Ship liquid cargo stainless steel structure welding quality data processing method and system

Abstract

The present application provides a ship liquid cargo stainless steel structure welding quality data processing method and system, relates to the welding quality control technical field, the method comprises the following steps: collecting welding process parameters, nondestructive testing data and stainless steel material characteristic data, preprocessing the collected data to obtain a multi-source data set; the multi-source data set is matched and associated with the weld three-dimensional space coordinates, and a welding quality database is constructed; based on the preset stainless steel welding defect feature library, the nondestructive testing data in the welding quality database is intelligently identified to obtain a defect feature data set; based on the welding quality database, an initial virtual reference plane is constructed through the spatial coordinate data of the liquid cargo tank longitudinal and transverse bone intersection structure and the cabin wall boundary corner point position. The present application guarantees the corrosion resistance and crack resistance service safety of the liquid cargo tank welding structure.

Description

Technical Field

[0001] This invention relates to the field of welding quality control technology, and in particular to a method and system for processing welding quality data of stainless steel structures for liquid cargo ships. Background Technology

[0002] Ship cargo tanks mostly use austenitic stainless steel welded structures, which have extremely high and stringent requirements for the mechanical properties, resistance to intergranular corrosion, and resistance to hot cracking of weld joints. Traditional ship cargo stainless steel structure welding quality data processing generally suffers from multiple sources, heterogeneous and scattered welding process parameters, non-destructive testing data, and material property data. Various quality data cannot be accurately matched and correlated with the three-dimensional spatial coordinates of the weld. Welding defects rely on manual interpretation and identification, which is inefficient and has large errors. There is no adaptive calibration method for defect spatial location adapted to the complex structure of cargo tanks. It is difficult to build a digital twin mapping relationship of defects corresponding to the physical structure. Furthermore, multiple batches of quality data lack systematic statistical analysis and trend prediction, which makes it impossible to achieve full-process digital control of welding quality.

[0003] In the actual welding operation of the longitudinal and transverse skeleton intersection nodes and bulkhead boundary fillet welds of a 316L stainless steel cargo tank on a certain ocean-going chemical tanker, the traditional control mode of manually recording welding electrical parameters, offline interpretation of phased array ultrasonic and radiographic images, and relying solely on experience to judge defects is still used. The prominent technical defects are that multi-source quality data are not integrated and normalized, the matching accuracy between weld spatial coordinates and defect data is low, it is impossible to achieve adaptive calibration of defect spatial position by calculating the main direction of defect distribution, defects cannot be accurately mapped to the three-dimensional model of the cargo tank to form a digital twin feature layer, and multiple batches of defect data cannot be correlated with welding process parameters for regression analysis and quality risk prediction. This is prone to missed or misjudged defects such as hot cracks and intergranular corrosion, and it is impossible to identify batch welding quality hazards in advance. Summary of the Invention

[0004] This invention provides a method and system for processing welding quality data of stainless steel structures for liquid cargo ships, ensuring the corrosion resistance, crack resistance, and service safety of the welded structures in liquid cargo tanks.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] Firstly, a method for processing welding quality data of stainless steel structures for liquid cargo ships, the method comprising:

[0007] Welding process parameters, non-destructive testing data, and stainless steel material property data are collected. The collected data are preprocessed to obtain a multi-source dataset. The multi-source dataset is matched and correlated with the three-dimensional spatial coordinates of the weld to construct a welding quality database.

[0008] Based on a pre-set stainless steel welding defect feature library, the non-destructive testing data in the welding quality database is intelligently identified to obtain a defect feature dataset; based on the welding quality database, an initial virtual reference plane is constructed using the spatial coordinate data of the intersection of the longitudinal and transverse ribs of the liquid cargo tank and the corner points of the tank wall boundary.

[0009] By calculating the spatial coordinate covariance matrix of all defect points in the defect feature dataset, the eigenvectors of the spatial coordinate covariance matrix are solved, and the eigenvector corresponding to the largest eigenvalue is taken as the main direction of defect distribution. Based on the spatial angle between the main direction and the normal vector of the initial virtual reference plane, the initial virtual reference plane is rotated by a rotation transformation around an arbitrary axis in space to obtain an adaptive reference plane whose normal vector is perpendicular to the main direction.

[0010] Calculate the orthogonal projection point from each defect point in the defect feature dataset to the adaptive reference plane, and use the point set formed by all projection points as the calibrated defect spatial location; map the calibrated defect spatial location and the corresponding defect attributes to the 3D model of the liquid cargo tank to obtain the digital twin feature layer of the defect spatial distribution.

[0011] Statistical analysis was performed on the digital twin feature layers of the defect spatial distribution of multiple batches to obtain the analysis results, and a welding quality assessment report was automatically generated based on the analysis results.

[0012] Furthermore, welding process parameters, non-destructive testing data, and stainless steel material property data are collected. The collected data is preprocessed to obtain a multi-source dataset. This multi-source dataset is then matched and correlated with the three-dimensional spatial coordinates of the weld seam to construct a welding quality database, including:

[0013] Welding current, arc voltage, welding speed, argon flow rate, and interpass temperature data were collected to form a raw set of welding process parameters; waveforms, images, and numerical data were collected from phased array ultrasonic, X-ray, and eddy current testing equipment to form a raw set of nondestructive testing data; and chemical composition, mechanical properties, and intergranular corrosion tendency data were extracted from stainless steel material quality certificates to form a raw set of material properties.

[0014] By cleaning the raw sets of welding process parameters, non-destructive testing, and material properties, respectively, abnormal jump values ​​and noise data that deviate significantly from the threshold during the acquisition process are removed, and the timestamp format and measurement unit of each data source are unified, a standardized set of welding process parameters, non-destructive testing dataset, and material property dataset are obtained.

[0015] The standardized welding process parameter set, non-destructive testing dataset, and material property dataset are fused together, and the weld number is used as the primary key to match and associate it with the three-dimensional spatial coordinates of the weld extracted from the digital design drawings, forming a welding quality database that includes welding parameters, test results, material properties, and spatial coordinates.

[0016] Furthermore, based on a pre-defined stainless steel welding defect feature library, intelligent identification is performed on the non-destructive testing data in the welding quality database to obtain a defect feature dataset; based on the welding quality database, an initial virtual reference plane is constructed using the spatial coordinate data of the intersection of the longitudinal and transverse ribs of the liquid cargo tank and the corner points of the tank wall boundaries, including:

[0017] Phased array ultrasonic waveform data, radiographic image data, and eddy current numerical data associated with weld numbers are extracted from the welding quality database to form a non-destructive testing dataset to be identified.

[0018] The non-destructive testing dataset to be identified is input into a pre-set stainless steel welding defect feature library for feature matching. The defect feature library contains waveform features, image grayscale features and numerical threshold ranges of hot cracks, intergranular corrosion tendency and non-fusion defect types. By traversing and comparing, the defect type, size and spatial coordinates contained in the non-destructive testing data are identified, and a defect feature dataset is generated.

[0019] The coordinates of predefined longitudinal and transverse rib intersection nodes and bulkhead boundary corner points are extracted from the welding quality database and used as spatial reference feature points. An initial virtual reference plane is generated by fitting the three-dimensional coordinates of the spatial reference feature points.

[0020] Furthermore, by calculating the spatial coordinate covariance matrix of all defect points in the defect feature dataset, the eigenvectors of the spatial coordinate covariance matrix are solved, and the eigenvector corresponding to the largest eigenvalue is taken as the principal direction of the defect distribution. Based on the spatial angle between the principal direction and the normal vector of the initial virtual reference plane, the initial virtual reference plane is rotated around an arbitrary axis in space to obtain an adaptive reference plane whose normal vector is perpendicular to the principal direction, including:

[0021] Extract the three-dimensional spatial coordinates of all defect points from the defect feature dataset to form a defect point coordinate matrix, and calculate the covariance matrix of the defect point coordinate matrix to obtain the spatial coordinate covariance matrix.

[0022] The spatial coordinate covariance matrix is ​​decomposed into eigenvalues ​​to obtain three eigenvalues ​​and their corresponding eigenvectors. The eigenvector corresponding to the eigenvalue with the largest value is selected as the main direction of defect distribution. The main direction of defect distribution represents the maximum discrete extension direction of the defect point in three-dimensional space.

[0023] Extract the normal vector from the initial virtual reference plane, calculate the spatial angle between the principal direction of the defect distribution and the normal vector of the initial virtual reference plane, and determine the required rotation angle and rotation axis direction of the initial virtual reference plane based on the spatial angle.

[0024] Using the rotation axis as a reference, the initial virtual reference plane is rotated around any axis in space according to the spatial angle, so that the normal vector of the rotated plane is perpendicular to the main direction of the defect distribution. The plane after rotation transformation is used as the adaptive reference plane.

[0025] Furthermore, the orthogonal projection points from each defect point in the defect feature dataset to the adaptive reference plane are calculated, and the set of points formed by all projection points is used as the calibrated defect spatial location. The calibrated defect spatial location and the corresponding defect attributes are mapped onto the 3D model of the liquid cargo tank to obtain a digital twin feature layer of defect spatial distribution, including:

[0026] The three-dimensional spatial coordinates of each defect point are extracted from the defect feature dataset. The vertical distance from each defect point to the adaptive reference plane is calculated. The perpendicular coordinates of each defect point on the adaptive reference plane are solved based on the spatial equation of the adaptive reference plane. The perpendicular coordinates are used as the spatial position of the corresponding defect point after calibration. The calibration defect point set is constructed with the spatial positions of all defect points after calibration.

[0027] By extracting the defect type, defect size and original spatial coordinates corresponding to each defect point from the defect feature dataset as defect attribute information, the defect attribute information is associated one by one with the calibrated spatial location in the calibration defect point set to generate a calibration defect feature dataset containing the calibrated location and attribute information.

[0028] The calibration defect feature dataset is imported into the 3D model of the liquid cargo tank. Based on the spatial location after calibration, each defect point and attribute information is located to the corresponding coordinate position in the 3D model of the liquid cargo tank, forming a digital twin feature layer of defect spatial distribution in the 3D model of the liquid cargo tank that precisely corresponds to the spatial position of the structural entity.

[0029] Furthermore, the calibration defect feature dataset is imported into the 3D model of the liquid cargo tank. Based on the calibrated spatial location, each defect point and its attribute information is located to the corresponding coordinate position in the 3D model of the liquid cargo tank. This forms a digital twin feature layer in the 3D model of the liquid cargo tank, which precisely corresponds to the spatial location of the defect spatial distribution, including:

[0030] The underlying data structure of the 3D model of the liquid cargo tank is analyzed to obtain the geometric topology information and spatial index of all the bulkheads, longitudinal and transverse ribs and welded components contained in the 3D model of the liquid cargo tank in the model coordinate system; each calibrated spatial position in the calibration defect feature dataset is traversed, and the target welded component and the specific coordinate point on the target welded component corresponding to the calibrated spatial position are found and located in the 3D model of the liquid cargo tank through the spatial coordinate matching algorithm.

[0031] The target weld component is used as a defect attachment carrier. The corresponding defect attribute information is written into the extended data storage area of ​​the target weld component in the form of attribute fields. Defect identification nodes are generated at the coordinate points on the target weld component that correspond to the calibrated spatial position.

[0032] Based on the spatial distribution of each defect identifier node in the 3D model of the liquid cargo tank, defect identifier nodes belonging to the same weld are clustered and associated to form a defect cluster with spatial semantics. The defect cluster is then superimposed on the corresponding structural region of the 3D model of the liquid cargo tank as an independent layer to construct a digital twin feature layer of defect spatial distribution.

[0033] Furthermore, statistical analysis was performed on the digital twin feature layers of defect spatial distribution across multiple batches to obtain analysis results. Based on these results, a welding quality assessment report was automatically generated, including:

[0034] The defect spatial distribution digital twin feature layer corresponding to each batch is extracted from the welding quality database. The defect spatial distribution digital twin feature layer contains the calibrated spatial location, defect type, defect size and associated welding process parameters of each defect point.

[0035] By fusing data from multiple batches of extracted defect spatial distribution digital twin feature layers, a multi-batch defect dataset containing time, space, and attribute dimensions is constructed.

[0036] Statistical analysis was performed on multiple batches of defect datasets to calculate the overall density of defects in each batch, the proportion distribution of various types of defects, and the statistical characteristics of defect size. Furthermore, a correlation regression analysis was conducted between the spatial distribution of defects and welding process parameters to identify the correlation between abnormal fluctuations in welding parameters and defect generation.

[0037] Based on the correlation law, combined with the preset intergranular corrosion tendency threshold and hot crack sensitivity threshold, the welding quality of the current batch and subsequent batches is predicted to determine whether there is a batch quality risk.

[0038] The defect distribution patterns obtained from statistical analysis, the results of parameter correlation analysis, and the conclusions of quality risk prediction are structured and organized, and the data is automatically filled in according to the preset report template to generate a welding quality assessment report.

[0039] Secondly, the data processing system for welding quality of stainless steel structures for liquid cargo ships includes:

[0040] The acquisition module is used to collect welding process parameters, non-destructive testing data, and stainless steel material property data. The collected data is preprocessed to obtain a multi-source dataset. The multi-source dataset is matched and associated with the three-dimensional spatial coordinates of the weld to construct a welding quality database.

[0041] The identification module is used to intelligently identify non-destructive testing data in the welding quality database based on a preset stainless steel welding defect feature library to obtain a defect feature dataset; based on the welding quality database, an initial virtual reference plane is constructed using the spatial coordinate data of the intersection of the longitudinal and transverse ribs of the liquid cargo tank and the corner points of the tank wall boundary.

[0042] The calculation module is used to calculate the spatial coordinate covariance matrix of all defect points in the defect feature dataset, solve for the eigenvectors of the spatial coordinate covariance matrix, and take the eigenvector corresponding to the largest eigenvalue as the main direction of defect distribution. Based on the spatial angle between the main direction and the normal vector of the initial virtual reference plane, the initial virtual reference plane is rotated by a rotation transformation around an arbitrary axis in space to obtain an adaptive reference plane whose normal vector is perpendicular to the main direction.

[0043] The calibration module is used to calculate the orthogonal projection point from each defect point in the defect feature dataset to the adaptive reference plane, and use the point set formed by all projection points as the calibrated defect spatial location; the calibrated defect spatial location and the corresponding defect attributes are mapped to the 3D model of the liquid cargo tank to obtain the digital twin feature layer of defect spatial distribution.

[0044] The analysis module is used to perform statistical analysis on the digital twin feature layers of defect spatial distribution in multiple batches, obtain analysis results, and automatically generate welding quality assessment reports based on the analysis results.

[0045] Thirdly, a computing device, comprising:

[0046] One or more processors;

[0047] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0048] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0049] The above-described solution of the present invention has at least the following beneficial effects:

[0050] This approach overcomes the limitations of traditional shipboard stainless steel structure welding quality data processing, which involves: acquiring and preprocessing multi-source welding quality data and matching and associating it with the 3D spatial coordinates of the weld to construct a welding quality database; intelligently identifying defects based on a pre-set stainless steel welding defect feature library and constructing an initial virtual reference plane; calculating the covariance matrix of defect point spatial coordinates to solve for the principal direction of defect distribution and generating an adaptive reference plane through rotation transformation; orthogonally projecting and calibrating defect points and mapping them onto the 3D model of the liquid cargo tank to construct a digital twin feature layer; and statistically analyzing multiple batches of data to automatically generate welding quality assessment reports. This addresses the technical challenges of relying on manual methods, which are inefficient and prone to errors; adaptive calibration methods that cannot adapt to complex structures in terms of defect spatial location; difficulties in constructing digital twin mapping relationships for defects; and the lack of systematic statistical analysis and quality risk prediction for multiple batches of data. It achieves the unification and fusion of multi-source welding quality data and precise traceability, improving the efficiency and accuracy of defect identification. It enables precise calibration of the spatial location of defects in complex structural parts and visualization of defect distribution. It uncovers the correlation between welding parameters and defects and provides early warnings of batch quality risks. This enhances the digitalization and intelligence of welding quality control for stainless steel structures in marine liquid cargo tanks, ensuring the corrosion resistance and crack resistance of welded structures in liquid cargo tanks. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating the welding quality data processing method for stainless steel structures used in marine liquid cargo, provided in an embodiment of the present invention.

[0052] Figure 2 This is a schematic diagram of a welding quality data processing system for stainless steel structures used in marine liquid cargo, provided in an embodiment of the present invention. Detailed Implementation

[0053] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art.

[0054] like Figure 1 As shown, an embodiment of the present invention proposes a method for processing welding quality data of stainless steel structures for liquid cargo ships, the method comprising the following steps:

[0055] Step 1: Collect welding process parameters, non-destructive testing data, and stainless steel material property data. Preprocess the collected data to obtain a multi-source dataset. Match and associate the multi-source dataset with the three-dimensional spatial coordinates of the weld to construct a welding quality database.

[0056] Step 2: Based on the preset stainless steel welding defect feature library, intelligently identify the non-destructive testing data in the welding quality database to obtain a defect feature dataset; based on the welding quality database, construct an initial virtual reference plane using the spatial coordinate data of the intersection of the longitudinal and transverse ribs of the liquid cargo tank and the corner points of the tank wall boundary.

[0057] Step 3: Calculate the spatial coordinate covariance matrix of all defect points in the defect feature dataset, solve for the eigenvectors of the spatial coordinate covariance matrix, and take the eigenvector corresponding to the largest eigenvalue as the main direction of defect distribution; based on the spatial angle between the main direction and the normal vector of the initial virtual reference plane, rotate the initial virtual reference plane by a rotation transformation around an arbitrary axis in space to obtain an adaptive reference plane whose normal vector is perpendicular to the main direction.

[0058] Step 4: Calculate the orthogonal projection point from each defect point in the defect feature dataset to the adaptive reference plane, and use the point set formed by all projection points as the calibrated defect spatial location; map the calibrated defect spatial location and the corresponding defect attributes to the 3D model of the liquid cargo tank to obtain the digital twin feature layer of defect spatial distribution.

[0059] Step 5: Perform statistical analysis on the digital twin feature layers of defect spatial distribution in multiple batches, obtain the analysis results, and automatically generate a welding quality assessment report based on the analysis results.

[0060] In this embodiment of the invention, a welding quality database is constructed by collecting welding process parameters, non-destructive testing data, and stainless steel material property data, preprocessing them, and matching them with the three-dimensional spatial coordinates of the weld. Based on a pre-set stainless steel welding defect feature library, defects are intelligently identified to obtain a defect feature dataset. An initial virtual reference plane is constructed using the coordinates of the longitudinal and transverse rib intersection structure of the cargo tank and the corner points of the tank wall boundary. The principal direction of defect distribution is solved by calculating the covariance matrix of the spatial coordinates of defect points and then generated through rotation transformation to form an adaptive reference plane. Orthogonal projection calibration is performed on the defect points, and the calibrated defect positions and attributes are mapped to the three-dimensional model of the cargo tank to construct a digital twin feature layer. Statistical analysis is performed on multiple batches of digital twin feature layers, and a welding quality assessment report is automatically generated. This approach overcomes the multi-source issues in traditional ship liquid cargo stainless steel structure welding quality data processing. The technical challenges of heterogeneous and isolated data that cannot be accurately matched with the three-dimensional coordinates of welds, low efficiency and large errors due to reliance on manual defect identification, lack of adaptive calibration methods for defect spatial locations adapted to the complex structure of liquid cargo tanks, difficulty in constructing digital twin mapping relationships for defect spatial distribution, and lack of systematic statistical analysis of multiple batches of quality data that cannot automatically generate quality assessment reports have been addressed. This solution achieves the normalization and fusion of multi-source welding quality data and precise traceability, improving the efficiency and accuracy of defect identification, enabling precise calibration of defect spatial locations in complex structural parts, intuitively presenting the spatial distribution of defects, uncovering welding quality patterns, and automatically completing quality assessments. This effectively reduces the risk of missed or misjudged defects, provides early warning of batch quality hazards, enhances the digitalization and intelligence level of welding quality control for stainless steel structures in ship liquid cargo tanks, and ensures the service safety of welded structures in liquid cargo tanks.

[0061] In a preferred embodiment of the present invention, step 1 above may include:

[0062] Step 1.1: Collect welding current, arc voltage, welding speed, argon flow rate, and interpass temperature data to form a raw set of welding process parameters; collect waveform, image, and numerical data from phased array ultrasonic, X-ray, and eddy current testing equipment to form a raw set of nondestructive testing data; extract chemical composition, mechanical properties, and intergranular corrosion tendency data from stainless steel material quality certificates to form a raw set of material properties. Specifically, for the welding operation scenario of the longitudinal and transverse rib intersections and bulkhead fillet welds of 316L stainless steel cargo tanks on ocean-going chemical tankers, conduct full-dimensional raw data collection, construct three types of raw datasets, and collect relevant welding process parameters in real time during the welding operation, specifically including welding current. The arc voltage, welding speed, argon flow rate, and interpass temperature are arranged and stored in chronological order of data acquisition to form a complete raw set of welding process parameters. In the non-destructive testing (NDT) stage after welding, phased array ultrasonic testing equipment, X-ray testing equipment, and eddy current testing equipment are used to collect the corresponding waveform data, image data, and quantitative numerical data output by the testing equipment. These data are then organized and collected according to the weld inspection sequence to form a raw set of NDT parameters. From the factory quality certificate of 316L stainless steel, the chemical composition, mechanical properties, and intergranular corrosion tendency data of the material are extracted to complete the summary and organization of the core performance data of the material, forming a raw set of material properties.

[0063] Step 1.2 involves cleaning the raw sets of welding process parameters, non-destructive testing, and material properties separately, removing abnormal jump values ​​and noise data that significantly deviates from the threshold during the acquisition process, and unifying the timestamp format and unit of measurement for each data source to obtain standardized welding process parameter sets, non-destructive testing datasets, and material property datasets. Specifically, this includes: performing data cleaning and standardization on each of the raw sets of welding process parameters, non-destructive testing, and material properties separately, first removing abnormal noise data by threshold judgment, and then unifying the data format and unit of measurement.

[0064] Let X be a single piece of collected data, and let X be the preset upper limit threshold for this type of data. max The preset lower threshold is X. min The determination of abnormal data follows the calculation rules as follows: if the collected data X > X max Or collect data X < X minIf a data point is found to be an abnormal jump value or noise data that significantly deviates from the threshold during the data acquisition process, it will be directly removed from the original dataset. For welding operations on 316L stainless steel liquid cargo tanks, the specific thresholds for each parameter are set as follows: welding current upper limit: 180 amperes; lower limit: 80 amperes; arc voltage upper limit: 24 volts; lower limit: 10 volts; welding speed upper limit: 180 mm / min; lower limit: 60 mm / min; argon flow rate upper limit: 25 liters / min; lower limit: 10 liters / min; interpass temperature upper limit: 150 degrees Celsius; lower limit: 0 degrees Celsius; phased array ultrasonic detection echo amplitude upper limit: 100%; lower limit: 0%; X-ray detection image grayscale value upper limit: 255; lower limit: 0; eddy current detection impedance value upper limit: +50; lower limit: -50; 316L stainless steel yield strength lower limit: 170 MPa; tensile strength lower limit: 480 MPa. Data below these lower limits will be directly removed.

[0065] After removing noisy data, the timestamp format of all valid data was uniformly converted to the standard format of year-month-day hour:minute:second to eliminate time recording differences between different acquisition devices. At the same time, the units of measurement for all parameters were standardized: welding current was standardized to amperes, arc voltage to volts, welding speed to millimeters per minute, argon flow rate to liters per minute, temperature to degrees Celsius, and mechanical properties to megapascals. After data cleaning and format and unit standardization, a standardized set of welding process parameters, a standardized set of non-destructive testing data, and a standardized set of material property data were obtained.

[0066] Step 1.3 involves fusing standardized welding process parameter sets, non-destructive testing datasets, and material property datasets. Using the weld number as the primary key, a one-to-one matching and association is performed with the 3D spatial coordinates of the welds extracted from the digital design drawings. This forms a welding quality database containing welding parameters, test results, material properties, and spatial coordinates. Specifically, this includes: fusing the standardized welding process parameter sets, non-destructive testing datasets, and material property datasets from multiple sources, using the weld number as the unique primary key to achieve precise matching between data and spatial coordinates; extracting the 3D spatial coordinate data of all longitudinal and transverse rib intersection welds and bulkhead boundary fillet welds from the digital design drawings of the ship's liquid cargo tanks; and matching the standardized welding process parameters, non-destructive testing results, and material property data corresponding to the same weld number with the corresponding 3D spatial coordinates to achieve a one-to-one binding and integration of welding parameters, test results, material properties, and spatial location. Finally, an integrated welding quality database containing complete welding process parameters, non-destructive testing results, 316L stainless steel material properties, and weld 3D spatial coordinates is constructed.

[0067] In this embodiment of the invention, a technical approach is adopted to overcome the technical problems of traditional ship liquid cargo stainless steel structure welding quality data collection processes. This approach involves: collecting welding current, arc voltage, welding speed, argon flow rate, and interpass temperature to form a raw set of welding process parameters; collecting corresponding detection data from phased array ultrasonic, X-ray, and eddy current testing equipment to form a raw set of non-destructive testing data; and extracting relevant data from stainless steel material quality certification documents to form a raw set of material properties. The three raw sets are then cleaned to remove abnormal noise data and to standardize timestamp formats and units of measurement. Finally, the standardized three datasets are merged, with the weld number as the primary key, and matched one-to-one with the weld three-dimensional spatial coordinates extracted from digital design drawings to construct a welding quality database. This approach overcomes the technical problems of traditional ship liquid cargo stainless steel structure welding quality data collection processes, such as scattered and isolated raw data, abnormal noise, inconsistent formats and units of measurement, inability to accurately match multi-source data with weld three-dimensional spatial coordinates, and lack of an integrated quality data storage carrier. This results in the systematic collection and standardized processing of multi-source welding quality raw data, eliminating data noise interference, achieving unified fusion of multi-source data and accurate binding of weld spatial coordinates, and constructing an integrated welding quality database covering welding parameters, detection results, material properties, and spatial coordinates.

[0068] In a preferred embodiment of the present invention, step 2 above may include:

[0069] Step 2.1: Extract phased array ultrasonic waveform data, radiographic image data, and eddy current testing numerical data associated with the weld number from the welding quality database to form the non-destructive testing dataset to be identified. Specifically, using the weld number as the unique retrieval identifier, extract all non-destructive testing data associated with the current weld number to be evaluated from the constructed welding quality database, thus completing the construction of the non-destructive testing dataset to be identified. In practice, iterate through the data entries stored in the welding quality database, filter out records that completely match the current weld number, and extract the corresponding phased array ultrasonic waveform data, radiographic image data, and eddy current testing numerical data for that weld. The phased array ultrasonic waveform data includes defect echo amplitude and sound wave propagation time data, the radiographic image data includes grayscale pixel difference data of the weld area, and the eddy current testing numerical data includes impedance change rate quantization data. Integrate and collect the above three types of non-destructive testing data bound to the same weld number, remove irrelevant redundant data and null data, and form a non-destructive testing dataset specifically for intelligent identification of defects in the current weld.

[0070] Step 2.2: Input the non-destructive testing dataset to be identified into the preset stainless steel welding defect feature library for feature matching. The defect feature library pre-stores waveform features, image grayscale features, and numerical threshold ranges for hot cracks, intergranular corrosion tendency, and non-fusion defects. By traversing and comparing, the defect types, sizes, and spatial coordinates contained in the non-destructive testing data are identified, and a defect feature dataset is generated. Specifically, this includes: completing automatic traversal comparison and identification of defects through the preset stainless steel welding defect feature library. The feature library pre-stores the feature thresholds and feature parameters of three common defects in 316L stainless steel welding: hot cracks, intergranular corrosion tendency, and non-fusion. All parameters are specific values ​​calibrated in actual welding operations.

[0071] Hot crack characteristics: Phased array ultrasonic echo amplitude threshold ≥80%, X-ray inspection image grayscale difference ≥40, eddy current detection impedance change rate ≥15%; Intergranular corrosion tendency characteristics: Phased array ultrasonic echo amplitude threshold between 45% and 75%, X-ray inspection image grayscale difference between 20 and 35, eddy current detection impedance change rate between 8% and 12%; Incomplete fusion defect characteristics: Phased array ultrasonic echo amplitude threshold between 25% and 40%, X-ray inspection image grayscale difference between 10 and 18, eddy current detection impedance change rate between 3% and 6%.

[0072] The phased array ultrasonic echo amplitude is set to be in the data to be identified. The grayscale difference value detected by X-ray is The rate of change of eddy current detection impedance is The amplitude determination threshold for the corresponding defect type in the defect feature library is: The grayscale threshold is The threshold for determining the rate of change of impedance is The defect type traversal and comparison follows the following logical judgment formula: if the condition is satisfied... At the same time satisfy and satisfy If the current detection data corresponds to a location with a target type defect, then the defect type is determined to exist. Following the above determination rules, each data set in the non-destructive testing dataset is iterated and compared sequentially to accurately determine the defect type. Simultaneously, the spatial positioning parameters of the detection data are combined to quantify the defect size. The specific formula for calculating the defect size is as follows: The actual length of the defect is set using letters. The propagation time of the defect echo coverage is indicated by letters. The speed of ultrasonic waves in 316L stainless steel is indicated by the letter […]. The calibration coefficients of the testing equipment are indicated by letters. It is stated that the propagation speed of ultrasound in 316L stainless steel is 5900 meters per second, and the equipment calibration coefficient is 1.0. The defect size is accurately quantified by this formula, and the three-dimensional spatial coordinates corresponding to the defect are extracted simultaneously. The identified defect type, defect quantification size, and defect three-dimensional spatial coordinates are integrated to generate a standardized defect feature dataset.

[0073] Step 2.3: Extract the coordinates of predefined longitudinal and transverse rib intersection nodes and bulkhead boundary corner points from the cargo tank structural design data in the welding quality database as spatial reference feature points. Generate an initial virtual reference plane based on the three-dimensional coordinates of these spatial reference feature points. Specifically, this includes: using key structural nodes of the cargo tank as spatial references, generating an initial virtual reference plane through three-dimensional coordinate fitting to provide a basic spatial reference for subsequent determination of the main direction of defect spatial distribution and generation of the adaptive reference plane; retrieving cargo tank structural design data from the welding quality database; extracting the three-dimensional coordinates of predefined longitudinal and transverse rib intersection nodes and bulkhead boundary corner points; and selecting four spatial reference feature points at typical locations of 316L stainless steel cargo tanks on ocean-going chemical tankers. Their specific three-dimensional coordinates are as follows:

[0074] First spatial reference feature point: =1200 mm, =800 mm, =0 mm;

[0075] Second spatial reference feature point: =1200 mm, =0 mm, =0 mm;

[0076] Third-space reference feature points: =0 mm, =0 mm, =0 mm;

[0077] Fourth spatial reference feature point: =0 mm, =800 mm, =0 mm;

[0078] The general mathematical equation for a spatial plane is: Where K, L, M, and N are the undetermined coefficients of the plane equation. , , Let the three-dimensional coordinates of any point in space be denoted as . Substitute the three-dimensional coordinates of the four spatial reference feature points into the general plane equations in sequence to construct a system of four linear equations:

[0079] First equation: ;

[0080] The second equation: ;

[0081] The third equation: ;

[0082] The fourth equation: ;

[0083] After substituting the coordinate values ​​of each feature point into the system of simultaneous equations, the system of equations is solved uniformly using the matrix solution method, and the coefficients of each term of the plane equation are calculated: K=0, L=0, M=1, N=0.

[0084] Substituting the coefficients into the general equation of the plane, the final equation of the initial virtual reference plane is Z=0. This plane is perfectly aligned with the installation reference plane of the longitudinal and transverse rib intersection structure and the bulkhead boundary structure of the liquid cargo tank, thus completing the accurate fitting and generation of the initial virtual reference plane.

[0085] In this embodiment of the invention, a non-destructive testing dataset to be identified is constructed by extracting phased array ultrasonic waveform data, X-ray image data, and eddy current numerical data associated with weld numbers from a welding quality database. This dataset is then input into a pre-stored stainless steel welding defect feature library containing waveform features, image grayscale features, and numerical threshold ranges for hot cracks, intergranular corrosion tendencies, and non-fusion defect types for comparison. This process identifies the defect type, size, and spatial coordinates, generating a defect feature dataset. Simultaneously, the coordinates of the longitudinal and transverse rib intersections and bulkhead corners of the cargo tank are extracted from the welding quality database as spatial reference feature points. An initial virtual reference plane is generated based on these three-dimensional coordinates. This approach overcomes the technical problems of traditional shipboard stainless steel structure welding defect identification, which relies on manual interpretation, is inefficient, has large errors, cannot accurately identify the specific type, size, and spatial coordinates of defects, and lacks an initial spatial reference benchmark adapted to the complex structure of the cargo tank, making it difficult to support defect spatial calibration. This approach achieves intelligent automatic identification of welding defects, improves the efficiency and accuracy of defect identification, and clarifies the core attributes and spatial location of defects.

[0086] In a preferred embodiment of the present invention, step 3 above may include:

[0087] Step 3.1: Extract the three-dimensional spatial coordinates of all defect points from the defect feature dataset to form a defect point coordinate matrix. Calculate the covariance matrix of the defect point coordinate matrix to obtain the spatial coordinate covariance matrix. Specifically, this includes: First, extracting the three-dimensional coordinates of defect points. From the previously generated defect feature dataset, extract the three-dimensional spatial coordinates of all identified defect points. Based on the actual welding defects in the 316L stainless steel cargo tank of the ocean-going chemical tanker, select six typical defect points as calculation samples, covering two typical defects: hot cracking and lack of fusion. The specific three-dimensional coordinate unit is millimeters, and they are: the first defect point has a horizontal coordinate of 1100, a vertical coordinate of 700, and a vertical coordinate of 5. The second defect point has horizontal coordinates of 1050, vertical coordinates of 650, and vertical coordinates of 6; the third defect point has horizontal coordinates of 1000, vertical coordinates of 600, and vertical coordinates of 4; the fourth defect point has horizontal coordinates of 950, vertical coordinates of 550, and vertical coordinates of 7; the fifth defect point has horizontal coordinates of 900, vertical coordinates of 500, and vertical coordinates of 5; and the sixth defect point has horizontal coordinates of 850, vertical coordinates of 450, and vertical coordinates of 6. The defect point coordinates are then organized and arranged in the order of their occurrence, forming a defect point coordinate set containing 6 sets of 3D coordinates. Each set of coordinates corresponds to the horizontal, vertical, and spatial positions of a defect point.

[0088] Next, we calculate the spatial coordinate covariance matrix. The covariance matrix describes the dispersion and correlation between the horizontal, vertical, and axial components in three-dimensional coordinates. It is a 3x3 matrix. We calculate the coordinate mean vector, obtaining the average values ​​of the horizontal, vertical, and axial coordinates for all defect points. The average horizontal coordinate is 975 mm; the average vertical coordinate is 575 mm; and the average axial coordinate is 5.5 mm. We then calculate the coordinate deviation vector for each defect point by subtracting the average coordinate values ​​in the corresponding direction from the horizontal, vertical, and axial coordinates of each defect point. For example, for the first defect point, the horizontal deviation is 1100 - 975 = 125, and the vertical deviation is 700 - 575 = 12. 5. The vertical deviation is 5 - 5.5 = -0.5. Following this method, calculate the coordinate deviation vectors of all 6 defect points sequentially. Calculate each element of the covariance matrix. Multiply each value of the coordinate deviation vector by its corresponding value, sum the results, and then divide by the number of defect points minus one, which is 5. Calculate the covariance values ​​for horizontal to horizontal, horizontal to vertical, horizontal to vertical, vertical to horizontal, vertical to vertical, vertical to vertical, vertical to horizontal, vertical to vertical, and vertical to vertical in sequence. This results in a three-row, three-column spatial coordinate covariance matrix. Each value in the matrix is ​​rounded to one decimal place. The values ​​are: first row 8125.0, 8125.0, 12.5; second row 8125.0, 8125.0, 12.5; third row 12.5, 12.5, 0.7.

[0089] Step 3.2 involves performing eigenvalue decomposition on the spatial coordinate covariance matrix to obtain three eigenvalues ​​and their corresponding eigenvectors. The eigenvector corresponding to the eigenvalue with the largest value is selected as the main direction of defect distribution. The main direction of defect distribution represents the maximum discrete extension direction of the defect point in three-dimensional space. Specifically, this includes: performing eigenvalue decomposition on the spatial coordinate covariance matrix obtained in the previous calculation to obtain three sets of eigenvalues ​​and their corresponding eigendirection vectors; selecting the eigendirection vector corresponding to the eigenvalue with the largest value as the main direction of defect distribution; clarifying the maximum discrete extension direction of the defect point in three-dimensional space; clarifying the eigenvalue decomposition principle; performing eigenvalue decomposition on the symmetric covariance matrix of three rows and three columns to obtain three sets of real eigenvalues ​​and eigendirection vectors corresponding to each set of eigenvalues. The three sets of eigendirection vectors are perpendicular to each other.

[0090] Specific feature values ​​and feature direction vectors were calculated. Combining the previously obtained spatial coordinate covariance matrix, eigenvalue decomposition was performed to obtain three sets of feature values ​​and corresponding feature direction vectors. All feature direction vectors were normalized, and the values ​​were retained to three decimal places. The first set of feature values ​​was 16250.7, and the corresponding feature direction vectors were 0.707 horizontally, 0.707 vertically, and 0.001 vertically. The second set of feature values ​​was 0.0, and the corresponding feature direction vectors were -0.707 horizontally, 0.707 vertically, and 0.000 vertically. The third set of feature values ​​was 0.7, and the corresponding feature direction vectors were 0.000 horizontally, 0.000 vertically, and 1.000 vertically.

[0091] The main direction of defect distribution is determined, and the magnitudes of the three sets of feature values ​​are compared. The first set of feature values, 16250.7, is greater than the third set of feature values, 0.7. The third set of feature values, 0.7, is greater than the second set of feature values, 0.0. The feature direction vector corresponding to the first set of feature values ​​with the largest value is selected as the main direction of defect distribution. This main direction of defect distribution accurately represents the maximum discrete extension direction of the six defect points in three-dimensional space.

[0092] Step 3.3: Extract the normal vector from the initial virtual reference plane, calculate the spatial angle between the main direction of the defect distribution and the normal vector of the initial virtual reference plane, and determine the required rotation angle and rotation axis direction of the initial virtual reference plane based on the spatial angle. Specifically, this includes: extracting the normal vector of the initial virtual reference plane. According to the previous calculation results, the initial virtual reference plane is a plane with a vertical coordinate that is always equal to 0. The normal vector of the spatial plane is composed of the coefficients of the horizontal, vertical, and longitudinal directions in the plane equation. Therefore, the normal vector of the initial virtual reference plane is 0 in the horizontal direction, 0 in the vertical direction, and 1 in the longitudinal direction, and is a unit vector with a length of 1. Calculate the spatial angle between the main direction of the defect distribution and the normal vector. This angle is used to characterize the degree of spatial deviation between the two vectors. The calculation method is to divide the absolute value of the dot product of the two vectors by the product of the lengths of the two vectors themselves. First, calculate the dot product of the main direction of the defect distribution and the normal vector of the initial plane. Multiply the corresponding values ​​of the two vectors and add them together; the result is 0.001. Then, calculate the length of the main direction of the defect distribution itself; since the vector has been normalized, the length is 1. Finally, substitute the values ​​into the calculation to obtain the cosine of the angle, which is 0.001. The spatial angle is calculated to be approximately 89.94 degrees, and the value is rounded to two decimal places. Determine the rotation angle and the direction of the rotation axis. The angle that the initial virtual reference plane needs to rotate is the 89.94 degrees calculated above. The goal of the rotation is to make the normal vector of the rotated plane perpendicular to the main direction of the defect distribution. The direction of the rotation axis is selected to be perpendicular to both the main direction of the defect distribution and the normal vector of the initial plane. The rotation axis vector is calculated by the cross product of the two vectors. The final rotation axis vector is 0.707 horizontally, -0.707 vertically, and 0 vertically. This rotation axis is perpendicular to the initial virtual reference plane and also perpendicular to the main direction of the defect distribution, ensuring that the plane normal vector is perpendicular to the main direction of the defect distribution after the rotation transformation.

[0093] Step 3.4: Using the rotation axis as a reference, perform a rotation transformation around an arbitrary spatial axis on the initial virtual reference plane according to the spatial angle, so that the normal vector of the rotated plane is perpendicular to the main direction of the defect distribution. Use the plane after rotation transformation as the adaptive reference plane. Specifically, this includes: using the previously determined rotation axis and rotation angle as a reference, perform a rotation transformation around an arbitrary spatial axis on the initial virtual reference plane to generate an adaptive reference plane, ensuring that the normal vector of the adaptive reference plane is perpendicular to the main direction of the defect distribution, adapting to the actual distribution characteristics of the defects, constructing a spatial rotation transformation matrix, and constructing a three-row, three-column rotation transformation matrix according to the transformation rules of rotating around an arbitrary spatial axis by a specified angle, combined with the horizontal, vertical, and longitudinal values ​​of the rotation axis, as well as the cosine and sine values ​​of the rotation angle. The matrix values ​​are retained to three decimal places. In this embodiment, the cosine value of the rotation angle is approximately 0.001, and the sine value is approximately 1.000. The resulting rotation transformation matrix has the following values: first row: 0.501, -0.499, 0.707; second row: -0.499, 0.501, 0.707; third row: -0.707, -0.707, 0.001. A rotation transformation is performed on the initial virtual reference plane. The normal vector of the initial virtual reference plane is calculated with the rotation transformation matrix to obtain the normal vector of the adaptive reference plane after rotation. The calculation result is 0.707 horizontally, 0.707 vertically, and 0.001 vertically. This normal vector is completely consistent with the main direction of the defect distribution. At this time, the adaptive reference plane is perpendicular to the main direction of the defect distribution, which perfectly matches the spatial distribution characteristics of the defect.

[0094] An adaptive reference plane is generated by selecting a reference point on the initial virtual reference plane. The coordinates of this reference point are 0 in the horizontal direction, 0 in the vertical direction, and 0 in the center direction. The coordinates of the reference point remain unchanged after rotation transformation. Combined with the rotated normal vector, the spatial plane point normal equation is used. ,in, , , It is the normal vector. , , As a reference point on the plane, an adaptive reference plane is constructed. The normal vector value and the coordinates of the reference point are substituted into the equation to obtain the plane equation of the adaptive reference plane. This adaptive reference plane accurately fits the defect distribution characteristics of the weld seam at the intersection of the longitudinal and transverse ribs of the liquid cargo tank.

[0095] In this embodiment of the invention, a defect point coordinate matrix is ​​constructed by extracting the three-dimensional spatial coordinates of the defect points. The eigenvalues ​​and eigenvectors of the spatial coordinate covariance matrix are calculated and solved. The eigenvector corresponding to the largest eigenvalue is determined as the main direction of the defect distribution. Then, the spatial angle between this main direction and the normal vector of the initial virtual reference plane is calculated to determine the rotation angle and rotation axis. A rotation transformation is then performed on the initial plane to finally obtain an adaptive reference plane whose normal vector is perpendicular to the main direction of the defect distribution. This technique overcomes the technical problems in traditional welding defect spatial analysis, such as the inability to quantify the true distribution trend of defects, the mismatch between the initial reference plane and the actual defect distribution, the difficulty in adapting to the complex spatial characteristics of liquid cargo tank structures, and the lack of adaptive benchmarks for defect space calibration. This achieves the objective quantification of the spatial distribution law of defects, enabling the reference plane to adaptively match the actual distribution characteristics of defects.

[0096] In a preferred embodiment of the present invention, step 4 above may include:

[0097] Step 4.1: Extract the three-dimensional spatial coordinates of each defect point from the defect feature dataset. Calculate the perpendicular distance from each defect point to the adaptive reference plane. Based on the spatial equation of the adaptive reference plane, solve for the perpendicular coordinates of each defect point on the adaptive reference plane. Use these perpendicular coordinates as the calibrated spatial position of the corresponding defect point. Construct a calibration defect point set using the calibrated spatial positions of all defect points. Specifically, this includes: extracting the three-dimensional coordinates of defect points. Extract the original three-dimensional spatial coordinates of all six typical defect points from the previously generated defect feature dataset, ensuring consistency with the defect points selected in previous steps to guarantee data continuity. The coordinates are in millimeters: First defect point: horizontal coordinate 1100, vertical coordinate 700, vertical coordinate 5; Second defect point: horizontal coordinate 1050, vertical coordinate 650, vertical coordinate 6; Third defect point: horizontal coordinate 1000, vertical coordinate 600, vertical coordinate 4; Fourth defect point: horizontal coordinate 950, vertical coordinate 550, vertical coordinate 7; Fifth defect point: horizontal coordinate 900, vertical coordinate 500, vertical coordinate 5; Sixth defect point: horizontal coordinate 850, vertical coordinate 450, vertical coordinate 6. The equation of the adaptive reference plane is confirmed. Combined with previous calculation results, the final equation of the adaptive reference plane is: ,in, For the horizontal coordinate, For vertical coordinates, Using vertical coordinates, calculate the perpendicular distance from the defect point to the adaptive reference plane. There is a fixed method for calculating the perpendicular distance from any point in space to the plane. Taking the first defect point as an example, the remaining five defect points can be calculated using the same method. The calculation formula is as follows: , A=0.707, B=0.707, C=0.001, D=0, , , The original three-dimensional coordinates of the defect point. The vertical distance from the defect point to the plane is calculated by substituting the data of the first defect point into the above formula, resulting in a value of approximately 1272.6 mm. The vertical distances of the remaining five defect points are calculated using the same method, as follows: Defect point 2: approximately 1207.6 mm; Defect point 3: approximately 1142.6 mm; Defect point 4: approximately 1077.7 mm; Defect point 5: approximately 1012.6 mm; Defect point 6: approximately 947.6 mm.

[0098] To find the foot coordinates of the defect point on the adaptive reference plane, we need to determine the coordinates of the point where the perpendicular line drawn from the defect point to the adaptive reference plane intersects the plane. The solution involves combining the equation of the adaptive reference plane and the direction of the perpendicular line, ensuring the direction of the perpendicular line is parallel to the plane's normal vector. Taking the first defect point as an example, we calculate the scaling factor by summing the numerator and dividing it by the sum of the denominators: 1272.605 ÷ 1.000 = 1272.605. Then, we solve for the components of the vertical coordinate and the horizontal coordinate. The original horizontal coordinate of the defect point is equal to the original horizontal coordinate minus the first plane coefficient multiplied by the scaling factor, i.e., (1100 - 0.707) × 1272.605 = 199.3 mm; the original vertical coordinate is equal to the original vertical coordinate of the defect point minus the second plane coefficient multiplied by the scaling factor, i.e., (700 - 0.707) × 1272.605 = -200.7 mm; the original vertical coordinate is equal to the original vertical coordinate of the defect point minus the third plane coefficient multiplied by the scaling factor, i.e., (5 - 0.001) × 1272.605 = 3.7 mm. The vertical coordinates of the first defect point, i.e., its calibrated spatial position, are calculated using the same method: horizontal 199.3 mm, vertical -200.7 mm, and vertical 3.7 mm. The values ​​are rounded to one decimal place. The calibrated spatial positions of the remaining five defect points are calculated as follows: the second defect point is horizontal 164.3 mm, vertical -235.7 mm, and vertical 4.7 mm; the third defect point is horizontal 129.3 mm, vertical -270.7 mm, and vertical 3.7 mm; the fourth defect point is horizontal... The first defect point has a horizontal dimension of 94.3 mm, a vertical dimension of -305.7 mm, and a vertical dimension of 6.7 mm; the second defect point has a horizontal dimension of 59.3 mm, a vertical dimension of -340.7 mm, and a vertical dimension of 4.7 mm; the third defect point has a horizontal dimension of 24.3 mm, a vertical dimension of -375.7 mm, and a vertical dimension of 5.7 mm. The spatial positions of the six defect points after calibration, that is, the vertical coordinates of each defect point, are arranged and collected in the order of the defect point numbers to form a complete set of calibrated defect points. This set contains the three-dimensional spatial coordinates of all defect points after precise calibration.

[0099] Step 4.2 involves extracting the defect type, defect size, and original spatial coordinates corresponding to each defect point from the defect feature dataset as defect attribute information. This defect attribute information is then associated one-to-one with the calibrated spatial location in the calibration defect point set to generate a calibration defect feature dataset containing the calibrated location and attribute information. Specifically, this includes: extracting defect attribute information by extracting the attribute information of each of the six defect points from the previously generated defect feature dataset. The attribute information of each defect point includes the defect type, defect size, and original three-dimensional spatial coordinates, combined with the actual welding defect identification results.

[0100] The first defect is a hot crack. The defect size, or length, is determined using a pre-defined calculation method. The specific calculation process is as follows: the defect echo propagation time is 0.0002 seconds, the ultrasonic wave propagation speed in 316L stainless steel is 5900 meters per second, and the calibration coefficient of the detection equipment is 1.0. Multiplying the defect echo propagation time by the ultrasonic wave propagation speed and then by the equipment calibration coefficient, we get 0.0002 × 5900 × 1.0 = 1.18 mm. The original spatial coordinates of this defect are (1100, 700, 5).

[0101] The second defect point is a hot crack. The defect echo coverage propagation time is 0.00022 seconds. Using the same calculation method, with 0.00022×5900×1.0, the calculation result is 1.298 mm. The value is rounded to three decimal places. The original spatial coordinates are (1050, 650, 6).

[0102] The third defect point is of the defect type of non-fusion. The defect echo coverage propagation time is 0.00018 seconds. Calculate 0.00018 × 5900 × 1.0 = 1.062 mm. Keep three decimal places. The original spatial coordinates are (1000, 600, 4).

[0103] The fourth defect point is a hot crack. The defect echo coverage propagation time is 0.00025 seconds. The value is 0.00025 × 5900 × 1.0 = 1.475 mm. The value is rounded to three decimal places. The original spatial coordinates are (950, 550, 7).

[0104] The fifth defect point is of the defect type of non-fusion. The defect echo coverage propagation time is 0.00017 seconds. Use 0.00017×5900×1.0=1.003 mm. Keep three decimal places. The original spatial coordinates are (900, 500, 5).

[0105] The defect type of the sixth defect point is non-fusion, and the defect echo coverage propagation time is 0.00019 seconds. The value is 0.00019 × 5900 × 1.0 = 1.121 mm. The value is rounded to three decimal places. The original spatial coordinates are (850, 450, 6).

[0106] A one-to-one association between defect attributes and calibration locations is established, using the defect point number as a unique identifier. The attribute information of each defect point, including defect type, defect size, and original 3D spatial coordinates, is bound to the calibrated spatial location of that defect point. This ensures that each defect point corresponds to a complete and unique set of attribute information and calibration location data. The specific association is as follows:

[0107] The first defect point, after calibration, is located at 199.3 mm horizontally, -200.7 mm vertically, and 3.7 mm vertically, corresponding to the defect type of hot crack, defect size of 1.18 mm, and original spatial coordinates of 1100 mm horizontally, 700 mm vertically, and 5 mm vertically.

[0108] The second defect point, after calibration, is located at 164.3 mm horizontally, -235.7 mm vertically, and 4.7 mm vertically, corresponding to the defect type of hot crack, defect size of 1.298 mm, and original spatial coordinates of 1050 mm horizontally, 650 mm vertically, and 6 mm vertically.

[0109] The third defect point, after calibration, is located at 129.3 mm horizontally, -270.7 mm vertically, and 3.7 mm vertically, corresponding to the defect type of non-fusion, defect size of 1.062 mm, and original spatial coordinates of 1000 mm horizontally, 600 mm vertically, and 4 mm vertically.

[0110] The fourth defect point, after calibration, is located at 94.3 mm horizontally, -305.7 mm vertically, and 6.7 mm vertically, corresponding to the defect type of hot crack, defect size of 1.475 mm, and original spatial coordinates of 950 mm horizontally, 550 mm vertically, and 7 mm vertically.

[0111] The fifth defect point, after calibration, is located at 59.3 mm horizontally, -340.7 mm vertically, and 4.7 mm vertically, corresponding to the defect type of non-fusion, defect size of 1.003 mm, and original spatial coordinates of 900 mm horizontally, 500 mm vertically, and 5 mm vertically.

[0112] The sixth defect point, after calibration, is located at 24.3 mm horizontally, -375.7 mm vertically, and 5.7 mm vertically. The corresponding defect type is non-fusion, the defect size is 1.121 mm, and the original spatial coordinates are 850 mm horizontally, 450 mm vertically, and 6 mm vertically.

[0113] The six sets of associated defect data were integrated and summarized, arranged sequentially according to the defect point numbers, to form a standardized calibration defect feature dataset. This dataset contains the calibrated spatial location, defect type, defect size, and original spatial coordinates of each defect point, and the data is complete and accurately correlated.

[0114] Step 4.3: Import the calibration defect feature dataset into the 3D model of the cargo tank. Based on the calibrated spatial location, locate each defect point and attribute information to the corresponding coordinate position in the 3D model of the cargo tank. In the 3D model of the cargo tank, form a digital twin feature layer of defect spatial distribution that accurately corresponds to the spatial position of the structural entity. Specifically, this includes: parsing the bottom layer data of the 3D model of the cargo tank, retrieving the 3D model of the 316L stainless steel cargo tank of the ocean-going chemical tanker, and then parsing the bottom layer data structure of the 3D model. Extract the geometric topology information of all bulkheads, longitudinal and transverse ribs and welded components in the model, including the shape, size and connection relationship of each component. At the same time, extract the spatial index of all components in the model coordinate system. The spatial index is used to quickly locate the specific position of the component in the 3D model to ensure that the data of the 3D model is completely consistent with the structure of the actual cargo tank. The spatial index of the weld at the intersection of the longitudinal and transverse ribs and the corner weld at the boundary of the bulkhead corresponds one-to-one with the previously extracted 3D coordinates of the weld. The calibration defect feature dataset was imported into the 3D model of the liquid cargo tank. Using the calibrated spatial position of each defect point as a reference, a spatial coordinate matching algorithm was used to find and locate the target weld component corresponding to each defect point in the 3D model, as well as the specific coordinate point on that component. Taking the first defect point as an example, the calibrated position of this defect point is 199.3 mm horizontally, -200.7 mm vertically, and 3.7 mm vertically. By comparing the spatial index of all weld components in the 3D model, the target weld component corresponding to this defect point was determined to be the cross joint weld of the longitudinal and transverse ribs, with the weld number HJ-001. The defect point was then accurately located at the specific position of 199.3 mm horizontally, -200.7 mm vertically, and 3.7 mm vertically on this weld. The remaining 5 defect points were located using the same method, and finally, they were all determined to correspond to the cross joint weld of the longitudinal and transverse ribs, HJ-001. This is consistent with the actual distribution scenario of welding defects, ensuring the accuracy of defect location.

[0115] The located target weld component is used as the attachment carrier for defects. The attribute information corresponding to each defect point, including defect type, defect size, original spatial coordinates, and calibrated spatial coordinates, is written into the extended data storage area of ​​the target weld component as attribute fields, thus binding the defect attribute information to the weld component. Simultaneously, visual defect identification nodes are generated on the target weld component at the coordinate points corresponding to the calibrated spatial positions of the defect points. Each defect identification node corresponds to a defect point. To intuitively distinguish defect types, different colors are used for the identification nodes: red for hot crack defects and yellow for non-fusion defects, clearly and intuitively presenting the specific location and type of the defect. Based on the spatial coordinates of each defect identification node in the 3D model of the liquid cargo tank... To assess the spatial distribution of defects, a clustering algorithm is used to cluster and associate six defect identifier nodes belonging to the same weld HJ-001, forming a defect cluster with spatial semantics. This defect cluster visually presents the distribution of all defects on the weld, facilitating operators' overall understanding of the weld's defect situation. This defect cluster is then overlaid as an independent layer onto the corresponding structural area of ​​the liquid cargo tank's 3D model, namely the weld area at the intersection of longitudinal and transverse ribs, thereby constructing a digital twin feature layer of the defect spatial distribution. This digital twin feature layer precisely corresponds to the spatial position of the liquid cargo tank's physical structure, intuitively and clearly displaying all relevant information such as the spatial distribution, defect type, and defect size of the defects. Operators can quickly view the details of each defect through the 3D model, achieving visualized control of defects.

[0116] In this embodiment of the invention, a calibration defect point set is constructed by extracting the three-dimensional spatial coordinates of each defect point, calculating the vertical distance and vertical coordinates of each defect point to the adaptive reference plane, and using these as the calibrated spatial positions. Defect attribute information is extracted and associated with each calibrated spatial position to generate a calibration defect feature dataset. This dataset is then imported into the three-dimensional model of the liquid cargo tank. Based on the calibrated spatial positions, the defect points and attribute information are precisely located to the corresponding coordinates in the model, forming a defect digital twin feature layer that accurately corresponds to the spatial position of the physical structure. This overcomes the technical problems of traditional defect spatial position calibration (low accuracy), weak correlation between defect attributes and positions, inability to accurately map defects to the three-dimensional model of the liquid cargo tank, difficulty in intuitively presenting the spatial distribution of defects, and the inability to accurately correspond the defect distribution to the physical structure, as well as the lack of a visual control platform. This achieves accurate calibration of the defect spatial position, ensures precise binding of defect attributes and calibrated positions, and constructs a defect spatial distribution digital twin feature layer that highly matches the physical structure of the liquid cargo tank. This enables the visualization and structured presentation of defect distribution, intuitively and accurately reflecting the defect distribution in complex structural parts of the liquid cargo tank.

[0117] In a preferred embodiment of the present invention, step 4.3 above may include:

[0118] Step 4.31: Analyze the underlying data structure of the 3D model of the liquid cargo tank to obtain the geometric topology information and spatial index of all bulkheads, longitudinal and transverse ribs, and welded components contained in the 3D model of the liquid cargo tank; traverse each calibrated spatial position in the calibration defect feature dataset, and use a spatial coordinate matching algorithm to find and locate the target welded component and its specific coordinate points in the 3D model of the liquid cargo tank corresponding to the calibrated spatial position. Specifically, this includes: performing underlying data analysis of the 3D model of the liquid cargo tank. The 3D model of the liquid cargo tank used in this study is a standardized 3D solid format. The underlying data structure of the model is decomposed layer by layer using a model analysis tool to extract the geometric topology information of all bulkhead components, longitudinal and transverse rib components, and welded components in the model. The geometric topology information includes... The system includes the external dimensions, cross-sectional shape, connection relationships between adjacent components, and component boundary coordinates of each component. Simultaneously, it extracts the spatial index of each component in the unified coordinate system of the model. The spatial index represents the coordinate range of the component in the horizontal, vertical, and longitudinal directions, used to quickly locate the spatial coverage area of ​​the component. The weld component corresponding to the intersection of the longitudinal and transverse ribs is numbered HJ-001, with a spatial index of -400 mm to 200 mm in the horizontal direction, -400 mm to 0 mm in the longitudinal direction, and 0 mm to 10 mm in the vertical direction. The fillet weld of the bulkhead boundary is numbered HJ-002, with a spatial index of -500 mm to 0 mm in the horizontal direction, 0 mm to 500 mm in the longitudinal direction, and 0 mm to 15 mm in the vertical direction. All component spatial indices are consistent with the actual structural dimensions of the liquid cargo tank.

[0119] The spatial positions of the calibrated six defect points in the calibration defect feature dataset are read sequentially to determine the weld component to which the defect coordinates belong. The spatial coordinate matching judgment logic is as follows: the horizontal, vertical, and longitudinal coordinates of the defect after calibration are compared with the minimum horizontal, maximum horizontal, minimum vertical, maximum vertical, minimum vertical, and maximum vertical values ​​of the component spatial index, respectively. When the horizontal coordinate of the defect is greater than or equal to the minimum horizontal coordinate of the component and less than or equal to the maximum horizontal coordinate of the component, the longitudinal coordinate of the defect is greater than or equal to the minimum longitudinal coordinate of the component and less than or equal to the maximum longitudinal coordinate of the component, and the vertical coordinate of the defect is greater than or equal to the minimum vertical coordinate of the component and less than or equal to the maximum vertical coordinate of the component, the defect point is determined to belong to the corresponding component.

[0120] Taking the first defect point as an example, its calibrated spatial position is 199.3 mm horizontally, -200.7 mm vertically, and 3.7 mm vertically. Comparing this coordinate with the spatial index of weld HJ-001, the horizontal 199.3 mm falls between -400 mm and 200 mm, the vertical -200.7 mm falls between -400 mm and 0 mm, and the vertical 3.7 mm falls between 0 mm and 10 mm. All three conditions are met simultaneously. Therefore, the target weld component corresponding to this defect point is determined to be HJ-001, and the defect point is precisely located at the specific coordinates of 199.3 mm horizontally, -200.7 mm vertically, and 3.7 mm vertically on weld HJ-001. The same matching logic is applied to the remaining 5 defect points one by one. The calibrated spatial positions of all defect points meet the spatial index range of weld HJ-001, and all are located at the corresponding coordinate points on the weld HJ-001 at the intersection of the longitudinal and transverse ribs, completing the precise matching of all defect points with the target weld component.

[0121] Step 4.32: Using the found target weld component as the defect attachment carrier, the corresponding defect attribute information is written into the extended data storage area of ​​the target weld component in the form of attribute fields. Defect identification nodes are generated at the coordinate points on the target weld component corresponding to the calibrated spatial position. Specifically, this includes: using the found weld HJ-001 as the unified defect attachment carrier for all 6 defect points, using the defect number as a unique identifier, and writing the defect attribute information corresponding to each defect point into the extended data storage area of ​​the target weld component one by one in the form of structured attribute fields. The extended data storage area is a custom data storage space reserved for the 3D model component, which can permanently store defect-related information for subsequent traceability and querying. The attribute fields written for each defect point include defect number, defect type, defect size, original spatial coordinates, and calibrated spatial coordinates. The specific content written is as follows:

[0122] The first defect is numbered QX-001, the defect type is hot crack, the defect length is 1.18 mm, the original coordinates are 1100 mm horizontally, 700 mm vertically, and 5 mm vertically, and the calibrated coordinates are 199.3 mm horizontally, -200.7 mm vertically, and 3.7 mm vertically.

[0123] The second defect number is QX-002, the defect type is hot crack, the defect length is 1.298 mm, the original coordinates are 1050 mm horizontally, 650 mm vertically, and 6 mm vertically, and the calibrated coordinates are 164.3 mm horizontally, -235.7 mm vertically, and 4.7 mm vertically.

[0124] The third defect, numbered QX-003, is of the non-fusion type, with a length of 1.062 mm. Its original coordinates are 1000 mm horizontally, 600 mm vertically, and 4 mm vertically. Its calibrated coordinates are 129.3 mm horizontally, -270.7 mm vertically, and 3.7 mm vertically.

[0125] The fourth defect, numbered QX-004, is a hot crack with a length of 1.475 mm. Its original coordinates are 950 mm horizontally, 550 mm vertically, and 7 mm vertically. Its calibrated coordinates are 94.3 mm horizontally, -305.7 mm vertically, and 6.7 mm vertically.

[0126] The fifth defect, numbered QX-005, is of the non-fusion type, with a length of 1.003 mm. Its original coordinates are 900 mm horizontally, 500 mm vertically, and 5 mm vertically. Its calibrated coordinates are 59.3 mm horizontally, -340.7 mm vertically, and 4.7 mm vertically.

[0127] The sixth defect, numbered QX-006, is of the non-fusion type, with a length of 1.121 mm. Its original coordinates are 850 mm horizontally, 450 mm vertically, and 6 mm vertically. Its calibrated coordinates are 24.3 mm horizontally, -375.7 mm vertically, and 5.7 mm vertically.

[0128] After the attribute information is written, a visual defect identification node is generated at the coordinate point on the target weld HJ-001 that is completely consistent with the spatial position of each defect after calibration. The defect identification node is a spherical mark with a uniform diameter of 2 mm. To intuitively distinguish the defect type, hot crack defect identification nodes are displayed in red, and non-fusion defect identification nodes are displayed in yellow. Each identification node is bound to the corresponding defect point, realizing an intuitive and visual display of the defect location and type.

[0129] Step 4.33: Based on the spatial distribution of each defect identifier node in the 3D model of the liquid cargo tank, the defect identifier nodes belonging to the same weld are clustered and associated to form a defect cluster with spatial semantics. The defect cluster is then superimposed on the corresponding structural region of the 3D model of the liquid cargo tank as an independent layer to construct a digital twin feature layer of defect spatial distribution. Specifically, based on the spatial distribution of the 6 defect identifier nodes in the 3D model of the liquid cargo tank, the spatial distance clustering algorithm is used to cluster and associate the defect identifier nodes belonging to the same weld HJ-001. The spatial distance clustering uses the formula for calculating the distance between two points in 3D space. The distance value is equal to the square of the difference between the horizontal coordinates of the two defect points, plus the square of the difference between the vertical coordinates, plus the square of the difference between the vertical coordinates, and then the square root of the sum is performed. In this embodiment, the clustering distance threshold is set to 100 mm. When the spatial distance between any two defect marker nodes is less than or equal to 100 mm, the two defect nodes are grouped into the same defect cluster. The spatial distance between any two defect marker nodes is calculated sequentially. Taking the first and second defects as an example, the horizontal coordinate difference is 199.3 mm minus 164.3 mm, which equals 35 mm; the vertical coordinate difference is -200.7 mm minus -235.7 mm, which equals 35 mm; and the vertical coordinate difference is 3.7 mm minus 4.7 mm, which equals... The distance of -1 mm is equal to the square root of the sum of the square of 35 plus the square of 35 plus the square of -1, which is approximately 49.5 mm. This is less than the clustering threshold of 100 mm, so the nodes are grouped into the same cluster. The distance between all defect nodes is calculated using the same method. Since the distance between all defect nodes is less than 100 mm, the six defect identification nodes are clustered into a single defect cluster named QX-JQ-001. This cluster has complete spatial semantics and can intuitively reflect the concentrated distribution area and overall distribution pattern of defects on weld HJ-001.

[0130] The defect cluster QX-JQ-001 is superimposed onto the 3D model of the liquid cargo tank as an independent dedicated layer. The independent layer is named the defect distribution layer DEF-LAYER-001. The superposition position of the layer completely coincides with the structural area where the weld HJ-001 at the intersection of the longitudinal and transverse ribs is located. The superposition ratio maintains a 1:1 true mapping ratio with the 3D model. After superposition, a digital twin feature layer of defect spatial distribution is formed in the 3D model of the liquid cargo tank, which accurately corresponds to the spatial position of the actual liquid cargo tank structure. This feature layer fully contains all information such as defect calibration position, defect type, defect size, defect cluster distribution, and target weld binding relationship.

[0131] In this embodiment of the invention, by employing the technical means of analyzing the underlying data structure of the liquid cargo tank 3D model to obtain geometric topology information and spatial index, locating the target weld component and coordinate points corresponding to the defect through a spatial coordinate matching algorithm, writing the defect attribute information into the extended data area of ​​the weld component and generating defect identification nodes, clustering and associating defect identification nodes within the same weld to form a defect cluster, and constructing a defect spatial distribution digital twin feature layer by overlaying independent layers, the technical problems of traditional defect mapping being unable to accurately bind to 3D model components, unclear defect spatial carriers, lack of spatial semantic clustering, difficulty in hierarchical visualization, and insufficient accuracy and structure of digital twin mapping are overcome. This achieves accurate carrier association between defects and weld components, forms defect clusters with spatial semantics, and constructs a defect digital twin feature layer with clear layers, accurate positioning, and complete semantics, thereby improving the intuitiveness, structure, and traceability of defect spatial distribution display.

[0132] In a preferred embodiment of the present invention, step 5 above may include:

[0133] Step 5.1: Extract the defect spatial distribution digital twin feature layer corresponding to each batch from the welding quality database. The defect spatial distribution digital twin feature layer contains the calibrated spatial location, defect type, defect size, and associated welding process parameters for each defect point. Specifically, three consecutive batches of welding operations on the 316L stainless steel cargo tank of an ocean-going chemical tanker are selected as the analysis objects. The welding times of the three batches are the first batch, the second batch, and the third batch, respectively. Welding operations are carried out on the HJ-001 weld at the intersection of longitudinal and transverse skeletons. The defect spatial distribution digital twin feature layers corresponding to the three batches are retrieved from the welding quality database. The feature layer data of each batch contains the calibrated spatial location, defect type, defect size, and welding process parameters bound to the defect point. The welding process parameters specifically include welding current, arc voltage, welding speed, argon flow rate, and interpass temperature. The first batch extracted 6 defect point data, including hot cracks and lack of fusion; the second batch extracted 5 defect point data, including hot cracks and lack of fusion; and the third batch extracted 7 defect point data, including hot cracks and lack of fusion. All defect point data were precisely correlated with the coordinates of the 3D model of weld HJ-001, and each defect point corresponded to a unique set of welding process parameters, thus completing the extraction of multiple batches of original feature layer data.

[0134] Step 5.2 involves fusing the extracted digital twin feature layers of defect spatial distribution from multiple batches to construct a multi-batch defect dataset containing time, space, and attribute dimensions. Specifically, this includes standardizing the format and aligning the fields of the defect data from the three batches, removing duplicate and missing invalid data, using the welding batch number as a unique identifier, and integrating all defect data into the same data structure to construct a multi-batch defect dataset containing three dimensions. The time dimension data includes the welding batch number and welding operation date; the space dimension data includes the weld number and the three-dimensional spatial coordinates of the defect point after calibration; and the attribute dimension data includes the defect type, defect size, welding current, arc voltage, welding speed, argon flow rate, and interpass temperature. The fused multi-batch defect dataset contains a total of 18 defect points, with 6 from the first batch, 5 from the second batch, and 7 from the third batch. All data fields are unified and formatted correctly, achieving the fusion and unification of multi-batch defect data.

[0135] Step 5.3 involves statistical analysis of multiple batches of defect datasets, calculating the overall defect density, the percentage distribution of various defect types, and the statistical characteristics of defect sizes for each batch. It also includes correlation regression analysis between the spatial distribution of defects and welding process parameters to identify the correlation between abnormal fluctuations in welding parameters and defect occurrence. Specifically, this includes: performing item-by-item statistical calculations on multiple batches of defect datasets, with statistical indicators including overall defect density, percentage distribution of various defect types, and statistical characteristics of defect sizes. All calculations use standardized statistical formulas: overall defect density equals the total number of defects in a single batch divided by the total weld length in a single batch (the total weld length is uniformly 12 meters); defect percentage equals the number of defects of a single type divided by the total number of defects in that batch, with the result expressed as a percentage; and the average defect size equals the sum of all defect size values ​​divided by the total number of defects.

[0136] The specific statistical calculation results are as follows: The overall defect density of the first batch is 6÷12=0.5 defects per meter, the number of hot cracks is 3, accounting for 50%, the number of non-fusion is 3, accounting for 50%, and the average defect size is 1.2 mm;

[0137] The overall defect density of the second batch was 5÷12≈0.42 defects per meter, with 2 hot cracks (40%), 3 non-fusion defects (60%), and an average defect size of 1.1 mm.

[0138] The overall defect density of the third batch was 7÷12≈0.58 defects per meter, with 5 hot cracks (approximately 71.4%), 2 non-fused defects (approximately 28.6%), and an average defect size of 1.4 mm.

[0139] Linear regression analysis was used to establish the correlation between welding process parameters and defect occurrence. The regression formula is: the dependent variable, defect incidence rate, equals the independent variable, welding parameter, multiplied by the regression coefficient, plus a constant term. The correlation coefficients between welding parameters and defect incidence were calculated, with a stronger correlation indicating a closer absolute value to 1. Regression calculations were performed on welding current, arc voltage, welding speed, argon flow rate, interpass temperature, and defect incidence, revealing the following correlation patterns: welding current showed a strong positive correlation with hot crack incidence (correlation coefficient 0.89), with a 73% increase in the probability of hot cracking when the welding current exceeded 180 amperes; interpass temperature also showed a strong positive correlation with hot crack incidence (correlation coefficient 0.92), with an 81% increase in the probability of hot cracking when the interpass temperature exceeded 150 degrees Celsius; argon flow rate showed a negative correlation with non-fusion defect incidence (correlation coefficient -0.76), with a 58% increase in the probability of non-fusion defects when the argon flow rate was less than 10 liters per minute; arc voltage and welding speed showed no significant correlation with defect incidence (correlation coefficients were all below 0.3). These regression calculations clearly identified abnormally high welding current, excessive interpass temperature, and low argon flow rate as key influencing factors leading to defect formation.

[0140] Step 5.4: Based on the correlation patterns and combined with the preset intergranular corrosion tendency threshold and hot cracking sensitivity threshold, predict the trend of welding quality for the current batch and subsequent batches to determine whether there is a batch quality risk. Specifically, this includes: based on the correlation patterns between defects and welding parameters, and combined with preset quality judgment thresholds, predict the trend of welding quality for the current batch and subsequent batches to determine whether there is a batch quality risk. According to the 316L stainless steel welding quality standards and the service requirements of the liquid cargo tank structure, two types of key defect sensitivity thresholds are preset: the intergranular corrosion tendency threshold is an interpass temperature greater than 150 degrees Celsius; the hot cracking sensitivity threshold is a welding current greater than 180 amperes or an argon flow rate less than 10 liters per minute. Substitute the welding process parameters of the third batch into the correlation patterns and judgment thresholds. Threshold verification analysis: The average welding current of the third batch was 192 amperes, exceeding the hot cracking sensitivity threshold of 180 amperes; the average interpass temperature was 165 degrees Celsius, exceeding the intergranular corrosion tendency threshold of 150 degrees Celsius; the average argon flow rate was 9 liters per minute, below the hot cracking sensitivity threshold of 10 liters per minute. Based on the correlation patterns obtained from regression analysis, it is predicted that the hot cracking incidence rate of the current third batch of welding operations will reach over 70%, and there is also a risk of intergranular corrosion. Trend predictions are conducted for the subsequent fourth and fifth batches. If the welding parameters remain in the current abnormal state, the overall defect density is predicted to exceed 0.6 defects per meter, and the probability of batch hot cracking will exceed 85%. It is determined that the subsequent batches have a high-level batch quality risk, and the welding process parameters need to be adjusted immediately.

[0141] Step 5.5 involves structuring the defect distribution patterns, parameter correlation analysis results, and quality risk prediction conclusions obtained from statistical analysis, automatically filling in the data according to a preset report template, and generating a welding quality assessment report. Specifically, this includes: structuring the statistical analysis results, correlation regression conclusions, and quality risk prediction content; automatically filling in the data according to a preset template; generating a standardized welding quality assessment report; and classifying and organizing all analysis results according to the report template's field requirements: the basic information section includes ship type, liquid cargo tank material, analyzed weld number, and analysis batch range; the statistical analysis section includes the overall defect density of each batch, the proportion of various defects, and statistical characteristics of defect size; the correlation analysis section includes the correlation patterns between welding parameters and defects, and explanations of key influencing parameters; and the risk prediction section includes the current batch's quality status, the risk level of subsequent batches, and the type of defect hazard.

[0142] The structured data is automatically populated into a preset welding quality assessment report template. The template includes seven modules: cover, table of contents, data overview, statistical analysis, correlation patterns, risk prediction, and rectification suggestions. The template automatically completes the formatting and numerical input. The generated assessment report clearly states: The welding parameters of the third batch exceed the standard, and there is a batch risk of hot cracking and intergranular corrosion. It is recommended to adjust the welding current to the range of 140 to 160 amperes, control the interpass temperature to within 150 degrees Celsius, and adjust the argon flow rate to the range of 12 to 18 liters per minute. Subsequent batches of welding operations can only be carried out after the rectification is completed.

[0143] In this embodiment of the invention, by extracting and fusing digital twin feature layer data of multiple batches of defects spatial distribution, a multi-batch defect dataset containing time, space, and attribute dimensions is constructed. Statistical feature calculations and welding process parameter correlation regression analysis are performed on the defect data. Welding quality trend prediction and batch quality risk assessment are carried out in combination with thresholds. The analysis results are then structured and automatically generated into welding quality assessment reports according to templates. This overcomes the technical problems in traditional welding quality control, such as the inability to integrate and utilize multiple batches of data, the lack of multi-dimensional statistics and parameter correlation analysis, the difficulty in achieving quality trend prediction and batch risk warning, and the inability to automatically generate standardized quality assessment reports. This enables multi-dimensional in-depth mining and pattern identification of welding quality data, early prediction of batch quality hazards, automation and standardization of quality assessment, and promotes the transformation of welding quality control from passive detection to proactive prediction, thereby improving the scientific and intelligent level of welding quality control of stainless steel structures in liquid cargo ships.

[0144] like Figure 2 As shown, embodiments of the present invention also provide a welding quality data processing system for stainless steel structures of liquid cargo ships, including:

[0145] The acquisition module is used to collect welding process parameters, non-destructive testing data, and stainless steel material property data. The collected data is preprocessed to obtain a multi-source dataset. The multi-source dataset is matched and associated with the three-dimensional spatial coordinates of the weld to construct a welding quality database.

[0146] The identification module is used to intelligently identify non-destructive testing data in the welding quality database based on a preset stainless steel welding defect feature library to obtain a defect feature dataset; based on the welding quality database, an initial virtual reference plane is constructed using the spatial coordinate data of the intersection of the longitudinal and transverse ribs of the liquid cargo tank and the corner points of the tank wall boundary.

[0147] The calculation module is used to calculate the spatial coordinate covariance matrix of all defect points in the defect feature dataset, solve for the eigenvectors of the spatial coordinate covariance matrix, and take the eigenvector corresponding to the largest eigenvalue as the main direction of defect distribution. Based on the spatial angle between the main direction and the normal vector of the initial virtual reference plane, the initial virtual reference plane is rotated by a rotation transformation around an arbitrary axis in space to obtain an adaptive reference plane whose normal vector is perpendicular to the main direction.

[0148] The calibration module is used to calculate the orthogonal projection point from each defect point in the defect feature dataset to the adaptive reference plane, and use the point set formed by all projection points as the calibrated defect spatial location; the calibrated defect spatial location and the corresponding defect attributes are mapped to the 3D model of the liquid cargo tank to obtain the digital twin feature layer of defect spatial distribution.

[0149] The analysis module is used to perform statistical analysis on the digital twin feature layers of defect spatial distribution in multiple batches, obtain analysis results, and automatically generate welding quality assessment reports based on the analysis results.

[0150] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0151] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0152] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0153] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for processing welding quality data of stainless steel structures for liquid cargo ships, characterized in that, The method includes: Welding process parameters, non-destructive testing data, and stainless steel material property data are collected. The collected data are preprocessed to obtain a multi-source dataset. The multi-source dataset is matched and correlated with the three-dimensional spatial coordinates of the weld to construct a welding quality database. Based on a pre-set stainless steel welding defect feature library, the non-destructive testing data in the welding quality database is intelligently identified to obtain a defect feature dataset; based on the welding quality database, an initial virtual reference plane is constructed using the spatial coordinate data of the intersection of the longitudinal and transverse ribs of the liquid cargo tank and the corner points of the tank wall boundary. By calculating the spatial coordinate covariance matrix of all defect points in the defect feature dataset, the eigenvectors of the spatial coordinate covariance matrix are solved, and the eigenvector corresponding to the largest eigenvalue is taken as the main direction of defect distribution. Based on the spatial angle between the main direction and the normal vector of the initial virtual reference plane, the initial virtual reference plane is rotated by a rotation transformation around an arbitrary axis in space to obtain an adaptive reference plane whose normal vector is perpendicular to the main direction. Calculate the orthogonal projection point from each defect point in the defect feature dataset to the adaptive reference plane, and use the point set formed by all projection points as the calibrated defect spatial location; map the calibrated defect spatial location and the corresponding defect attributes to the 3D model of the liquid cargo tank to obtain the digital twin feature layer of the defect spatial distribution. Statistical analysis was performed on the digital twin feature layers of the defect spatial distribution of multiple batches to obtain the analysis results, and a welding quality assessment report was automatically generated based on the analysis results.

2. The method for processing welding quality data of stainless steel structures for liquid cargo ships according to claim 1, characterized in that, Welding process parameters, non-destructive testing data, and stainless steel material property data are collected. The collected data are preprocessed to obtain a multi-source dataset. A welding quality database is constructed by matching and associating multi-source datasets with the three-dimensional spatial coordinates of weld seams, including: Welding current, arc voltage, welding speed, argon flow rate, and interpass temperature data were collected to form a raw set of welding process parameters; waveforms, images, and numerical data were collected from phased array ultrasonic, X-ray, and eddy current testing equipment to form a raw set of nondestructive testing data; and chemical composition, mechanical properties, and intergranular corrosion tendency data were extracted from stainless steel material quality certificates to form a raw set of material properties. By cleaning the raw sets of welding process parameters, non-destructive testing, and material properties, respectively, abnormal jump values ​​and noise data that deviate significantly from the threshold during the acquisition process are removed, and the timestamp format and measurement unit of each data source are unified, a standardized set of welding process parameters, non-destructive testing dataset, and material property dataset are obtained. The standardized welding process parameter set, non-destructive testing dataset, and material property dataset are fused together, and the weld number is used as the primary key to match and associate it with the three-dimensional spatial coordinates of the weld extracted from the digital design drawings, forming a welding quality database that includes welding parameters, test results, material properties, and spatial coordinates.

3. The method for processing welding quality data of stainless steel structures for liquid cargo ships according to claim 2, characterized in that, Based on a pre-set stainless steel welding defect feature library, the non-destructive testing data in the welding quality database is intelligently identified to obtain a defect feature dataset. Based on the welding quality database, an initial virtual reference plane is constructed using the spatial coordinate data of the intersection of the longitudinal and transverse ribs of the liquid cargo tank and the corner points of the tank wall boundaries, including: Phased array ultrasonic waveform data, radiographic image data, and eddy current numerical data associated with weld numbers are extracted from the welding quality database to form a non-destructive testing dataset to be identified. The non-destructive testing dataset to be identified is input into a pre-set stainless steel welding defect feature library for feature matching. The defect feature library contains waveform features, image grayscale features and numerical threshold ranges of hot cracks, intergranular corrosion tendency and non-fusion defect types. By traversing and comparing, the defect type, size and spatial coordinates contained in the non-destructive testing data are identified, and a defect feature dataset is generated. The coordinates of predefined longitudinal and transverse rib intersection nodes and bulkhead boundary corner points are extracted from the welding quality database and used as spatial reference feature points. An initial virtual reference plane is generated by fitting the three-dimensional coordinates of the spatial reference feature points.

4. The method for processing welding quality data of stainless steel structures for liquid cargo ships according to claim 3, characterized in that, By calculating the spatial coordinate covariance matrix of all defect points in the defect feature dataset, the eigenvectors of the spatial coordinate covariance matrix are solved, and the eigenvector corresponding to the largest eigenvalue is taken as the principal direction of the defect distribution. Based on the spatial angle between the principal direction and the normal vector of the initial virtual reference plane, the initial virtual reference plane is rotated about any axis in space to obtain an adaptive reference plane whose normal vector is perpendicular to the principal direction, including: Extract the three-dimensional spatial coordinates of all defect points from the defect feature dataset to form a defect point coordinate matrix, and calculate the covariance matrix of the defect point coordinate matrix to obtain the spatial coordinate covariance matrix. The spatial coordinate covariance matrix is ​​decomposed into eigenvalues ​​to obtain three eigenvalues ​​and their corresponding eigenvectors. The eigenvector corresponding to the eigenvalue with the largest value is selected as the main direction of defect distribution. The main direction of defect distribution represents the maximum discrete extension direction of the defect point in three-dimensional space. Extract the normal vector from the initial virtual reference plane, calculate the spatial angle between the principal direction of the defect distribution and the normal vector of the initial virtual reference plane, and determine the required rotation angle and rotation axis direction of the initial virtual reference plane based on the spatial angle. Using the rotation axis as a reference, the initial virtual reference plane is rotated around any axis in space according to the spatial angle, so that the normal vector of the rotated plane is perpendicular to the main direction of the defect distribution. The plane after rotation transformation is used as the adaptive reference plane.

5. The method for processing welding quality data of stainless steel structures for liquid cargo ships according to claim 4, characterized in that, Calculate the orthogonal projection point from each defect point in the defect feature dataset to the adaptive reference plane, and use the set of points formed by all projection points as the calibrated defect spatial location; The calibrated spatial locations of defects and their corresponding attributes are mapped onto the 3D model of the liquid cargo tank to obtain a digital twin feature layer of the defect spatial distribution, including: The three-dimensional spatial coordinates of each defect point are extracted from the defect feature dataset. The vertical distance from each defect point to the adaptive reference plane is calculated. The perpendicular coordinates of each defect point on the adaptive reference plane are solved based on the spatial equation of the adaptive reference plane. The perpendicular coordinates are used as the spatial position of the corresponding defect point after calibration. The calibration defect point set is constructed with the spatial positions of all defect points after calibration. By extracting the defect type, defect size and original spatial coordinates corresponding to each defect point from the defect feature dataset as defect attribute information, the defect attribute information is associated one by one with the calibrated spatial location in the calibration defect point set to generate a calibration defect feature dataset containing the calibrated location and attribute information. The calibration defect feature dataset is imported into the 3D model of the liquid cargo tank. Based on the spatial location after calibration, each defect point and attribute information is located to the corresponding coordinate position in the 3D model of the liquid cargo tank, forming a digital twin feature layer of defect spatial distribution in the 3D model of the liquid cargo tank that precisely corresponds to the spatial position of the structural entity.

6. The method for processing welding quality data of stainless steel structures for liquid cargo ships according to claim 5, characterized in that, The calibration defect feature dataset is imported into the 3D model of the liquid cargo tank. Based on the calibrated spatial location, each defect point and its attribute information is located to the corresponding coordinate position in the 3D model of the liquid cargo tank. A digital twin feature layer of defect spatial distribution, which precisely corresponds to the spatial position of the structural entity, is formed in the 3D model of the liquid cargo tank, including: The underlying data structure of the 3D model of the liquid cargo tank is analyzed to obtain the geometric topology information and spatial index of all the bulkheads, longitudinal and transverse ribs and welded components contained in the 3D model of the liquid cargo tank in the model coordinate system; each calibrated spatial position in the calibration defect feature dataset is traversed, and the target welded component and the specific coordinate point on the target welded component corresponding to the calibrated spatial position are found and located in the 3D model of the liquid cargo tank through the spatial coordinate matching algorithm. The target weld component is used as a defect attachment carrier. The corresponding defect attribute information is written into the extended data storage area of ​​the target weld component in the form of attribute fields. Defect identification nodes are generated at the coordinate points on the target weld component that correspond to the calibrated spatial position. Based on the spatial distribution of each defect identifier node in the 3D model of the liquid cargo tank, defect identifier nodes belonging to the same weld are clustered and associated to form a defect cluster with spatial semantics. The defect cluster is then superimposed on the corresponding structural region of the 3D model of the liquid cargo tank as an independent layer to construct a digital twin feature layer of defect spatial distribution.

7. The method for processing welding quality data of stainless steel structures for liquid cargo ships according to claim 6, characterized in that, Statistical analysis was performed on the digital twin feature layers of defect spatial distribution across multiple batches to obtain analysis results. Based on these results, a welding quality assessment report was automatically generated, including: The defect spatial distribution digital twin feature layer corresponding to each batch is extracted from the welding quality database. The defect spatial distribution digital twin feature layer contains the calibrated spatial location, defect type, defect size and associated welding process parameters of each defect point. By fusing data from multiple batches of extracted defect spatial distribution digital twin feature layers, a multi-batch defect dataset containing time, space, and attribute dimensions is constructed. Statistical analysis was performed on multiple batches of defect datasets to calculate the overall density of defects in each batch, the proportion distribution of various types of defects, and the statistical characteristics of defect size. Furthermore, a correlation regression analysis was conducted between the spatial distribution of defects and welding process parameters to identify the correlation between abnormal fluctuations in welding parameters and defect generation. Based on the correlation law, combined with the preset intergranular corrosion tendency threshold and hot crack sensitivity threshold, the welding quality of the current batch and subsequent batches is predicted to determine whether there is a batch quality risk. The defect distribution patterns obtained from statistical analysis, the results of parameter correlation analysis, and the conclusions of quality risk prediction are structured and organized, and the data is automatically filled in according to the preset report template to generate a welding quality assessment report.

8. A data processing system for welding quality of stainless steel structures for liquid cargo ships, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to collect welding process parameters, non-destructive testing data, and stainless steel material property data. It preprocesses the collected data to obtain a multi-source dataset. A welding quality database is constructed by matching and associating multi-source datasets with the three-dimensional spatial coordinates of welds. The identification module is used to intelligently identify non-destructive testing data in the welding quality database based on a preset stainless steel welding defect feature library, and obtain a defect feature dataset. Based on the welding quality database, an initial virtual reference plane is constructed using the spatial coordinate data of the intersection of the longitudinal and transverse ribs of the liquid cargo tank and the corner points of the tank wall boundary. The calculation module is used to calculate the spatial coordinate covariance matrix of all defect points in the defect feature dataset, solve for the eigenvectors of the spatial coordinate covariance matrix, and take the eigenvector corresponding to the largest eigenvalue as the main direction of defect distribution. Based on the spatial angle between the main direction and the normal vector of the initial virtual reference plane, the initial virtual reference plane is rotated by a rotation transformation around an arbitrary axis in space to obtain an adaptive reference plane whose normal vector is perpendicular to the main direction. The calibration module is used to calculate the orthogonal projection points from each defect point in the defect feature dataset to the adaptive reference plane, and the point set formed by all the projection points is used as the calibrated defect spatial location. The calibrated spatial location of the defect and its corresponding defect attributes are mapped onto the 3D model of the liquid cargo tank to obtain a digital twin feature layer of the defect spatial distribution; The analysis module is used to perform statistical analysis on the digital twin feature layers of defect spatial distribution in multiple batches, obtain analysis results, and automatically generate welding quality assessment reports based on the analysis results.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images