Welding defect data management method and system based on equivalent crack model

Abstract

The invention relates to the technical field of industrial automation and data management, and discloses a welding defect data management method and system based on an equivalent crack model. The method comprises the following steps: acquiring working condition data and process parameters of welding defects; performing feature extraction, dynamic weight adjustment and risk quantitative calculation according to the data to obtain a risk score; high-risk defects are screened according to the risk scores, a management file is generated, feedback optimization is carried out based on the management file, and technological parameters are adjusted. According to the method, an intelligent management process from dynamic assessment to closed-loop feedback is constructed, so that the problem of low risk assessment precision in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the fields of industrial automation and data management technology, and in particular to a welding defect data management method and system based on an equivalent crack model. Background Technology

[0002] Currently, high-quality welding is a core element in ensuring the safety and service life of structural components in high-end manufacturing industries such as aerospace and energy equipment, while the scientific management and accurate assessment of welding defects are key to ensuring welding quality.

[0003] In existing technologies, defect data management largely relies on traditional manual inspection or simple automated inspection systems. These automated systems often employ fixed rules or early machine learning models, such as a decision tree based on a fixed threshold, to classify the collected geometric features of defects. However, in actual working conditions, the impact of welding defects on structural safety depends not only on their static geometric dimensions but also on complex factors such as dynamically changing stress field distribution, material properties, and service environment. Because the aforementioned existing technologies use fixed management strategies and model parameters, they are ill-suited to this multi-dimensional, dynamically changing assessment scenario. They cannot accurately characterize the complex mapping relationship between the geometric features of defects and their mechanical effects, leading to biased risk assessment results and even missed or misjudged truly dangerous defects.

[0004] Therefore, existing technologies suffer from low accuracy in assessing the risk of welding defects. Summary of the Invention

[0005] This invention provides a welding defect data management method and system based on an equivalent crack model to solve the problem of low accuracy in risk assessment of welding defect data in the existing technology.

[0006] In a first aspect, to address the aforementioned technical problems, this invention provides a welding defect data management method based on an equivalent crack model, comprising: Collect welding defect condition data and process parameters, and preprocess them to obtain an initial defect dataset; Based on the initial defect dataset, feature extraction and weight adjustment are performed to obtain a weighted feature vector; Based on the weighted feature vectors, clustering is performed to obtain preliminary risk categories; Based on the initial defect dataset, risk quantification calculation is performed to obtain a risk score; Based on the risk score and the preliminary risk category, a report is selected and generated to obtain the final management file; Based on the final management file and the initial defect dataset, data fusion and updates are performed to obtain an enhanced defect dataset; Based on the enhanced defect dataset and the final management file, feedback parameters are adjusted to obtain an optimized defect prevention strategy.

[0007] Preferably, the process of collecting welding defect data and process parameters, and preprocessing them to obtain an initial defect dataset, includes: Collect the geometric dimensions, material hardness values, stress peak points, defect location coordinates, ambient temperature and service environment data as the working condition data, and combine them with the process parameters to form raw data; The original data is cleaned and normalized to obtain the initial defect dataset.

[0008] Preferably, the step of performing feature extraction and weight adjustment based on the initial defect dataset to obtain a weighted feature vector includes: Extract the original physical features from the initial defect dataset and concatenate them to form an initial feature vector; If the stress peak point component in the initial feature vector exceeds a preset stress threshold, then a set of preset risk weight coefficients are multiplied with the predetermined risk-related components in the initial feature vector to obtain the weighted feature vector.

[0009] Preferably, the step of performing clustering processing based on the weighted feature vector to obtain preliminary risk categories includes: The weighted feature vectors are grouped to determine multiple cluster centers; Calculate the similarity between each weighted feature vector and each cluster center to obtain a similarity value; Based on the similarity value, the weighted feature vector is assigned to the category corresponding to the cluster center with the highest similarity, thus obtaining the preliminary risk category.

[0010] Preferably, the step of performing risk quantification calculation based on the initial defect dataset to obtain a risk score includes: The geometric dimensions are obtained from the initial defect dataset, and the equivalent parameters are calculated based on the geometric dimensions. The equivalent parameters are standardized to obtain a set of standardized quantized parameters; An initial risk score is obtained by performing a linear mapping calculation based on the standardized set of quantitative parameters and the material hardness values ​​in the initial defect dataset. The initial risk score is compared with a preset score range, and if the initial risk score exceeds the preset score range, it is adjusted to obtain an adjusted risk score. Based on the initial risk category, a preset category weighting factor is assigned to the adjusted risk score, and a weighted calculation is performed to obtain the risk score.

[0011] Preferably, the step of screening and generating a report based on the risk score and the preliminary risk category to obtain the final management file includes: The risk score is compared with a preset risk threshold, and defect data with risk scores higher than the preset risk threshold are filtered out to form a high-risk defect subset; Based on the high-risk defect subset and using a preset report template, an automated report set is generated; The automated report collection is stored to obtain the final management file.

[0012] Preferably, the step of performing data fusion and updating based on the final management file and the initial defect dataset to obtain the enhanced defect dataset includes: The process parameters are obtained from the final management file, and defect data containing the geometric dimensions are filtered from the initial defect dataset; The process parameters and the geometric dimensions are fused together to obtain a multidimensional data set; The multidimensional dataset is prioritized based on the risk scores associated with the geometric dimensions, and a preferred dataset is obtained by filtering. The initial defect dataset is updated based on the preferred dataset to obtain the enhanced defect dataset.

[0013] Preferably, the step of adjusting feedback parameters based on the enhanced defect dataset and the final management file yields an optimized defect prevention strategy: Defect data containing high stress peak points are filtered from the enhanced defect dataset and combined with the preliminary risk category in the final management file to obtain a preliminary defect set; The stress peak points and the preliminary risk categories in the preliminary defect set are integrated to obtain a multidimensional fusion dataset; Based on the severity of the preliminary risk categories, the multidimensional fusion datasets are prioritized to obtain the preferred fusion datasets; The optimized defect prevention strategy is obtained by performing parameter optimization processing based on the preferred fusion dataset and adjusting the welding process feedback parameters.

[0014] Secondly, the present invention provides a welding defect data management system based on an equivalent crack model, used to implement the above-mentioned welding defect data management method based on an equivalent crack model, comprising: The data acquisition and preprocessing module is used to collect working condition data and process parameters of welding defects, and perform preprocessing to obtain an initial defect dataset; The feature extraction and weighting module is used to extract features and adjust weights based on the initial defect dataset to obtain a weighted feature vector; The risk classification module is used to perform clustering based on the weighted feature vector to obtain preliminary risk categories; The risk quantification module is used to perform risk quantification calculations based on the initial defect dataset to obtain a risk score; The decision-making and archiving module is used to filter and generate reports based on the risk score and the preliminary risk category to obtain the final management file; The data enhancement module is used to perform data fusion and update based on the final management file and the initial defect dataset to obtain an enhanced defect dataset; The strategy generation module is used to adjust feedback parameters based on the enhanced defect dataset and the final management file to obtain an optimized defect prevention strategy.

[0015] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the welding defect data management method based on the equivalent crack model described in any one of the above.

[0016] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the welding defect data management method based on the equivalent crack model described above.

[0017] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention collects multi-dimensional working condition data, including geometric dimensions, material hardness, stress peak points and service environment, performs feature extraction and dynamic weight adjustment, and combines equivalent crack model to perform risk quantification calculation; this method improves the assessment of defects from a single static geometric dimension to a multi-dimensional dynamic assessment that comprehensively considers mechanical properties and environmental factors, and more profoundly reflects the true hazard of defects, thereby solving the problem of low risk assessment accuracy caused by the simple assessment model and inability to handle complex mapping relationships in the existing technology.

[0018] (2) By establishing a dual feedback closed-loop mechanism, the present invention first traces back the corresponding process parameters based on the archived high-risk defects to enhance and iterate the knowledge dataset. Then, based on the enhanced dataset, it performs in-depth analysis and actively generates optimization and prevention strategies that can be used to guide production. This mechanism transforms the system from a passive defect detector into an optimizer that can learn and evolve on its own, solving the problem of rigid data management strategies caused by the use of fixed parameters and static rules in the prior art.

[0019] (3) By constructing a complete and automated data management process from multi-dimensional data acquisition, dynamic risk quantification, decision archiving to dual closed-loop feedback optimization, this invention forms an intelligent system that can self-iterate and continuously optimize based on real-time analysis results. Compared with the isolated and static management of existing technologies, this invention has made progress in the overall operating efficiency of the system, the scientific nature of decision-making, and the reliability of long-term application, providing a brand-new technical paradigm for welding quality control in complex industrial scenarios. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the welding defect data management method based on the equivalent crack model provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the welding defect data management system based on the equivalent crack model provided in the second embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figure 1 The first embodiment of the present invention provides a welding defect data management method based on an equivalent crack model, comprising the following steps: S11: Collect the working condition data and process parameters of welding defects, and perform preprocessing to obtain the initial defect dataset; S12, Based on the initial defect dataset, perform feature extraction and weight adjustment to obtain a weighted feature vector; S13, perform clustering based on the weighted feature vector to obtain preliminary risk categories; S14. Based on the initial defect dataset, perform risk quantification calculation to obtain a risk score; S15, Based on the risk score and the preliminary risk category, filter and generate a report to obtain the final management file; S16, perform data fusion and update based on the final management file and the initial defect dataset to obtain an enhanced defect dataset; S17. Based on the enhanced defect dataset and the final management file, the feedback parameters are adjusted to obtain an optimized defect prevention strategy.

[0023] In step S11, the working condition data and process parameters of welding defects are collected and preprocessed to obtain an initial defect dataset, including: Collect the geometric dimensions, material hardness values, stress peak points, defect location coordinates, ambient temperature and service environment data as the working condition data, and combine them with the process parameters to form raw data; The original data is cleaned and normalized to obtain the initial defect dataset.

[0024] In one implementation, this embodiment utilizes a multi-source heterogeneous sensor network deployed on the welding production line or inspection station to collect the operating condition data and process parameters in real time. It should be noted that the operating condition data is a set of parameters describing the physical state of the defect and its environment. Specifically, it may include: the geometric dimensions obtained by a laser profilometer or industrial camera; the material hardness value obtained by a portable hardness tester; the stress peak point obtained by strain gauges attached to the surface of the structural component; the defect location coordinates obtained by ultrasonic testing equipment; and the ambient temperature and service environment data (such as humidity and vibration frequency) obtained by temperature and humidity sensors. The process parameters are production process records directly related to the occurrence of the defect, such as welding current, voltage, and welding speed read from the welding equipment controller. All these collected multi-dimensional, heterogeneous data collectively constitute the raw data.

[0025] It is worth noting that, in order to eliminate potential noise in the original data and unify the dimensions of different data, this embodiment performs data cleaning and normalization on the original data. The data cleaning specifically includes using the 3-sigma principle in statistics to calculate the mean (μ) and standard deviation (σ) of each numerical data sequence (such as material hardness values), and identifying any data points falling outside the (μ-3σ, μ+3σ) interval as outliers and removing or correcting them. The normalization process uses the Min-Max Normalization method to linearly transform all numerical data into a closed interval of [0,1]. For example, assuming a set of material hardness values ​​is collected in the range of 200-300 HV, after Min-Max Normalization, this range will be linearly mapped to the [0,1] interval. After the above cleaning and normalization processes, all data are transformed into dimensionless, uniformly formatted standardized values, ultimately forming a structured initial defect dataset that facilitates subsequent calculations and analysis.

[0026] In step S12, feature extraction and weight adjustment are performed based on the initial defect dataset to obtain a weighted feature vector, including: Extract the original physical features from the initial defect dataset and concatenate them to form an initial feature vector; If the stress peak point component in the initial feature vector exceeds a preset stress threshold, then a set of preset risk weight coefficients are multiplied with the predetermined risk-related components in the initial feature vector to obtain the weighted feature vector.

[0027] In one implementation, this embodiment first extracts all relevant original physical features from the initial defect dataset (which includes data collected from S11 and normalized), and concatenates them into a fixed-dimensional initial feature vector according to a preset order. An exemplary initial feature vector structure can be defined as: V_initial = (defect length, defect depth, material hardness value, stress peak point, ambient temperature, humidity). Here, defect length and defect depth are key components directly extracted from the geometric dimensions of the initial defect dataset, while the rest are corresponding values ​​directly extracted from the initial defect dataset.

[0028] In another implementation, this embodiment determines whether the stress peak point in the initial defect dataset exceeds a preset stress threshold, and adjusts the weights of the initial feature vector based on the determination result to obtain the weighted feature vector. It should be noted that the preset stress threshold is determined based on statistical analysis of fatigue test data from historical weldments. Specifically, this determination process involves collecting stress-life (SN) curve data of a large number of weldment samples under cyclic loading to identify the fatigue limit of the material; multiplying this fatigue limit by a safety factor (e.g., 0.75) determined according to industry design specifications (such as relevant standards from the American Society of Mechanical Engineers (ASME) or the American Welding Society (AWS)) and setting this as the preset stress threshold.

[0029] It is worth noting that the weight adjustment process is a component-conditional weighting process. Specifically, if the stress peak point does not exceed the preset stress threshold, the initial feature vector is directly used as the weighted feature vector; if the stress peak point exceeds the preset stress threshold, a set of preset risk weight coefficients are multiplied by the corresponding risk-related components in the initial feature vector (e.g., components corresponding to defect length, defect depth, stress peak point, and humidity) to amplify the values ​​of these components and obtain the final weighted feature vector.

[0030] It should be noted that the risk weight coefficients were determined in an offline modeling phase through multivariate logistic regression analysis of historical failure cases. This analysis used whether the welded component failed as the dependent variable and a set of original characteristics with clear physical meaning (e.g., defect length, defect depth, stress peak point, humidity) as independent variables to train the model and calculate the regression coefficients of each factor's contribution to the failure.

[0031] It should be noted that the specific calculation of the regression coefficients is determined by solving the log-likelihood function of the model using an iterative optimization algorithm (such as the L-BFGS algorithm) to find the maximum value; then, the absolute values ​​of each coefficient are normalized to obtain the final risk weight coefficients; wherein the loss function of the multivariate logistic regression is the negative log-likelihood. ,in , It is the first The true label of each sample The model predicts the first The probability that a sample belongs to the positive class has a range of [0, 1]. It is the weight coefficient vector of logistic regression, used to adjust the linear combination relationship of input features. It is the first The feature vector of each sample It is the negative log-likelihood loss function, which measures the difference between the model's predicted values ​​and the actual values; the smaller the value, the better. It is optimized iteratively using the L-BFGS algorithm, with a convergence tolerance of 1×10⁻⁶. -6 The maximum number of iterations is 1000 to ensure the stability of the coefficients.

[0032] In step S13, clustering is performed based on the weighted feature vector to obtain preliminary risk categories, including: The weighted feature vectors are grouped to determine multiple cluster centers; Calculate the similarity between each weighted feature vector and each cluster center to obtain a similarity value; Based on the similarity value, the weighted feature vector is assigned to the category corresponding to the cluster center with the highest similarity, thus obtaining the preliminary risk category.

[0033] In one implementation, this embodiment uses a k-means clustering algorithm to group the set of all weighted feature vectors generated in S12. This process is an unsupervised learning process that automatically groups vectors with similar positions in the feature space into one class, thereby identifying natural clusters in defective data. It should be noted that the core hyperparameter of the k-means clustering algorithm, namely the number of clusters k, is predetermined in an offline modeling phase by analyzing a large dataset of historical weighted feature vectors. Specifically, this determination process can employ the Elbow Method. In this embodiment, iterative calculations are performed for a preset range of k values ​​(e.g., from 2 to 10). For each k value, the k-means algorithm is run and its within-cluster sum of squared errors (SSE) is calculated. A curve of SSE changing with k values ​​is plotted, and the k value corresponding to the elbow where the rate of decline in the curve changes from steep to gentle is selected as the optimal number of clusters. This method ensures that the choice of the number of clusters maximizes the cluster discrimination while avoiding overfitting caused by an excessively large k value.

[0034] It is worth noting that k-means clustering uses 10 random initializations, selects the result with the smallest sum of squares (SSE) within the cluster as the final cluster center, sets the convergence tolerance threshold to 1e-4, and sets the maximum number of iterations to 300 to avoid local optima. After determining the number of clusters k, the specific implementation of the grouping process is an iterative optimization process, including: randomly initializing k cluster centers; entering an iterative loop, which includes an allocation step and an update step, wherein the allocation step assigns each weighted feature vector to the cluster center with the nearest distance, and the update step updates the position of each cluster center to the mean of all the vectors it contains; repeating this loop until the preset convergence condition is met, reaching a convergence state.

[0035] It should be noted that the specific method for determining the preset convergence condition is as follows: after each iteration's update step, calculate the Euclidean distance between the new position of each cluster center and its old position from the previous iteration; compare the maximum value among all k distances with a preset convergence tolerance threshold; if the maximum distance is less than the convergence tolerance threshold, the algorithm is determined to have converged, and the loop terminates. It should be noted that the convergence tolerance threshold is a very small positive number (e.g., 1e^(-1 / 2)). -4 The choice of its value ensures the stability of the clustering results while avoiding unnecessary iterative calculations. The k center points obtained after convergence are the final determined cluster centers.

[0036] It is worth noting that after obtaining multiple cluster centers through clustering, this embodiment also includes an offline analysis and labeling step. By analyzing the numerical distribution of each cluster center across various feature dimensions and combining it with its physical meaning (e.g., stress component values ​​are generally high), a clear physical label with engineering significance (e.g., "crack type", "porosity type", etc.) is assigned to each category represented by the cluster center.

[0037] In another implementation, this embodiment quantifies the similarity between the weighted feature vectors and the cluster centers by calculating the Euclidean distance between each weighted feature vector and each cluster center. It should be noted that Euclidean distance is a way to measure the straight-line distance between two points in multidimensional space; the smaller the distance value, the higher the similarity between the two vectors.

[0038] For example, assuming k=2 is determined using the aforementioned method, and the two cluster centers C1 and C2 are labeled as "crack class" and "porosity class" respectively through offline analysis and annotation. For a given weighted feature vector V, this embodiment calculates its similarity value with the two centers, resulting in d(V,C1)=0.5, d(V,C2)=1.2. Based on the similarity value, this embodiment assigns the weighted feature vector V to the category corresponding to the cluster center with the highest similarity (i.e., closest distance). In this example, since the similarity value of V and C1 is the minimum of 0.5, the initial risk category of V is determined as "crack class". After performing this assignment process on all weighted feature vectors, the resulting classification result set is the initial risk category.

[0039] In one implementation, to avoid the large differences in the numerical scales of the weighted feature components from adversely affecting subsequent clustering analysis, this embodiment performs a new min-max normalization process on the set of weighted feature vectors after obtaining all weighted feature vectors, so that all components are uniformly mapped to the [0,1] interval again.

[0040] In step S14, risk quantification calculation is performed based on the initial defect dataset to obtain a risk score, including: The geometric dimensions are obtained from the initial defect dataset, and the equivalent parameters are calculated based on the geometric dimensions. The equivalent parameters are standardized to obtain a set of standardized quantized parameters; An initial risk score is obtained by performing a linear mapping calculation based on the standardized set of quantitative parameters and the material hardness values ​​in the initial defect dataset. The initial risk score is compared with a preset score range, and if the initial risk score exceeds the preset score range, it is adjusted to obtain an adjusted risk score. Based on the initial risk category, a preset category weighting factor is assigned to the adjusted risk score, and a weighted calculation is performed to obtain the risk score.

[0041] In one implementation, this embodiment first obtains the geometric dimensions corresponding to each defect from the initial defect dataset. To unify the multidimensional geometric features into a single index that comprehensively reflects their severity, this embodiment calculates the equivalent parameter based on the geometric dimensions using a preset fracture mechanics model. It should be noted that the preset fracture mechanics model is a stress intensity factor calculation model selected from publicly available engineering standards or manuals (e.g., ASTM E399 standard) that matches the material type and geometry of the workpiece under test.

[0042] For example, to ensure that all defects of different morphologies (whether approximate planar cracks or volumetric defects) can be uniformly quantified for subsequent standardized comparisons, this embodiment preferably and uniformly adopts the equivalence quantification method based on defect projected area proposed by Murakami. Its equivalent crack size... (i.e., the equivalent parameter) can be obtained from the formula The calculation yielded, where Let be the projected area of ​​the defect on the plane perpendicular to the direction of the maximum principal stress. This method unifies the geometrical hazard of all three-dimensional defects into a single, comparable equivalent parameter.

[0043] In another implementation, this embodiment standardizes the equivalent parameters obtained from the aforementioned calculation. Specifically, the standardization process can employ a min-max normalization method, scaling the equivalent parameter values ​​to a uniform range of [0,1] through linear transformation to eliminate the influence of dimensions and obtain the standardized quantized parameter set. Subsequently, this embodiment uses the standardized quantized parameter set and the material hardness values ​​from the initial defect dataset as input, and performs linear mapping calculations using a pre-trained linear regression model to obtain the initial risk score.

[0044] It should be noted that the construction and training process of the linear regression model is as follows: First, a training dataset containing thousands of historical defect samples is obtained, in which each sample contains an equivalent parameter, a material hardness value, and a corresponding true risk value determined by a destructive physical test (such as a tensile fatigue test) as a label; then, the model is trained using the Ordinary Least Squares (OLS) method.

[0045] It should be noted that the linear regression model used in this embodiment can be expressed as follows: in, For the predicted risk value, For the standardized equivalent parameter in the set of standardized quantization parameters, The value is the hardness of the material. , , The coefficients to be solved in the model. This is the error term. The goal of OLS is to find a set of model coefficients that minimizes the sum of squared residuals (SSR) between the predicted and true risk values ​​for all training samples. This solution process can be achieved by solving the normal equation: The analytical calculation yielded the result, where For the true risk value vector, This is a feature matrix containing intercept terms, equivalent parameters, and material hardness values. Its first column is all 1s (corresponding to the intercept term). The second column contains the standardized equivalent parameters for all samples in the training set, and the third column contains the normalized material hardness values. This represents the optimal model coefficient vector. Training is complete when the mean squared error of the model on an independent test set is below a preset error threshold.

[0046] It should be noted that the error threshold is determined by plotting a curve showing the model's performance (such as prediction accuracy) on the test set as a function of model complexity (e.g., by introducing higher-order terms or interaction terms), and selecting the error value corresponding to the "inflection point" on the curve where performance no longer significantly improves, thus achieving a balance between the model's accuracy and generalization ability.

[0047] It is worth noting that, to ensure the rationality and stability of the risk score, this embodiment compares the calculated initial risk score with a preset score range. The preset score range is determined through statistical analysis of the initial risk score distributions of all "qualified" and "needs attention" level defects in historical data. For example, the 95th percentile of the "qualified" category score distribution is selected as the upper limit of the range, and the 5th percentile of the "needs attention" category score distribution is selected as the lower limit. If the initial risk score exceeds this range, it is adjusted, for example, by forcibly setting it to the boundary value of the range, resulting in the adjusted risk score.

[0048] Finally, based on the preliminary risk categories obtained in S13, this embodiment assigns a preset category weighting factor to the adjusted risk score and performs a weighted calculation to obtain the final risk score. It should be noted that the preset category weighting factor is a knowledge base that pre-configures a weight value for each preliminary risk category (such as "crack type" or "porosity type"). This weight value is determined based on the degree of harm different types of defects pose to structural safety, and is based on relevant industry safety standards and publicly available literature on material failure analysis.

[0049] For example, suppose a defect (whether a crack or a pore) has a projected area of ​​2.5 mm² measured by a laser profilometer or ultrasonic flaw detector on a plane perpendicular to the direction of maximum principal stress. According to the equivalence method used in this embodiment, its equivalent parameter is calculated as follows: The value of 1.581 mm is the equivalent parameter, used for subsequent standardization processing.

[0050] In step S15, based on the risk score and the preliminary risk category, a report is selected and generated to obtain the final management file, including: The risk score is compared with a preset risk threshold, and defect data with risk scores higher than the preset risk threshold are filtered out to form a high-risk defect subset; Based on the high-risk defect subset and using a preset report template, an automated report set is generated; The automated report collection is stored to obtain the final management file.

[0051] In one implementation, this embodiment compares each risk score calculated in S14 with a preset risk threshold. If the risk score is higher than the preset risk threshold, the defect data corresponding to that score (including all its operating data, process parameters, and analysis results) is filtered out, and all the filtered defect data constitute the high-risk defect subset. It should be noted that the determination of the preset risk threshold is a decision-making process that combines historical data statistics and risk cost analysis. Specifically, the determination process involves: first, collecting a large number of historical defect samples and their results on whether they ultimately lead to component failure; second, plotting a receiver operating characteristic (ROC) curve, with the true positive rate (the proportion of correctly identifying dangerous defects) at different thresholds on the vertical axis and the false positive rate (the proportion of misjudging safety defects as dangerous) on the horizontal axis; finally, combining the predefined costs incurred due to failure and the detection costs incurred due to misjudgment, selecting an operating point on the ROC curve that minimizes the total expected cost, and the risk score corresponding to this operating point is determined as the preset risk threshold.

[0052] In another implementation, this embodiment generates a structured, automated report for each of the high-risk defect subsets generated in the previous step, and all reports together constitute the automated report set. It should be noted that the report generation is based on a preset report template. This template is a standardized data structure (e.g., JSON or XML format) with predefined fields to fully record the entire lifecycle information of a high-risk defect. These fields include at least a unique defect identifier, a detection timestamp, complete operating condition data (including geometric dimensions, material hardness values, etc.), the process parameters corresponding to the discovery of the defect, the calculated risk score, and the preliminary risk category determined in S13.

[0053] It is worth noting that this embodiment persistently stores the generated automated report set to form the final management archive that can be traced and analyzed over a long period. This storage process is executed in a structured database (e.g., a time-series database or a relational database). Each report is stored as a record in the database and indexed based on key metadata such as its unique defect identifier, detection timestamp, and preliminary risk category to support efficient data retrieval and correlation analysis. The database also supports version control; if the status of a defect changes during subsequent detection or manual review, the system can update the corresponding report record, thereby establishing a dynamic and complete lifecycle archive for each high-risk defect. For example, suppose a defect's risk score is calculated to be 0.85, while the preset risk threshold is 0.7. Since 0.85 is higher than 0.7, this defect is filtered and included in the high-risk defect subset. The system then generates a report containing all information about the defect based on a preset JSON template. Finally, the JSON report was stored in the database and indexed as {"defect_id": "W-1024", "timestamp": "2025-10-12T12:30:05Z", "risk_category": "crack class"}, becoming part of the final management archive.

[0054] In step S16, data fusion and updating are performed based on the final management file and the initial defect dataset to obtain an enhanced defect dataset, including: The process parameters are obtained from the final management file, and defect data containing the geometric dimensions are filtered from the initial defect dataset; The process parameters and the geometric dimensions are fused together to obtain a multidimensional data set; The multidimensional dataset is prioritized based on the risk scores associated with the geometric dimensions, and a preferred dataset is obtained by filtering. The initial defect dataset is updated based on the preferred dataset to obtain the enhanced defect dataset.

[0055] In one implementation, to achieve feedback learning and data augmentation, this embodiment first extracts the process parameters corresponding to all high-risk defects from the final management file generated in S15. Simultaneously, this embodiment filters out defect data containing the geometric dimensions from the initial defect dataset based on the unique defect identifiers that correspond one-to-one with the high-risk defects in the final management file. It should be noted that the fusion process associates and matches the extracted process parameters with the geometric dimensions to construct a clear "process-defect" causal relationship dataset. Specifically, this process involves creating a new structured data record for each high-risk defect instance. This new record not only contains the process parameters extracted from the final management file but also all other operating condition data (such as geometric dimensions, material hardness values, stress peak points, etc.) corresponding to the defect retrieved from the initial defect dataset using the defect unique identifier, ensuring that the data structure of the new record is completely consistent with the initial defect dataset. All these newly generated, pattern-consistent, complete data records together constitute the multidimensional data set.

[0056] In another implementation, this embodiment prioritizes the generated multidimensional dataset based on the risk score associated with the geometric dimensions (e.g., using a quicksort algorithm). This sorting uses the risk score as the unique sorting key, arranging data in descending order to ensure that the data records with the highest risk and greatest learning value are at the top of the set. Subsequently, this embodiment filters the sorted set according to a preset data filtering rule to obtain the preferred dataset. It should be noted that the preset data filtering rule can be to select the top N% of the sorted data, where N is a hyperparameter determined through offline experiments based on the trade-off between the system's learning rate and data storage cost. For example, when N < 15%, the learning samples are insufficient; when N > 25%, noise is introduced; and 20% is the optimal balance point, which can be set to select the top 20% of the sorted data (N = 20).

[0057] It is worth noting that in this embodiment, the initial defect dataset is updated based on the preferred data set to obtain the enhanced defect dataset. This update process is a data appending operation, which involves adding all high-value "process-defect" data records from the preferred data set to the end of the original initial defect dataset. After this operation, the knowledge capacity of the dataset is expanded, containing verified high-risk causal samples; hence, it is called the enhanced defect dataset. This enhanced dataset will provide a high-quality data foundation for generating more accurate prevention strategies in S17.

[0058] For example, suppose the final management file records a high-risk crack with a risk score of 0.9, corresponding to process parameters of {current: 180A, speed: 0.5m / min}, and geometric dimensions of {length: 3.5mm, depth: 0.8mm}. This embodiment merges this information into a new data record. After sorting all similar records by risk score, this record is selected for inclusion in the preferred data set due to its high risk score (0.9). Finally, this record is appended to the initial defect dataset, completing one learning and enhancement of the dataset.

[0059] In step S17, based on the enhanced defect dataset and the final management file, feedback parameters are adjusted to obtain an optimized defect prevention strategy, including: Defect data containing high stress peak points are filtered from the enhanced defect dataset and combined with the preliminary risk category in the final management file to obtain a preliminary defect set; The stress peak points and the preliminary risk categories in the preliminary defect set are integrated to obtain a multidimensional fusion dataset; Based on the severity of the preliminary risk categories, the multidimensional fusion datasets are prioritized to obtain the preferred fusion datasets; The optimized defect prevention strategy is obtained by performing parameter optimization processing based on the preferred fusion dataset and adjusting the welding process feedback parameters.

[0060] In one implementation, this embodiment, to generate a precise prevention strategy, firstly, filters data related to the highest-risk operating conditions from the enhanced dataset generated in S16. This filtering is achieved by comparing the stress peak points in the data records with a preset high-stress threshold; only data with stress values ​​exceeding this threshold are selected. It should be noted that the determination of the high-stress threshold is based on statistical analysis of the distribution of all stress peak points in the enhanced dataset. Specifically, this determination process involves calculating the enhanced dataset and selecting the value corresponding to the 95th percentile of the distribution as the high-stress threshold. This method ensures that only statistically extreme and analytically valuable stress events are included in the subsequent optimization process.

[0061] In another implementation, this embodiment integrates the data selected in the previous step with the preliminary risk categories in the final management file and sorts them by severity. Specifically, this integration process involves creating an analysis record for each selected high-stress defect instance, containing its complete process parameters, peak stress value, and preliminary risk category label. All records together constitute the multidimensional fusion dataset. Subsequently, this embodiment sorts the multidimensional fusion dataset in descending order according to a preset "defect severity level table" and selects the top-ranked data to form the preferred fusion dataset. It should be noted that the defect severity level table is a knowledge base pre-established by industry experts based on fracture mechanics theory and engineering practice. It assigns a quantified severity score to each preliminary risk category (e.g., cracks = 10, lack of fusion = 7, porosity = 3). This sorting step ensures that the defect types posing the greatest threat to structural safety are prioritized for analysis and optimization.

[0062] It is worth noting that, in this embodiment, the optimized defect prevention strategy is generated by performing parameter optimization processing based on the preferred fusion dataset. The core of this processing is to construct a machine learning model capable of predicting stress levels under different process parameters. In one implementation, this parameter optimization processing can be performed by a gradient boosting regression model. The model's construction and training process is as follows: the preferred fusion dataset is used as training data, where process parameters (such as welding current, voltage, and speed) are used as input features, and the stress peak point is used as the output label that the model needs to predict; the model approximates the final prediction result by iteratively constructing a series of decision trees (called weak learners).

[0063] It should be noted that the iterative process involves first training a first decision tree to fit the training data; then, calculating the residuals between the predictions of the current ensemble model (initially consisting of only the first tree) and the true labels; in the next iteration, training a new decision tree to fit these residuals; adding the newly trained tree, weighted by a preset learning rate, to the ensemble model; repeating this process, with each subsequent tree focusing on fitting the residuals left after combining the predictions of all previous trees. Model training is completed when a preset stopping condition is met.

[0064] It should be noted that the stopping conditions may include reaching a preset maximum number of iterations (i.e., the maximum number of decision trees), or terminating early if a preset early stopping criterion is met; and the specific implementation of the early stopping criterion is to set a patience round number (e.g., 10 rounds) and a minimum improvement threshold (e.g., 10). -4During training, if the model's performance (e.g., mean squared error) on an independent validation set continuously exceeds the number of training rounds, and the improvement is less than the minimum improvement threshold, then the model's performance is determined to have plateaued, and training is terminated prematurely.

[0065] It should be noted that the key hyperparameters of this model, including the learning rate, maximum tree depth, maximum number of iterations, and the number of patience rounds and minimum improvement threshold used for early stopping criteria, were all determined in an offline modeling phase through joint optimization using a grid search combined with k-fold cross-validation, with the objective of minimizing the mean squared error (MSE) on the validation set. An exemplary grid search space can be set as follows: candidate values ​​for the learning rate are {0.01, 0.05, 0.1, 0.2}; candidate values ​​for the maximum tree depth are {3, 5, 7}; candidate values ​​for the maximum number of iterations are {100, 200, 500}; candidate values ​​for the number of patience rounds are {10, 20}; and candidate values ​​for the minimum improvement threshold are {10, 20}. -4 10 -5 Through this joint optimization process, this embodiment selects a set of hyperparameters that achieves the best balance between preventing model overfitting and ensuring sufficient training, and applies it to the final model training and early stopping judgment.

[0066] It should be noted that after the model training is completed, this embodiment employs an optimization search algorithm, such as the coordinate descent method, with a step size set to 1% of the parameter range. The convergence condition is that the improvement margin is less than 1e-4 for three consecutive iterations. The goal is to minimize the stress peak point predicted by the model, searching for the optimal parameter combination within the preset safe operating range of each process parameter. It should be noted that the safe operating range is preset according to the Welding Procedure Specification (WPS) and equipment physical limitations. For example, the safe range for welding current can be set to [150A, 200A], and the welding speed can be set to [0.4m / min, 0.6m / min]. The optimization search algorithm will search for the optimal solution within this multi-dimensional range. Starting from the current process parameter point, the algorithm adjusts each parameter in turn with small steps, and uses the trained model to predict the adjusted stress result until a parameter combination that minimizes the predicted stress value is found. The difference between this optimal parameter combination and the current parameters is formatted as a set of specific adjustment suggestions.

[0067] For example, suppose the analysis object is a welded joint of a high-strength low-alloy steel (e.g., Q550D, whose yield strength is not less than 550MPa). Its preferred fusion dataset contains multiple records, all pointing to "crack-like" defects under high stress (>550MPa), primarily caused by the process parameter combination {current: 180A, speed: 0.5m / min}. Based on this data, this embodiment trains a model and optimizes the search algorithm, finding that when the parameters are adjusted to {current: 172A, speed: 0.53m / min}, the model-predicted stress peak can be reduced to 440MPa (below the risk threshold). Therefore, the optimized defect prevention strategy ultimately generated by the system is a clear instruction: "For high-risk 'crack-like' scenarios, it is recommended to adjust the welding current to 172A and the welding speed to 0.53m / min."

[0068] It should be noted that after the optimized defect prevention strategy is generated, the system directly sends the optimized process parameter instructions to the welding equipment controller through the integrated industrial communication interface (such as OPC UA, Modbus TCP) to realize online automatic adjustment of process parameters; or the optimized strategy and adjustment suggestions are pushed to the production operator through the human-machine interface (HMI), and after manual confirmation, the operator manually inputs the data into the equipment for execution; through the above methods, the present invention realizes a complete closed-loop control from defect detection, risk assessment to process optimization.

[0069] In summary, this invention solves the problems of rigid management strategies and low accuracy of risk assessment in existing technologies by constructing a complete management process from multi-dimensional data dynamic perception and intelligent risk assessment to closed-loop feedback optimization, thereby achieving accurate quantification of welding defect risks and adaptive optimization of production processes.

[0070] Reference Figure 2 The second embodiment of the present invention provides a welding defect data management system based on an equivalent crack model, used to implement the above-mentioned welding defect data management method based on an equivalent crack model, including: The data acquisition and preprocessing module is used to collect working condition data and process parameters of welding defects, and perform preprocessing to obtain an initial defect dataset; The feature extraction and weighting module is used to extract features and adjust weights based on the initial defect dataset to obtain a weighted feature vector; The risk classification module is used to perform clustering based on the weighted feature vector to obtain preliminary risk categories; The risk quantification module is used to perform risk quantification calculations based on the initial defect dataset to obtain a risk score; The decision-making and archiving module is used to filter and generate reports based on the risk score and the preliminary risk category to obtain the final management file; The data enhancement module is used to perform data fusion and update based on the final management file and the initial defect dataset to obtain an enhanced defect dataset; The strategy generation module is used to adjust feedback parameters based on the enhanced defect dataset and the final management file to obtain an optimized defect prevention strategy.

[0071] It should be noted that the welding defect data management system based on the equivalent crack model provided in this embodiment of the invention is used to execute all the process steps of the welding defect data management method based on the equivalent crack model in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0072] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a welding defect data management program based on an equivalent crack model. When the processor executes the computer program, it implements the steps in the various embodiments of the welding defect data management method based on the equivalent crack model described above, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above system embodiments, such as the risk classification module.

[0073] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0074] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0075] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0076] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0077] If the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0078] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0079] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A welding defect data management method based on an equivalent crack model, characterized by, The method comprises the following steps: Collecting and preprocessing the working condition data and process parameters of welding defects to obtain an initial defect data set; According to the initial defect data set, feature extraction and weight adjustment are performed to obtain a weighted feature vector; According to the weighted feature vector, clustering processing is performed to obtain a preliminary risk category; According to the initial defect data set, risk quantification calculation is performed to obtain a risk score; According to the risk score and the preliminary risk category, a report is screened and generated to obtain a final management archive; According to the final management archive and the initial defect data set, data fusion update is performed to obtain an enhanced defect data set; According to the enhanced defect data set and the final management archive, feedback parameter adjustment is performed to obtain an optimized defect prevention strategy.

2. The welding defect data management method based on the equivalent crack model according to claim 1, characterized by, The method comprises the following steps: Collecting and preprocessing the working condition data and process parameters of welding defects to obtain an initial defect data set, comprising: Collecting geometric size, material hardness value, stress peak point, defect position coordinate, environmental temperature and service environment data as the working condition data, and combining the process parameters to form raw data; 3. The welding defect data management method based on the equivalent crack model according to claim 2, characterized by, The raw data is subjected to data cleaning and normalization processing to obtain the initial defect data set. The method comprises the following steps: From the initial defect data set, the original physical features are extracted and spliced to form an initial feature vector; 4. The welding defect data management method based on the equivalent crack model according to claim 1, characterized by, If the stress peak point component in the initial feature vector exceeds the preset stress threshold, a group of preset risk weight coefficients is multiplied with the predetermined risk-related components in the initial feature vector to obtain the weighted feature vector. The method comprises the following steps: Grouping processing is performed on the weighted feature vector to determine a plurality of cluster centers; The similarity between each weighted feature vector and each cluster center is calculated to obtain a similarity value; 5. The welding defect data management method based on the equivalent crack model according to claim 2, characterized by, According to the similarity value, the weighted feature vector is assigned to the category corresponding to the cluster center with the highest similarity to obtain the preliminary risk category. The method comprises the following steps: The equivalent parameter is calculated according to the geometric size; The equivalent parameter is subjected to standardization processing to obtain a standardized quantization parameter set; Linear mapping calculation is performed according to the standardized quantization parameter set and the material hardness value in the initial defect data set to obtain an initial risk score; The initial risk score is compared with a preset score interval, and the initial risk score is adjusted when it exceeds the preset score interval to obtain an adjusted risk score; 6. The weld defect data management method based on the equivalent crack model according to claim 1, characterized by, According to the preliminary risk category, a preset category weight factor is assigned to the adjusted risk score, and weighted calculation is performed to obtain the risk score. The method comprises the following steps: The risk score is compared with a preset risk threshold, defect data with a risk score higher than the preset risk threshold is screened out to form a high-risk defect subset; According to the high-risk defect subset, a preset report template is used to generate an automatic report set; The automatic report set is stored to obtain the final management archive.

7. The weld defect data management method based on the equivalent crack model according to claim 2, characterized by, According to the final management archive and the initial defect data set, data fusion update is performed to obtain an enhanced defect data set, including: The process parameters are obtained from the final management archive, and defect data containing the geometric shape size is screened from the initial defect data set; The process parameters and the geometric shape size are fused to obtain a multi-dimensional data set; The multi-dimensional data set is prioritized according to the risk score associated with the geometric shape size, and an optimal data set is obtained by screening; The initial defect data set is updated according to the optimal data set to obtain the enhanced defect data set.

8. The weld defect data management method based on the equivalent crack model according to claim 1, characterized by, According to the enhanced defect data set and the final management archive, feedback parameter adjustment is performed to obtain an optimized defect prevention strategy, including: Defect data containing a high stress peak point is screened from the enhanced defect data set, and a preliminary defect set is obtained by combining the risk preliminary category in the final management archive; The stress peak point and the risk preliminary category in the preliminary defect set are integrated to obtain a multi-dimensional fusion data set; The multi-dimensional fusion data set is prioritized according to the severity of the risk preliminary category to obtain an optimal fusion data set; The optimal fusion data set is used for parameter optimization processing, and the welding process feedback parameters are adjusted to obtain the optimized defect prevention strategy.

9. A weld defect data management system based on an equivalent crack model, characterized by, A welding defect data management method based on an equivalent crack model is implemented, including: A data acquisition and preprocessing module is used to acquire and preprocess the working condition data and process parameters of the welding defect to obtain an initial defect data set; A feature extraction and weighting module is used to extract features and adjust weights based on the initial defect data set to obtain a weighted feature vector; A risk classification module is used to perform clustering based on the weighted feature vector to obtain a risk preliminary category; A risk quantification module is used to perform risk quantification calculation based on the initial defect data set to obtain a risk score; A decision and archiving module is used to screen and generate reports based on the risk score and the risk preliminary category to obtain a final management archive; A data enhancement module is used to perform data fusion update based on the final management archive and the initial defect data set to obtain an enhanced defect data set; A strategy generation module is used to perform feedback parameter adjustment based on the enhanced defect data set and the final management archive to obtain an optimized defect prevention strategy.

Images