Embodiment 2. Based on a welding quality detection method for a kind of LNG ship pipe fitting in the foregoing embodiment with the same inventive concept, the present application also provides a welding quality detection platform for an LNG ship pipe fitting. Please refer to the appendix Figure 3 , including: A defect feature distribution acquisition module 11, configured to perform ray scanning and surface image acquisition on the target Y-shaped tee pipe fitting, and respectively perform internal defect recognition and surface defect recognition according to the pipe fitting ray image and the pipe fitting surface image to obtain the internal defect feature distribution and the surface defect feature distribution; a defect feature space generation module 12, configured to perform spatial fitting and maximum value retention processing on the internal defect feature distribution and the surface defect feature distribution to model and generate the pipe fitting defect feature space; a first performance impact coefficient obtaining module 13, configured to analyze and determine the welding damage strength based on the pipe fitting defect feature space, and call the bearing performance evaluation channel to perform performance impact analysis according to the welding damage strength, and output the first potential performance impact coefficient; a second performance impact coefficient obtaining module 14, configured to perform similar traversal comparison on the pipe fitting surface image with the standard pipe fitting welding image of the target Y-shaped tee pipe fitting as a reference, and determine the second potential performance impact coefficient according to the deviation area ratio; a pipe fitting damage strength determination module 15, configured to evaluate and determine the pipe fitting damage strength according to the first potential performance impact coefficient and the second potential performance impact coefficient. If the pipe fitting damage strength is greater than the risk strength threshold, the target Y-shaped tee pipe fitting is marked for maintenance.

[0050] Further, the welding quality inspection platform for LNG marine pipe fittings is also used for: based on a multi-scale convolutional neural network, for the purpose of identifying predetermined defect features, constructing an internal defect identification model and a surface defect identification model, wherein the predetermined defect features include defect types and size information, and the defect types at least include cracks, pores and slag inclusions; using the internal defect identification model to identify internal defects in the ray image of the pipe fitting and output the distribution of the internal defect features; using the surface defect identification model to identify surface defects in the surface image of the pipe fitting and output the distribution of the surface defect features.

[0051] Further, the welding quality inspection platform for LNG marine pipe fittings is also used for: based on the defect feature space of the pipe fitting, respectively performing horizontal defect feature analysis and vertical defect feature analysis to determine the proportion of the horizontal defect area, the dispersion degree of the horizontal defect distribution and the vertical defect depth; obtaining the material properties and working environment characteristics of the target Y-shaped tee pipe fitting, and performing corrosion intensity simulation analysis according to the material properties and working environment characteristics to determine the corrosion intensity coefficient; evaluating and determining the welding damage intensity according to the proportion of the horizontal defect area, the dispersion degree of the horizontal defect distribution, the vertical defect depth and the corrosion intensity coefficient.

[0052] Further, the welding quality inspection platform for LNG marine pipe fittings is also used for: pre-training a load-bearing performance evaluation channel, wherein the load-bearing performance evaluation channel includes P basic evaluators; obtaining the maximum welding damage intensity in a predetermined historical time period, evaluating and determining the number of calls of the basic evaluators of the load-bearing performance evaluation channel according to the welding damage intensity and the maximum welding damage intensity, and setting it as Q; randomly activating Q selected basic evaluators among the P basic evaluators, and using the Q selected basic evaluators to perform performance impact analysis on the defect feature space of the pipe fitting and output Q performance impact coefficients, and obtaining the first potential performance impact coefficient after calculating the mean value.

[0053] Further, the welding quality inspection platform for LNG marine pipe fittings is also used for: according to the historical maintenance log of the target Y-shaped tee pipe fitting, collecting a set of three-dimensional defect feature distributions of sample pipe fittings, and counting the proportion of pipe fitting load-bearing failures of different three-dimensional defect feature distributions of sample pipe fittings within a predetermined time window to obtain the sample pipe fitting load-bearing failure ratio, which is set as the sample performance impact coefficient, and obtaining a set of sample performance impact coefficients; integrating the set of three-dimensional defect feature distributions of sample pipe fittings and the set of sample performance impact coefficients to obtain a sample training set, and equally dividing it into P parts to obtain P training arrays; using the P training arrays to train the generator and discriminator of the generative adversarial network until convergence to obtain P basic evaluators, and integrating them to obtain the load-bearing performance evaluation channel.

[0054] Further, the welding quality detection platform for LNG ship pipe fittings is also used for: respectively cropping the welding areas of the standard pipe fitting welding image and the pipe fitting surface image to obtain a standard welding area image and a pipe fitting welding area image; performing gray-scale processing on the standard welding area image and the pipe fitting welding area image to obtain a standard welding area gray-scale image and a pipe fitting welding area gray-scale image; randomly selecting a first gray-scale area within the pipe fitting welding area gray-scale image according to a predetermined size, where the predetermined size is a pixel area of 5 by 5; traversing and comparing the first gray-scale area with the area of the same predetermined size within the standard welding area gray-scale image, and setting the maximum similarity as the first area similarity of the first gray-scale area; continuing to randomly select areas and perform traversing and comparing within the pipe fitting welding area gray-scale image to obtain multiple gray-scale areas and multiple area similarities; setting the gray-scale areas with area similarities less than the similarity threshold as height difference areas, calculating the ratio of the area of the height difference areas to the area of the pipe fitting welding area gray-scale image to obtain the deviation area ratio, and determining the second potential performance influence coefficient according to the deviation area ratio.

[0055] Further, the welding quality detection platform for LNG ship pipe fittings is also used for: configuring a first influence weight and a second influence weight, where the first influence weight is 0.85 and the second influence weight is 0.15; based on the first influence weight and the second influence weight, performing weighted calculation on the first potential performance influence coefficient and the second potential performance influence coefficient to obtain the pipe fitting damage intensity.

[0056] The various embodiments in this specification are described in a progressive manner. The key points of each embodiment are the differences from other embodiments. The welding quality detection method and specific examples in the first embodiment above are equally applicable to the welding quality detection platform in this embodiment. Through the detailed description of the welding quality detection method for LNG ship pipe fittings above, those skilled in the art can clearly know the welding quality detection platform for LNG ship pipe fittings in this embodiment. Therefore, for the sake of brevity of the specification, it will not be elaborated here. For the platform disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For related parts, refer to the description in the method part.

[0057] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. 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 application. Therefore, the present application will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0058] Obviously, those skilled in the art can make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of this application and its equivalent technologies, this application also intends to cover these changes and modifications.

Claims

1. The welding quality inspection method of LNG marine pipe fittings is characterized by: Methods include: Perform radiographic scanning and surface image acquisition on the target Y-type three-way pipe fitting, perform internal defect recognition and surface defect recognition based on the pipe fitting radiographic image and pipe fitting surface image, and obtain internal defect feature distribution and surface defect feature distribution; Performing spatial fitting and maximum value retention processing on the internal defect characteristic distribution and the surface defect characteristic distribution, and modeling and generating a pipe defect characteristic space; Determine the welding damage intensity based on the pipe defect feature space analysis, call the load-bearing performance evaluation channel to perform performance impact analysis according to the welding damage intensity, and output a first potential performance impact coefficient; Taking the standard pipe welding image of the target Y-type tee pipe fitting as a reference, performing similarity traversal comparison on the pipe fitting surface image, and determining the second potential performance impact coefficient according to the deviation area ratio; The pipe fitting damage intensity is determined based on the first potential performance impact coefficient and the second potential performance impact coefficient. If the pipe fitting damage intensity is greater than the risk intensity threshold, the target Y-type three-way pipe fitting is marked for maintenance.

2. The welding quality detection method for LNG marine pipe fittings according to claim 1 is characterized in that: According to the tube X-ray image and the tube surface image, internal defects and surface defects are identified respectively to obtain the internal defect feature distribution and the surface defect feature distribution, including: Based on a multi-scale convolutional neural network, an internal defect recognition model and a surface defect recognition model are constructed for the purpose of identifying predetermined defect features, wherein the predetermined defect features include defect type and size information, and the defect types include at least cracks, pores and slag inclusions; Using the internal defect recognition model to perform internal defect recognition on the tube radiographic image, and outputting the internal defect feature distribution; The surface defect recognition model is used to perform surface defect recognition on the pipe surface image, and the surface defect feature distribution is output.

3. The welding quality inspection method for LNG marine pipe fittings according to claim 1 is characterized in that: Determining the welding damage intensity based on the pipe defect feature space analysis includes: Based on the pipe defect feature space, transverse defect feature analysis and longitudinal defect feature analysis are performed respectively to determine the transverse defect area ratio, transverse defect distribution dispersion and longitudinal defect depth; Obtaining material properties and working environment characteristics of the target Y-type tee pipe fitting, performing corrosion intensity simulation analysis according to the material properties and working environment characteristics, and determining the corrosion intensity coefficient; The welding damage intensity is determined based on the transverse defect area ratio, transverse defect distribution dispersion, longitudinal defect depth and corrosion intensity coefficient.

4. The welding quality inspection method for LNG marine pipe fittings according to claim 3 is characterized in that: According to the welding damage intensity, the load-bearing performance evaluation channel is called to perform performance impact analysis, and the first potential performance impact coefficient is output, including: Pre-training a bearer performance evaluation channel, wherein the bearer performance evaluation channel includes P basis evaluators; Obtaining the maximum welding damage intensity in a predetermined historical time zone, and determining the number of calls of the base evaluator of the load-bearing performance evaluation channel according to the welding damage intensity and the maximum welding damage intensity, which is set to Q; Q selected base evaluators are randomly activated in the P base evaluators, and the performance impact analysis of the pipe defect feature space is performed using the Q selected base evaluators to output Q performance impact coefficients, and the first potential performance impact coefficient is obtained after mean calculation.

5. The welding quality inspection method for LNG marine pipe fittings according to claim 4 is characterized in that: Pre-training load performance evaluation channel, including: According to the historical maintenance log of the target Y-type tee pipe fitting, a three-dimensional feature distribution set of sample pipe fitting defects is collected, and the proportion of pipe fitting load-bearing failures in the three-dimensional feature distribution of different sample pipe fitting defects within a predetermined time window is statistically analyzed, and the proportion of sample pipe fitting load-bearing failures is obtained and set as the sample performance influence coefficient to obtain the sample performance influence coefficient set; Integrate the sample pipe defect three-dimensional feature distribution set and the sample performance influence coefficient set to obtain a sample training set, and divide it into P equal parts to obtain P training arrays; The generator and the discriminator of the generative adversarial network are trained using the P training arrays until P basis evaluators are obtained after convergence, and the bearing performance evaluation channel is obtained by integration.

6. The welding quality inspection method for LNG marine pipe fittings according to claim 1, characterized in that: Based on the standard pipe welding image of the target Y-type tee pipe fitting, a similar traversal comparison is performed on the pipe fitting surface image, and the second potential performance impact coefficient is determined according to the deviation area ratio, including: Performing welding area cropping on the standard pipe fitting welding image and the pipe fitting surface image respectively to obtain a standard welding area image and a pipe fitting welding area image; Performing grayscale processing on the standard welding area image and the pipe fitting welding area image to obtain a standard welding area grayscale image and a pipe fitting welding area grayscale image; Randomly selecting a first grayscale area in the grayscale image of the pipe welding area according to a predetermined size, wherein the predetermined size is a 5 by 5 pixel area; Performing a traversal similarity comparison between the first grayscale area and the area of ​​the predetermined size in the standard welding area grayscale image, and setting the maximum similarity as the first area similarity of the first grayscale area; Continue to randomly select regions and perform traversal similarity comparison in the grayscale image of the pipe welding region to obtain multiple grayscale regions and multiple region similarities; The grayscale area with a regional similarity less than a similarity threshold is set as a height difference area, and the area ratio of the height difference area to the grayscale image of the pipe welding area is calculated to obtain the deviation area ratio, and the second potential performance impact coefficient is determined according to the deviation area ratio.

7. The welding quality inspection method for LNG marine pipe fittings according to claim 1 is characterized in that: Evaluating and determining the damage intensity of the pipe fitting according to the first potential performance impact coefficient and the second potential performance impact coefficient includes: Configure a first influence weight and a second influence weight, wherein the first influence weight is 0.85 and the second influence weight is 0.15; Based on the first influence weight and the second influence weight, the first potential performance influence coefficient and the second potential performance influence coefficient are weightedly calculated to obtain the pipe damage strength.

8. The welding quality inspection platform for LNG marine pipe fittings is characterized by: The steps for implementing the welding quality detection method of LNG marine pipe fittings according to any one of claims 1 to 7 include: The defect feature distribution acquisition module is used to perform ray scanning and surface image acquisition on the target Y-type three-way pipe fitting, and perform internal defect recognition and surface defect recognition according to the pipe fitting ray image and the pipe fitting surface image, respectively, to obtain the internal defect feature distribution and the surface defect feature distribution; A defect feature space generation module is used to perform spatial fitting and maximum value retention processing on the internal defect feature distribution and the surface defect feature distribution, and to model and generate a pipe defect feature space; A first performance impact coefficient obtaining module is used to determine the welding damage intensity based on the pipe defect feature space analysis, and call the bearing performance evaluation channel to perform performance impact analysis according to the welding damage intensity, and output a first potential performance impact coefficient; A second performance impact coefficient obtaining module is used to perform similarity traversal comparison on the surface image of the target Y-type tee pipe fitting based on the standard pipe welding image of the target Y-type tee pipe fitting, and determine the second potential performance impact coefficient according to the deviation area ratio; The pipe fitting damage intensity determination module is used to determine the pipe fitting damage intensity based on the first potential performance impact coefficient and the second potential performance impact coefficient. If the pipe fitting damage intensity is greater than the risk intensity threshold, the target Y-type three-way pipe fitting is marked for maintenance.

Images