A nondestructive testing device and detection method for engineering machinery heterogeneous components

Abstract

The present invention discloses a non-destructive testing device and detection method for heterogeneous components of engineering machinery. The method comprises: obtaining material properties and historical failure data of heterogeneous components, extracting defect morphology and position characteristic parameters, constructing a statistical distribution model and generating a typical feature set; based on the statistical distribution model and material properties, using numerical simulation to generate a virtual defect data set containing morphology and position characteristics; performing finite element analysis on the virtual defect data set, calculating stress concentration effects and expansion trends, and outputting a quantitative indicator of defect hazard; if the hazard indicator exceeds a preset threshold, it is marked as a high-risk defect and a priority simulation target set is generated, etc. The method of the present invention covers all aspects of non-destructive testing from data acquisition to standard construction, forming a complete set of workflows that can systematically solve the problem of non-destructive testing of heterogeneous components of engineering machinery.

Description

Technical Field

[0001] The present invention relates to the technical field of non-destructive testing, and in particular to a non-destructive testing device and a testing method for a special component of an engineering machinery. Background Art

[0002] Nondestructive testing of mechanical components is a crucial research area for ensuring the safe operation of mechanical equipment. Its core focus is on detecting defects within or on the surface of materials through non-destructive means, ensuring the reliability and stability of critical components under complex operating conditions. Research in this area is directly related to the safety and efficiency of industrial production, especially in high-risk industries such as aviation, energy, and rail transit, where even the slightest defect can lead to catastrophic consequences. Its importance is self-evident.

[0003] However, current nondestructive testing methods have significant limitations in practical applications. Researchers typically verify the effectiveness of detection techniques by artificially creating defects in test samples. However, the morphology, size, and distribution of these artificial defects often differ significantly from those naturally formed in actual use. This makes it difficult for test results to truly reflect the performance of the device in real-world conditions, limiting the practical value of the detection methods.

[0004] A deeper analysis of this issue reveals that the core challenge lies in making artificial defects more closely resemble the characteristics of real defects. Due to the complex material composition and geometric shapes of different components, their sensitivity to defects varies significantly. The design of artificial defects lacks specificity and is unable to simulate the formation mechanisms and impacts of real defects. This design shortcoming further leads to a lack of comprehensiveness in the establishment of testing standards and evaluation systems, making it impossible to provide customized testing criteria for different component types, which in turn affects the accuracy and reliability of nondestructive testing in practical applications.

[0005] Therefore, how to design artificial defects that are highly similar to actual defects based on the specific characteristics of heterogeneous components, and on this basis build a comprehensive and reliable detection standard and evaluation system, has become a key issue that needs to be urgently solved in the field of non-destructive testing of mechanical heterogeneous components. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide a non-destructive testing device and detection method for heterogeneous components of engineering machinery, which can solve at least one technical problem mentioned in the background technology.

[0007] According to one aspect of the present invention, a non-destructive testing method for a foreign component of an engineering machinery is provided, the method comprising:

[0008] Obtain material properties and historical failure data of heterogeneous components, extract defect morphology and location characteristic parameters, build statistical distribution models and generate typical feature sets;

[0009] generating a virtual defect data set including morphological and positional characteristics using numerical simulation according to the statistical distribution model and material properties;

[0010] Performing finite element analysis on the virtual defect data set, calculating stress concentration effects and expansion trends, and outputting quantitative indicators of defect hazard;

[0011] If the hazard index exceeds a preset threshold, it is marked as a high-risk defect and a priority simulation target set is generated;

[0012] Based on the high-risk defect set, artificial defects matching the virtual defects are generated in the test sample using micromachining technology;

[0013] Perform multimodal nondestructive testing on test samples containing artificial defects. If the matching degree between the signal characteristics and the preset feature library is lower than the threshold, adjust the detection parameters and re-collect data;

[0014] Training a defect recognition model based on the optimized inspection data, and if the model accuracy does not reach a threshold, retroactively adjusting generation parameters of the virtual defect dataset;

[0015] Integrate updated defect datasets, optimize detection parameters and recognition models, and build customized non-destructive testing standards and evaluation frameworks.

[0016] In the above technical solution, this method is a systematic and comprehensive non-destructive testing method, covering the complete process from data acquisition, virtual defect generation, analysis and evaluation, artificial defect simulation, multimodal detection to model training and standard construction, aiming to effectively detect defects in heterogeneous components of engineering machinery and ensure their safety and reliability.

[0017] Data Acquisition and Feature Extraction: Obtaining material properties and historical failure data for heterogeneous components provides foundational information for subsequent analysis. Material properties help understand the physical characteristics of components, while historical failure data records past failures. Extracting defect morphology and location characteristic parameters from this data allows for targeted focus on key defect characteristics, building statistical distribution models, and generating a set of typical features, laying the foundation for subsequent virtual defect dataset generation.

[0018] Generation of virtual defect datasets: Based on the previously constructed statistical distribution model and material properties, numerical simulation technology is used to generate a virtual defect dataset containing morphological and positional characteristics. This can simulate various possible defect situations without destroying the actual components, enrich the defect data samples, and provide sufficient data support for subsequent analysis and model training.

[0019] Finite Element Analysis and Criticality Assessment: Finite element analysis is performed on virtual defect datasets to calculate stress concentration effects and expansion trends, and then output a quantitative indicator of defect criticality. This process assesses the impact of defects on components from a mechanical perspective and identifies which defects pose higher risks, providing a basis for subsequent focus and treatment. If the criticality indicator exceeds a preset threshold, it is marked as a high-risk defect and a priority simulation target set is generated, effectively screening and classifying defect risks.

[0020] Artificial Defect Generation and Multimodal Detection: Based on a set of high-risk defects, micromachining technology is used to generate artificial defects in test samples that match the virtual defects, enabling more realistic detection studies on actual test samples. Multimodal nondestructive testing is performed on test samples containing artificial defects. Multimodal testing can combine the advantages of different detection methods to improve the accuracy and reliability of defect detection. If the match between the signal characteristics and the preset feature library falls below a threshold, the detection parameters are adjusted and data is re-collected, ensuring the quality and validity of the test results.

[0021] Model training and parameter optimization: The defect recognition model is trained based on the optimized inspection data. If the model accuracy does not reach the threshold, the generation parameters of the virtual defect dataset are retroactively adjusted. This feedback optimization mechanism helps to continuously improve the performance of the defect recognition model, enabling it to more accurately identify various defect situations and ensure the reliability and applicability of the model.

[0022] Construction of customized testing standards and evaluation framework: Ultimately, the updated defect data set is integrated, and the testing parameters and identification models are optimized to build a customized non-destructive testing standard and evaluation framework. This can develop testing standards and evaluation methods that meet the characteristics and actual needs of specific engineering machinery components, provide scientific and standardized guidance for actual non-destructive testing work, improve testing efficiency and quality, and ensure the safe operation of engineering machinery.

[0023] The method of the present invention covers all aspects of non-destructive testing from data acquisition to standard construction, forming a complete set of work processes, which can systematically solve the problem of non-destructive testing of heterogeneous components of engineering machinery. Based on material properties, historical failure data, and advanced numerical simulation, finite element analysis and other technical means, the detection method has a solid scientific basis and improves the accuracy and reliability of the detection results. By marking and prioritizing high-risk defects, and building customized detection standards and evaluation frameworks, it is possible to conduct detection based on the characteristics and actual needs of heterogeneous components of engineering machinery, with strong pertinence and practicality. Introducing a feedback optimization mechanism in the process of model training and parameter adjustment to continuously optimize the generation parameters and detection parameters of the virtual defect data set will help improve the performance and effectiveness of the entire detection system.

[0024] In some embodiments, obtaining material properties and historical failure data of a foreign component, extracting defect morphology and location characteristic parameters, building a statistical distribution model, and generating a typical feature set include:

[0025] The initial data set is constructed by obtaining the material properties and historical failure records of heterogeneous components from the database;

[0026] Based on the initial data set, preprocessing is performed and relevant parameters of defect morphology and location characteristics are extracted to determine the characteristic parameter data set;

[0027] Based on the feature parameter data set, a probability density function is applied to construct a statistical distribution model to determine the distribution characteristics. If the distribution characteristics meet the preset threshold range, a typical feature set is generated based on the statistical distribution model. If not, the process returns to the previous step to re-extract the feature parameters.

[0028] Based on the typical feature set, combined with the position features and defect morphology, feature clustering analysis is performed, and the K-means clustering algorithm is used to divide the feature categories to obtain the classification results;

[0029] Based on the classification results, correlation analysis is performed on each type of features to determine the mapping relationship between the feature set and the failure record.

[0030] In the above technical solution, this example method is the preliminary data processing and feature analysis stage of the entire non-destructive testing method. Through a series of operations such as acquisition, preprocessing, feature extraction, statistical modeling, cluster analysis, and correlation analysis of the relevant data of heterogeneous components, it provides a key feature set and association mapping basis for subsequent virtual defect data set generation and other links. It is the foundation of the entire method process and ensures that subsequent work can be carried out based on accurate and representative features.

[0031] Initial dataset construction: The material properties and historical failure records of heterogeneous components are obtained from the database to construct the initial dataset. This is the data source for the entire process. The material property data can reflect the physical characteristics of the component, and the historical failure records provide specific case information on past defects that led to failure, providing rich background and basic data resources for subsequent analysis.

[0032] Data preprocessing and feature parameter extraction: The initial data set is preprocessed to remove noise, missing values, outliers, and other issues to ensure data quality and usability. Parameters related to defect morphology and location characteristics are then extracted to determine the feature parameter data set.

[0033] Statistical distribution model construction and typical feature set generation: Using the feature parameter data set, a statistical distribution model is constructed using the probability density function to determine the distribution characteristics. The probability density function can describe the probability distribution of feature parameters within different value ranges, thereby understanding the statistical laws of defect feature parameters. If the distribution characteristics meet the preset threshold range, a typical feature set is generated based on the statistical distribution model. These typical feature sets represent defect characteristics that are statistically representative and common. If not, the feature parameters are re-extracted. This reflects the strict control of the quality and representativeness of the feature parameters, ensuring that the generated typical features can accurately reflect the distribution of actual defect characteristics.

[0034] Feature Cluster Analysis: Based on a set of typical features, combined with location features and defect morphology, feature cluster analysis is performed. The K-means clustering algorithm is used to classify features into categories and generate classification results. Cluster analysis can group similar features together. The K-means algorithm is a commonly used clustering method that calculates similarities (such as distance) between features to classify them into different categories. This allows for further subdivision of defect feature types and patterns, providing a foundation for more targeted analysis and processing.

[0035] Correlation Analysis and Mapping Determination: Based on the classification results, correlation analysis is performed on each feature type to determine the mapping relationship between the feature set and failure records. This step aims to establish a connection between defect characteristics and actual failure situations. By analyzing the performance of components with different feature categories in historical failure records, it is possible to identify which feature combinations are more likely to lead to failure. This provides a basis for subsequent risk assessment and detection focus determination, and also helps to better understand the impact of defect characteristics on component performance.

[0036] The above method is based on actual material properties and historical failure data. Through scientific data processing and analysis methods, it extracts representative and critical defect characteristic parameters, so that subsequent inspection and evaluation work can closely match the defect conditions of the actual components. In the process of constructing the statistical distribution model and generating the typical feature set, a judgment and feedback adjustment mechanism for the distribution characteristics is set up to ensure that the extracted and generated features have good statistical rationality and representativeness, avoiding subsequent analysis errors caused by improper feature selection. By classifying the features through cluster analysis and further exploring the mapping relationship between the feature set and the failure record, we can deeply understand the role and connection of different defect features in the failure process, and provide strong support for the subsequent formulation of more targeted detection strategies and risk assessment standards.

[0037] In some embodiments, based on the statistical distribution model and material properties, a virtual defect dataset containing morphological and positional features is generated using numerical simulation, including:

[0038] Using the statistical distribution and material properties, a pre-established model building method is used to obtain initial defect distribution parameters and a preliminary distribution framework;

[0039] Based on the preliminary distribution framework and combined with numerical simulation technology, simulation calculations are performed on the morphological and positional characteristics to determine the geometric shape and spatial coordinates of the virtual defect;

[0040] Use simulation technology to iteratively optimize the distribution of virtual defects. If the calculated results do not match the preset threshold of the material properties, adjust the defect distribution parameters, reacquire the simulation data, and determine whether the distribution consistency is met.

[0041] Through the adjusted simulation data, a virtual data set containing various morphological features is constructed for defect distribution and feature extraction, thus obtaining a diverse set of defect samples.

[0042] Based on a diverse set of defect samples, combined with location features and attribute analysis, the random forest algorithm is used to classify the data set and determine the spatial distribution patterns of different types of defects;

[0043] Based on the spatial distribution law after classification processing, the final virtual defect dataset is generated for virtual defects and datasets to obtain a complete description of defect characteristics;

[0044] If there is a deviation between the feature description of the final data set and the preset statistical distribution, the simulation parameters of the morphological and positional features are readjusted through simulation technology to determine whether the consistency requirements are met.

[0045] In the above technical solution, the above method is the key link in the generation of virtual defect data sets. Based on the statistical distribution model and material properties constructed previously, through numerical simulation, analog calculation, iterative optimization and data classification and other technical means, a virtual defect data set containing morphological and positional characteristics is gradually generated, and its consistency with material properties and preset statistical distribution is ensured, providing accurate, diverse and representative defect data samples for subsequent finite element analysis and other steps, which plays an important supporting role in the effectiveness and reliability of the entire non-destructive testing method.

[0046] Initial defect distribution parameter acquisition: Leveraging statistical distributions and material properties, a pre-established model-building method is employed to obtain initial defect distribution parameters and establish a preliminary distribution framework. This step is the starting point for generating the entire virtual defect dataset. The choice of model-building method is crucial, as it needs to be able to rationally combine statistical distributions and material properties to convert them into specific defect distribution parameters, thus providing a foundational framework for subsequent simulation calculations.

[0047] Simulation and calculation of morphological and positional characteristics: Based on the preliminary distribution framework and incorporating numerical simulation techniques, simulation calculations are performed on the morphological and positional characteristics to determine the geometric shape and spatial coordinates of the virtual defects. Numerical simulation technology plays a key role here, transforming abstract distribution parameters into specific virtual defect morphology and location information through simulation calculations. The accuracy of the geometric shape directly impacts the results of subsequent finite element analysis and other methods for defect hazard assessment, while the spatial coordinates reflect the actual distribution of defects within the component.

[0048] Iterative Optimization and Distribution Consistency Assessment: Simulation technology is used to iteratively optimize the distribution of virtual defects. If the calculated results do not match the preset thresholds for material properties, the defect distribution parameters are adjusted, the simulated data is re-acquired, and the distribution consistency is determined. The iterative optimization process ensures that the virtual defect distribution matches the reasonable distribution corresponding to the actual material properties, avoiding defect distributions that do not conform to reality, thereby improving the authenticity and reliability of the virtual defect dataset. The preset thresholds must be determined based on professional knowledge of material properties and failure mechanisms to ensure that the optimized distribution is meaningful.

[0049] Diversified defect sample set construction: Using adjusted simulated data, we construct a virtual dataset containing a variety of morphological features based on defect distribution and feature extraction, resulting in a diverse defect sample set. This process aims to enrich the sample types and morphological features of the virtual defect dataset, enabling subsequent analysis and model training to be based on more comprehensive and representative data samples, thereby improving the ability to identify and evaluate various types of defects.

[0050] Spatial distribution pattern classification: Based on a diverse set of defect samples, combined with location characteristics and attribute analysis, the random forest algorithm is used to classify the dataset and determine the spatial distribution patterns of different defect categories. The random forest algorithm is a highly efficient machine learning classification method that can effectively process large datasets and discover spatial distribution patterns under different feature categories. This helps to deeply understand the distribution patterns and characteristics of different types of defects in components, providing a basis for further risk assessment and detection strategy development.

[0051] Final virtual defect dataset generation and verification: Based on the spatial distribution patterns after classification processing, the final virtual defect dataset is generated for the virtual defects and datasets, obtaining a complete description of the defect characteristics. If the characteristic description of the final dataset deviates from the preset statistical distribution, the simulation parameters of the morphological and positional characteristics are readjusted through simulation technology to determine whether consistency requirements are met. This step is the final verification and correction of the entire virtual defect dataset generation process, ensuring that the generated dataset matches the expected statistical distribution in terms of characteristic description and meets the requirements of non-destructive testing methods for data accuracy and representativeness.

[0052] The above method comprehensively considers the statistical distribution model and material properties, and organically combines the two through technical means such as numerical simulation, so that the generated virtual defect dataset can more realistically reflect the distribution of defects in the actual component, thereby improving the reliability and practicality of the dataset. Introducing an iterative optimization link in the simulation process of the virtual defect distribution can continuously correct and adjust the defect distribution parameters to ensure that they are consistent with the preset thresholds of the material properties, thereby ensuring a high degree of consistency between the virtual defect dataset and the actual component characteristics. Constructing a virtual dataset containing multiple morphological features enriches the sample types and features, which is conducive to improving the performance and generalization ability of subsequent defect recognition models and enhancing the detection and evaluation effects of different types of defects. The random forest algorithm is used for classification processing to reveal the spatial distribution patterns of different types of defects, providing strong support for subsequent more accurate risk assessment and detection decisions, and helping to achieve priority attention and treatment of high-risk defects.

[0053] In some embodiments, finite element analysis is performed on the virtual defect dataset to calculate stress concentration effects and expansion trends, and output a quantitative indicator of defect hazard, including:

[0054] By processing the virtual defect data set, the initial defect characteristic data is obtained, and the calculation model is constructed using the finite element analysis method to obtain the stress distribution status of the defect area;

[0055] Based on the stress distribution, the stress concentration effect is calculated, and local mesh encryption is performed on high-stress areas to determine the key locations of stress concentration. At these key locations of stress concentration, the defect expansion trend is simulated, and finite element iteration is used to obtain expansion path and velocity change data. From these expansion path and velocity change data, expansion trend characteristics are extracted and combined with a preset threshold range to determine the potential risk level of defect expansion.

[0056] If the potential risk level of defect expansion exceeds the preset threshold, a weighted analysis is performed on the expansion trend characteristics to obtain a preliminary value of the quantitative indicator of harmfulness;

[0057] The final defect criticality quantitative index is determined by performing multi-dimensional calibration on the preliminary value of the criticality quantitative index and integrating stress distribution and expansion trend data.

[0058] In this technical solution, the above-mentioned process is a key step in conducting in-depth analysis of virtual defect datasets to assess their criticality. Through finite element analysis, local mesh refinement, iterative calculations, and multi-dimensional calibration, the stress concentration effects and expansion trends in the defect area are accurately calculated, ultimately quantifying the criticality of the defect. This provides accurate data support for the subsequent identification of high-risk defects, the development of detection strategies, and the construction of an assessment framework, which is of great significance for ensuring the safety and reliability of heterogeneous components in construction machinery.

[0059] Initial defect characteristic data processing and stress distribution calculation: By processing the virtual defect data set, the initial defect characteristic data is obtained, and the calculation model is constructed using the finite element analysis method to obtain the stress distribution status of the defect area. This is the basic step of the entire analysis. The quality and accuracy of the virtual defect data set directly affect the construction and calculation results of the finite element analysis model. As a numerical calculation method, finite element analysis can discretize complex continua into a finite number of units and nodes, thereby solving its mechanical responses such as stress and strain. When constructing the calculation model, it is necessary to accurately incorporate information such as the geometric shape, spatial position, and material properties of the virtual defects into the model to ensure the accuracy of the stress distribution calculation.

[0060] Stress Concentration Effect Calculation and Local Mesh Refinement: Calculate stress concentration effects based on stress distribution. Stress concentration is a significant increase in stress around a defect area and is often the root cause of component failure. Local mesh refinement is performed in high-stress areas to improve the accuracy of the calculated stress distribution details. Local mesh refinement increases the number of elements in high-stress areas and reduces element size, allowing for more accurate capture of subtle stress variations and pinpointing of key stress concentration locations, providing a more reliable basis for subsequent defect growth trend simulations.

[0061] Defect expansion trend simulation and feature extraction: Targeting key locations of stress concentration, defect expansion trends are simulated, and finite element iterative calculations are used to obtain expansion path and velocity change data. Finite element iterative calculations can gradually update the geometry and position of defects during the simulation process based on the physical laws of defect expansion and material failure criteria, thereby predicting their expansion path and velocity changes. Extracting expansion trend features from the expansion path and velocity change data is a step designed to transform the complex expansion process into quantifiable characteristic parameters, such as expansion direction and expansion rate, to facilitate subsequent analysis and risk assessment. Combining the preset threshold range to determine the potential risk level of defect expansion helps to classify the severity of defect expansion and provide guidance for subsequent risk management.

[0062] Preliminary calculation and calibration of the criticality quantification index: If the potential risk level of defect expansion exceeds the preset threshold, a weighted analysis of the expansion trend characteristics is performed to obtain a preliminary value for the criticality quantification index. The weighted analysis considers the degree of impact of different expansion trend characteristics on criticality, assigning different weights to comprehensively assess the criticality of the defect. A multi-dimensional calibration of the preliminary value of the criticality quantification index is performed, integrating stress distribution and expansion trend data to determine the final defect criticality quantification index. Multi-dimensional calibration comprehensively considers multiple factors such as the defect's stress concentration level, expansion trend, material properties, and the component's operating environment, thereby more comprehensively and accurately quantifying the defect's criticality, providing a scientific basis for subsequent high-risk defect identification and detection parameter optimization.

[0063] The above method uses finite element analysis to accurately calculate the stress distribution and stress concentration effects in defect areas. Combined with local mesh encryption technology, it improves the calculation accuracy of key areas, thereby more accurately assessing the potential risks of defects and providing reliable data support for subsequent inspection and evaluation. By simulating the expansion trend of defects through iterative finite element calculations, obtaining expansion path and speed change data, and extracting expansion trend characteristics, it is possible to deeply understand the possible development direction and speed of defects during use, which helps to take preventive measures in advance to avoid further deterioration of defects and cause component failure. Using a method that combines weighted analysis and multi-dimensional calibration, the various characteristics and factors of defects are comprehensively considered to obtain the final quantitative index of defect hazard, which can more comprehensively reflect the impact of defects on component safety and provide a scientific basis for risk management and decision-making.

[0064] In some embodiments, if the hazard indicator exceeds a preset threshold, it is marked as a high-risk defect and a priority simulation target set is generated, including: obtaining hazard indicator data from the system, performing preliminary cleaning and formatting processing on it, and obtaining a normalized indicator data set; if the value in the normalized indicator data set exceeds the preset threshold, it is marked as a potential high-risk defect, and a corresponding risk determination result is generated; based on the risk determination result, a pre-established classification model is used to classify the potential high-risk defects and determine the specific defect category information; for the determined defect category information, a priority evaluation mechanism is constructed, and a priority simulation defect subset is obtained by comparing the priority level data; key features are extracted from the priority simulation defect subset, and the features are analyzed using a support vector machine algorithm to determine the direction of the applicable simulation strategy; based on the simulation strategy direction obtained by the analysis, a target set is constructed to generate the final priority simulation target set data; by storing and indexing the priority simulation target set data, a structured data set that can be used for subsequent calls is generated.

[0065] In this technical solution, the method primarily targets defects whose hazard indicators exceed a preset threshold. Through a series of steps, including data cleaning, risk assessment, defect classification, priority assessment, and simulation strategy analysis, it ultimately generates a prioritized set of simulation target data for subsequent use. This process provides clear priority guidance for subsequent simulation and testing work, helping to focus resources on analyzing and addressing high-risk defects, thereby improving the efficiency and relevance of the entire nondestructive testing process.

[0066] Hazard indicator data acquisition and cleaning: Obtaining hazard indicator data from the system is the foundation for subsequent processing, and the accuracy and completeness of the data are crucial. Initial cleaning and formatting are performed to remove errors, missing values, duplicate values, and other issues. The data is then converted to a unified format to obtain a standardized indicator dataset, providing high-quality data support for subsequent analysis and ensuring accurate risk assessment and other steps.

[0067] Potentially high-risk defect marking and risk assessment: If the value in the normalized indicator dataset exceeds the preset threshold, it is marked as a potential high-risk defect, and the corresponding risk assessment result is generated. The preset threshold is determined based on relevant standards, specifications, and engineering experience. It provides a clear quantitative standard for determining whether a defect is high-risk. Through this assessment step, defects that may pose a significant threat to component safety can be quickly screened out, providing a basis for subsequent focus and treatment.

[0068] Classification of Potential High-Risk Defects: Based on the risk assessment results, a pre-established classification model is used to classify potential high-risk defects and determine the specific defect category. This classification model can be based on machine learning algorithms (such as decision trees and neural networks) or traditional classification rules. Its purpose is to categorize high-risk defects of different types and characteristics, enabling more targeted simulation and detection strategies for each category. This also facilitates subsequent in-depth defect analysis and management.

[0069] Priority Assessment and Derivation of Prioritized Defect Subsets for Simulation: A priority assessment mechanism is established for identified defect categories. By comparing priority data, a subset of defects for simulation is determined. This priority assessment mechanism comprehensively considers multiple factors, such as the severity of the defect, probability of occurrence, impact on critical components, and difficulty of repair. Priority levels are assigned to different defects to determine which defects warrant prioritization for simulation and analysis, ensuring that limited resources are prioritized to address the most critical issues.

[0070] Key Feature Extraction and Simulation Strategy Analysis: Key features are extracted from the priority defect subset and analyzed using a support vector machine algorithm to determine the appropriate simulation strategy. The support vector machine algorithm is an effective supervised learning method that can classify and analyze existing data features. By interpreting its analysis results, the appropriate simulation strategy for these key features can be determined, such as whether a finer mesh, more complex material models, or more advanced boundary conditions are needed to ensure the accuracy and reliability of the simulation results.

[0071] Prioritized simulation target set construction and data processing: Based on the simulation strategy direction derived from the analysis, a target set is constructed, generating the final prioritized simulation target set data. This data set integrates the screened and analyzed high-risk defect information and the corresponding simulation strategy, providing clear guidance and basis for subsequent simulation work. By storing and indexing the prioritized simulation target set data, a structured data set is generated for subsequent access. This step facilitates quick and convenient access to required data during subsequent simulation and testing processes, improving work efficiency and standardization.

[0072] The above method uses preset thresholds to quickly screen the hazard indicators of defects, mark potential high-risk defects, and further determine the subset of defects that are prioritized for simulation. This allows limited resources to be concentrated on the analysis and treatment of key defects, improving the efficiency and pertinence of the entire non-destructive testing process and avoiding ineffective calculation and analysis of a large number of low-risk defects. Potential high-risk defects are classified using a pre-established classification model, and combined with a priority assessment mechanism, corresponding simulation strategies can be formulated based on the characteristics and priority levels of different types of defects, improving the accuracy and effectiveness of the simulation and making subsequent inspection and evaluation work more targeted and practical. A support vector machine algorithm is used to analyze key features and determine the direction of the simulation strategy. The introduction of intelligent data analysis methods can better mine the potential information in the data, provide a scientific basis for the selection of simulation strategies, and improve the intelligence level and decision-making support capabilities of the entire process.

[0073] In some embodiments, based on the high-risk defect set, artificial defects matching virtual defects are generated in a test sample using micromachining technology, including:

[0074] By extracting characteristic data from a set of high-risk defects, a digital model of virtual defects is constructed;

[0075] Micromachining technology is used to perform preliminary surface treatment on the test sample to obtain a sample matrix suitable for processing;

[0076] For the digital model of the virtual defect, processing instructions that match the high-risk defect are generated and the processing path is determined. If the processing instruction does not meet the preset accuracy threshold, the instruction parameters are adjusted through the information processing link to obtain an optimized processing plan. According to the optimized processing plan, artificial defects are generated on the test sample using micro-machining technology to determine whether the generated results are consistent. If the deviation between the generated artificial defect and the virtual defect feature exceeds the preset range, the source of the deviation is analyzed through the information processing link to determine the correction parameters. The micro-machining technology is calibrated using the correction parameters, and the artificial defect is regenerated on the test sample to obtain the final matching result.

[0077] In the above technical solution, the above method is mainly based on a set of high-risk defects, and uses micromachining technology to generate artificial defects in the test sample that match the virtual defects, so as to provide physical sample support for subsequent multimodal non-destructive testing, ensuring that the detection methods and models can be effectively verified and optimized under actual conditions. It is a key link connecting virtual simulation and actual testing, and is of great significance to the practicality and reliability of the entire non-destructive testing method.

[0078] Constructing a digital model of virtual defects: Extracting feature data from a collection of high-risk defects forms the foundation for subsequent generation of artificial defects. The accuracy and completeness of this feature data directly impacts the quality of the digital model. Constructing a digital model of virtual defects requires integrating and modeling the extracted feature data. This model should accurately reflect key features of the virtual defects, such as their geometry, size, and location, providing a clear blueprint for subsequent micromachining operations.

[0079] Initial surface treatment of test samples: Micromachining techniques are used to perform preliminary surface treatment on test samples to obtain a suitable substrate for processing. Different micromachining techniques have different requirements for the surface condition of the substrate. For example, some photolithography techniques require a flat, smooth surface to ensure accurate pattern transfer, while some etching techniques may require a surface with specific chemical properties or microstructures. Therefore, initial surface treatment is crucial to creating favorable conditions for subsequent precision machining, improving both the success rate and quality of the process.

[0080] Machining Instructions and Path Generation: Based on the digital model of the virtual defect, machining instructions matching the high-risk defects are generated and the machining path is determined. This step requires converting the abstract features in the digital model into specific operational instructions that can be executed by the micromachining equipment. These instructions include the machining sequence, tool path, and machining parameters (such as speed, depth, and force). Machining path planning should fully consider the kinematic characteristics of the micromachining equipment, machining efficiency, and its impact on the sample substrate to ensure smooth machining and accurate results.

[0081] Processing instruction accuracy assessment and optimization: If a processing instruction does not meet the preset accuracy threshold, the instruction parameters are adjusted through information processing to obtain an optimized processing solution. The preset accuracy threshold is set based on the actual application's requirements for artificial defect accuracy. When the generated processing instruction fails to meet this accuracy requirement, the instruction parameters must be adjusted. The information processing stage may involve technologies such as error analysis and compensation algorithms. By optimizing the instruction parameters, the accuracy and feasibility of the processing instruction are improved, ensuring that the processing process can be carried out according to the expected accuracy requirements.

[0082] Artificial defect generation and result evaluation: Based on the optimized processing plan, micromachining technology is used to generate artificial defects on test samples to determine whether the generated defects are consistent. When executing processing instructions, micromachining technology may be affected by various factors, such as the mechanical precision of the equipment, environmental conditions, and material properties. This can lead to some deviation between the actual artificial defects generated and the virtual defects. Therefore, the generated results need to be tested and evaluated by comparing the actual characteristics of the artificial defects with the digital model of the virtual defects to assess their consistency. If the deviation is within an acceptable range, the generated results are considered consistent; otherwise, further analysis of the cause of the deviation and correction are required.

[0083] Deviation Analysis and Correction Parameter Determination: If the deviation between the generated artificial defect and the virtual defect characteristics exceeds the preset range, the information processing stage analyzes the source of the deviation and determines the correction parameters. Deviations may arise from multiple factors, such as inaccurate machining instructions, errors in the machining path, drift of micromachining equipment, and anisotropy of the material. The information processing stage requires a systematic analysis of these possible sources of deviation to identify the primary influencing factors and determine the correction parameters accordingly. These correction parameters include adjustments to machining instructions, optimization of machining paths, and calibration of equipment parameters. The goal is to reduce or eliminate deviations during subsequent machining processes and improve the accuracy of artificial defect generation.

[0084] Calibration of micromachining technology and obtaining final matching results: The micromachining technology is calibrated using correction parameters, and artificial defects are regenerated on the test sample to obtain the final matching results. The calibration process adjusts and optimizes the micromachining equipment and processing technology. By applying correction parameters, the performance and processing accuracy of the micromachining technology can be improved. The regenerated artificial defects should more closely match the characteristics of the virtual defects. If the machining results after calibration still do not meet the requirements, deviation analysis and correction parameter determination are required again until the final matching result is obtained, that is, the characteristics of the artificial defects are consistent with the virtual defects within the specified accuracy range.

[0085] The above method extracts characteristic data from a set of high-risk defects to construct a digital model. Combined with micromachining technology, it can accurately generate artificial defects that match the virtual defects, providing highly simulated physical samples for subsequent nondestructive testing, helping to improve the practicality and reliability of the testing methods and models. Judgment and feedback mechanisms are introduced in the processing instruction generation, result judgment, and deviation analysis stages to promptly detect and correct problems that arise during the processing. By continuously optimizing the processing plan and calibrating the micromachining technology, the resulting artificial defects are ensured to meet precision requirements, improving the efficiency and quality of the entire generation process. This process combines virtual defect design with actual micromachining technology, achieving the transformation from virtual models to physical samples. This builds a bridge for the research and application of nondestructive testing methods, enabling virtual simulation and actual testing to verify and complement each other, helping to promote the development and innovation of nondestructive testing technology.

[0086] In some embodiments, a multimodal nondestructive test is performed on a test sample containing an artificial defect. If the matching degree between a signal feature and a preset feature library is lower than a threshold, the test parameters are adjusted and data is re-collected, including:

[0087] Performing an initial scan on the test sample using multimodal detection technology to obtain a first set of signal characteristic data;

[0088] If the matching degree between the first set of signal feature data and the preset feature library is lower than the matching degree threshold, the parameter adjustment mechanism is triggered to generate adjusted detection parameters, and data is re-collected based on the adjusted parameters to obtain a second set of signal feature data; for the second set of signal feature data, the support vector machine algorithm is used to perform feature classification processing to determine whether there is an abnormal signal pattern related to artificial defects; based on the classification processing results, if an abnormal signal pattern is detected, the corresponding defect feature template is extracted from the preset feature library to determine the specific defect type of the abnormal signal; after obtaining the determined defect type, the test sample is locally enhanced scanned in combination with multiple signal sources of multimodal detection to obtain a third set of high-resolution signal feature data; through in-depth analysis of the third set of signal feature data, the convolutional neural network model is used to accurately locate the defect position and determine the distribution of defects in the test sample; based on the defect distribution, the corresponding defect recognition result is generated to complete a comprehensive evaluation of the test sample.

[0089] In the above technical solution, the above method is to perform multimodal non-destructive testing on test samples containing artificial defects, and adjust the detection parameters and resample the data based on the matching degree between the signal characteristics and the preset feature library. The signal characteristics are classified, analyzed and located through a variety of advanced algorithms, and finally the defect identification results are generated. The test samples are comprehensively evaluated to ensure the accuracy, reliability and comprehensiveness of the detection results, providing key data support for the subsequent optimization of detection parameters and training of defect recognition models.

[0090] Initial Scan and Signal Feature Acquisition: Multimodal testing technology is used to perform an initial scan of the test sample to obtain the first set of signal feature data. By integrating multiple testing methods (such as ultrasonic testing, X-ray testing, and eddy current testing), multimodal testing can obtain rich information about the test sample from different angles and physical principles, improving the comprehensiveness and accuracy of the test. The results of the initial scan provide the raw data foundation for subsequent signal feature matching and analysis.

[0091] Matching Degree Determination and Parameter Adjustment Triggering: If the matching degree between the first set of signal feature data and the preset feature library falls below the matching degree threshold, the parameter adjustment mechanism is triggered. The preset feature library contains typical signal feature patterns for various known defect types. The matching degree threshold is used to determine the reliability of the inspection data. If the matching degree falls below the threshold, it indicates that the current inspection parameters may not be able to effectively capture the characteristics of the artificial defect. Adjusted inspection parameters need to be generated and data collected again to improve inspection quality.

[0092] Re-collection and feature classification: Data is re-collected based on the adjusted parameters to obtain a second set of signal feature data. This data set is then classified using a support vector machine algorithm to determine whether there are abnormal signal patterns associated with artificial defects. The support vector machine algorithm, with its excellent classification performance, can effectively identify abnormal patterns in signal features and distinguish normal signals from abnormal signals associated with defects.

[0093] Abnormal signal pattern analysis and defect type determination: Based on the classification results, if an abnormal signal pattern is detected, the corresponding defect feature template is extracted from the preset feature library to determine the specific defect type of the abnormal signal. This step, by comparing the detected abnormal signal pattern with the template in the feature library, can quickly and accurately identify the defect type, providing a specific target for further analysis and processing.

[0094] Local Enhanced Scanning and High-Resolution Data Acquisition: After determining the defect type, a local enhanced scan of the test sample is performed using a combination of multiple signal sources from multimodal testing to obtain a third set of high-resolution signal signature data. Local enhanced scanning focuses on areas where defects may exist. By leveraging the complementary and interoperable signal sources of multimodal testing technology, the detection resolution of defective areas is improved, resulting in more detailed and accurate signal signature data, providing stronger support for subsequent defect location and assessment.

[0095] In-depth analysis and precise defect location: Through in-depth analysis of the third set of signal feature data, a convolutional neural network model is used to accurately locate defects and determine their distribution within the test sample. Convolutional neural network models have powerful capabilities in image recognition and positioning. They can deeply mine the spatial information and characteristic patterns in signal feature data to accurately determine the location and distribution of defects, providing critical positioning information for the final defect identification results.

[0096] Defect identification result generation and comprehensive evaluation: Based on the defect distribution, corresponding defect identification results are generated to complete a comprehensive evaluation of the test sample. This step integrates all previous test data and analysis results to form a complete defect identification report, covering key information such as defect type, location, and distribution. This provides a comprehensive and accurate reference for subsequent test parameter optimization, model training, and the development of non-destructive testing standards.

[0097] The above method uses a combination of various non-destructive testing technologies, giving full play to the advantages of each technology, complementing and verifying each other, and can obtain more comprehensive and richer signal feature data, effectively improving the detection accuracy and reliability of artificial defects, and reducing the risk of missed detection and false detection. With the help of advanced machine learning algorithms such as support vector machine algorithms and convolutional neural network models, signal features are accurately classified and located. It can deeply mine the potential information in the data, identify weak or complex abnormal signal patterns, and achieve accurate identification and location of defects, thereby improving the intelligence level and accuracy of detection. By setting the matching threshold and triggering the parameter adjustment mechanism, the detection quality can be evaluated in real time during the detection process, and the detection parameters can be dynamically adjusted according to the evaluation results to ensure that the collected data is always of high quality and reliability, thereby improving the adaptability and flexibility of the entire detection process.

[0098] In some embodiments, a defect recognition model is trained based on the optimized detection data. If the model accuracy does not reach a threshold, the generation parameters of the virtual defect dataset are retroactively adjusted, including:

[0099] By processing the inspection data, an initial defect recognition model is constructed, preliminary results are obtained during the training process, and it is determined whether the output of the model meets the expected standards;

[0100] If the recognition accuracy of the preliminary results does not reach the preset threshold, feature extraction is performed on the detection data to determine the key influencing factors and obtain the analysis results of the feature distribution;

[0101] Based on the analysis results of the feature distribution, the generation parameters of the virtual defect dataset are adjusted, the updated dataset content is obtained, and a new training basis is determined;

[0102] Using the updated data set content, re-execute the defect recognition model training process to obtain a new recognition accuracy value and determine whether it meets the preset standards;

[0103] If the new recognition accuracy value still does not reach the preset threshold, the adjustment direction of the generated parameters is analyzed through the backtracking mechanism to obtain the optimized parameter configuration;

[0104] Generate an improved virtual defect dataset based on the optimized parameter configuration to obtain data content that is more in line with the actual scenario and determine the final training input;

[0105] Through the final training input, the iterative training of the defect recognition model is run, and the support vector machine algorithm is used for classification processing to obtain the final model output result.

[0106] In the above technical solution, the above method mainly revolves around training a defect recognition model based on optimized inspection data, and introduces a backtracking adjustment mechanism to dynamically adjust the generation parameters of the virtual defect data set according to whether the model accuracy meets the standards, thereby continuously improving the performance of the defect recognition model, ensuring that it can accurately and reliably identify defects, and laying a solid foundation for the subsequent construction of customized non-destructive testing standards and evaluation frameworks.

[0107] Initial model construction and training: By processing inspection data, an initial defect recognition model is constructed. Preliminary training results are obtained to determine whether the model's output meets expected standards. This is the starting point of the entire training process. The quality and representativeness of the inspection data are directly related to the performance of the initial model. Preliminary model training results can reflect the basic effectiveness of the current data and model architecture on the defect recognition task, providing a reference for subsequent optimization work.

[0108] Feature Extraction and Key Factor Identification: If the initial recognition accuracy results do not meet the preset threshold, feature extraction is performed on the test data to identify key influencing factors and generate feature distribution analysis results. This step aims to deeply explore useful information in the test data and identify key features that influence the model's recognition accuracy. This feature distribution analysis provides a better understanding of the data's inherent structure and patterns, providing a basis for subsequent adjustments to the generation parameters of the virtual defect dataset.

[0109] Updating the Virtual Defect Dataset: Based on the analysis of feature distribution, adjust the generation parameters of the virtual defect dataset, obtain the updated dataset content, and determine the new training basis. The generation parameters of the virtual defect dataset have a significant impact on the quality and representativeness of the dataset. By adjusting these parameters, a dataset that better meets actual needs and model training requirements can be generated, providing better data support for retraining the defect recognition model.

[0110] Model retraining and accuracy assessment: Using the updated dataset, retrain the defect recognition model to obtain a new accuracy value and determine whether it meets the preset standards. This step verifies the effectiveness of the updated dataset in improving model performance. If the new accuracy value still does not meet the preset threshold, further analysis is required and appropriate optimization measures should be implemented.

[0111] Backtracking mechanism and parameter optimization: If the new recognition accuracy value still does not reach the preset threshold, the backtracking mechanism analyzes the parameter adjustment direction and obtains the optimized parameter configuration. The backtracking mechanism can help us summarize the lessons learned from the previous parameter adjustment process and find more effective adjustment directions.

[0112] Improved dataset generation and final training: Based on the optimized parameter configuration, we generate an improved virtual defect dataset, obtain data content that is more relevant to the actual scenario, and determine the final training input. By continuously optimizing the parameters for generating the virtual defect dataset, we can generate a dataset that is more consistent with the actual application scenario, thereby improving the performance and reliability of the model in actual use.

[0113] Iterative training and final model output: The defect recognition model is iteratively trained using the final training input, and classification is performed using the support vector machine algorithm to obtain the final model output. The iterative training process continuously optimizes the model's parameters and performance. The excellent performance of the support vector machine algorithm in classification tasks helps improve defect recognition accuracy. The final model output is used in actual defect detection and assessment work.

[0114] Based on the feature extraction and analysis results of the detection data, the above method dynamically adjusts the generation parameters of the virtual defect data set, so that the training data can better fit the actual needs and model training requirements, thereby continuously improving the performance of the defect recognition model. This process fully reflects the idea of ​​data-driven optimization. The introduction of a retrospective adjustment mechanism allows timely analysis of the cause and adjustment of the generation parameters when the model accuracy does not meet expectations, avoiding the blind repetition of training and data collection processes, improving the efficiency and pertinence of model optimization, and ensuring that resources can be effectively utilized in key optimization links. Through continuous iterations, from initial model construction to final model output, each step is dedicated to solving current problems and gradually improving the recognition accuracy of the model. This gradual optimization method helps to achieve a stable improvement in model performance and ultimately achieve satisfactory recognition results.

[0115] In some embodiments, the updated defect dataset, optimized inspection parameters, and identification models are integrated to build customized nondestructive testing standards and assessment frameworks, including:

[0116] By obtaining the latest defect data set from the repository, the initial data is sorted and classified to obtain a structured defect data set;

[0117] Based on the structured defect data set, a preset feature extraction method is used to analyze the data distribution characteristics and determine the adjustment range of key detection parameters;

[0118] If the adjustment range exceeds the preset threshold, the abnormal data points are eliminated through the information filtering link, the parameter range is recalculated, and the optimized detection parameter combination is obtained;

[0119] For the optimized detection parameter combination, the pre-trained convolutional neural network model is called, the structured defect data set is input, and the preliminary defect recognition results are output;

[0120] Based on the preliminary defect recognition results, the recognition error distribution is analyzed and the internal weights of the model are adjusted to obtain an updated recognition model version;

[0121] Generate customized draft NDT specifications using the updated recognition model version and optimized test parameter combinations to determine their applicability.

[0122] If the applicability of the draft specification passes the test in the information verification phase, all processing results will be integrated to build the final evaluation system framework and output complete testing standards.

[0123] In this technical solution, the method aims to integrate updated defect datasets, optimize detection parameters and identification models, and construct customized nondestructive testing standards and evaluation frameworks that meet practical application needs. Through a series of rigorous steps, including data acquisition and organization, parameter optimization, model adjustment, and draft specification generation and verification, the finalized testing standards are guaranteed to be both accurate and reliable, with good applicability, effectively guiding nondestructive testing practices, improving detection efficiency and quality, and ensuring the safety and reliability of engineering machinery components.

[0124] Initial Data Collation and Classification: Retrieve the latest defect dataset from the repository and complete the initial data collation and classification to obtain a structured defect data set. This step is the foundation of the entire process. The latest defect dataset contains defect data that has been optimized and updated through all previous steps. Collation and classification make it more organized and usable, facilitating subsequent analysis and processing. This structured defect data set will provide high-quality data support for the construction of customized nondestructive testing standards and assessment frameworks.

[0125] Data distribution characteristic analysis and parameter adjustment range determination: Based on the structured defect data set, a preset feature extraction method is used to analyze the data distribution characteristics and determine the adjustment range of key detection parameters. Feature extraction can uncover key information in the data, while data distribution characteristic analysis can help us understand the distribution patterns and characteristics of defect data across different feature dimensions. Based on these analysis results, the appropriate adjustment range of key detection parameters can be determined, providing a basis for subsequent optimization of detection parameters.

[0126] Abnormal data point removal and parameter range optimization: If the adjustment range exceeds the preset threshold, the system removes the abnormal data points through information filtering, recalculates the parameter range, and obtains the optimized detection parameter combination. The preset threshold is set to ensure that the adjustment range of the detection parameters is within a reasonable and controllable range. When the threshold is exceeded, some abnormal data points may have a significant impact on the parameter range. By removing these abnormal data points and recalculating the parameter range, a more stable and reliable optimized detection parameter combination can be obtained, improving the accuracy and stability of detection.

[0127] Generating preliminary defect identification results: Based on the optimized inspection parameter combination, the pre-trained convolutional neural network model is invoked, fed with a structured defect data set, and outputs preliminary defect identification results. Leveraging its powerful feature learning and classification capabilities, the convolutional neural network model effectively processes and analyzes the structured defect data set, generating preliminary defect identification results. This step verifies the effectiveness of the optimized inspection parameter combination and pre-trained model on the defect identification task, providing a foundation for subsequent model adjustments.

[0128] Adjusting and updating model internal weights: Based on the initial defect identification results, we analyze the recognition error distribution and adjust the model's internal weights to produce an updated recognition model. Recognition error distribution analysis helps us understand where the model has deficiencies and deviations. By adjusting the model's internal weights, we can effectively optimize model performance, reduce recognition errors, and improve the model's recognition accuracy and generalization capabilities.

[0129] Customized NDT draft specification generation and applicability assessment: Using the updated recognition model version and optimized test parameter combinations, a customized NDT draft specification is generated and its applicability assessed. This step transforms the optimized technical achievements into a practical test draft specification, clarifying the test process, methods, parameter settings, and defect identification standards and requirements. Assessing the applicability of the draft specification verifies whether it meets the needs of actual NDT work and demonstrates its feasibility and effectiveness.

[0130] Final Assessment System Framework Construction and Test Standard Output: If the draft specification passes the information verification phase, all processing results are integrated to construct the final assessment system framework and output the complete test standard. The information verification phase conducts comprehensive and rigorous testing and verification of the draft specification's applicability. Only after the draft specification passes the testing can it be integrated into the final assessment system framework. The output of the complete test standard will provide detailed guidance and specifications for nondestructive testing practices, ensuring standardized, scientific, and standardized testing work.

[0131] The entire process of the above method covers the complete process from data collation to the output of the final inspection standard. The various steps are closely connected and interrelated, forming a systematic customized non-destructive testing standard and evaluation framework construction process, which can comprehensively consider various influencing factors and ensure the quality and reliability of the final inspection standard. Based on the structured defect data set, through data analysis, parameter optimization, model adjustment and other steps, the inspection parameters and recognition model are continuously optimized, and the applicability of the draft specification is verified, which fully reflects the data-driven optimization idea, so that the final inspection standard can better meet the actual application needs and have high accuracy and reliability. The application of the convolutional neural network model provides strong technical support for defect recognition. Its excellent feature learning and classification capabilities can improve the accuracy and efficiency of defect recognition. At the same time, by analyzing the distribution of recognition errors to adjust the internal weights of the model, the performance of the model is further improved, reflecting the application advantages of intelligent technology in the field of non-destructive testing.

[0132] According to another aspect of the present invention, a non-destructive testing device for a foreign component of an engineering machinery is provided, wherein the device is based on the above-mentioned method; the device comprises:

[0133] The acquisition module is used to obtain the material properties and historical failure data of heterogeneous components, extract defect morphology and location characteristic parameters, build a statistical distribution model and generate a typical feature set;

[0134] A simulation module, configured to generate a virtual defect data set including morphological and positional features using numerical simulation based on the statistical distribution model and material properties;

[0135] An indicator module is used to perform finite element analysis on the virtual defect data set, calculate stress concentration effects and expansion trends, and output a quantitative indicator of defect hazard;

[0136] A first determination module is configured to mark a defect as high-risk and generate a priority simulation target set if the hazard index exceeds a preset threshold;

[0137] a defect generation module, configured to generate artificial defects matching the virtual defects in a test sample using micromachining technology based on the high-risk defect set;

[0138] The second determination module is used to perform multimodal nondestructive testing on the test sample containing artificial defects. If the matching degree between the signal characteristics and the preset feature library is lower than the threshold, the detection parameters are adjusted and the data is re-collected;

[0139] a third determination module, configured to train a defect recognition model based on the optimized detection data, and retroactively adjust generation parameters of the virtual defect dataset if the model accuracy does not reach a threshold;

[0140] A building block is used to integrate updated defect datasets, optimize detection parameters and recognition models, and build customized nondestructive testing standards and evaluation frameworks.

[0141] In the above technical solution, in order to better use the above method, this application proposes a non-destructive testing device for different components of engineering machinery. Each module corresponds to each step of the above method. The specific principles have been described above and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0142] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0143] Figure 1 This is a flow chart of an embodiment of a nondestructive testing method for a foreign component of an engineering machinery according to the present invention;

[0144] Figure 2 It is a structural schematic diagram of an embodiment of a non-destructive testing device for a different component of an engineering machinery according to the present invention. DETAILED DESCRIPTION

[0145] The present invention will be described in further detail below with reference to the accompanying drawings and examples. It is particularly noted that the following examples are intended only to illustrate the present invention and are not intended to limit the scope of the present invention. Similarly, the following examples are only some embodiments of the present invention and are not intended to be exhaustive. All other embodiments obtained by those of ordinary skill in the art without creative effort are intended to fall within the scope of protection of the present invention.

[0146] Example 1

[0147] See also Figure 1 A nondestructive testing method for a foreign component of an engineering machinery, the method comprising:

[0148] S1. Obtain material properties and historical failure data of heterogeneous components, extract defect morphology and location characteristic parameters, build a statistical distribution model and generate a typical feature set;

[0149] In this embodiment, S1. Obtain the material properties and historical failure data of the heterogeneous components, extract the defect morphology and position characteristic parameters, construct a statistical distribution model and generate a typical feature set, including: S11. Construct an initial data set by obtaining the material properties and historical failure records of the heterogeneous components from a database; S12. Preprocess the initial data set and extract relevant parameters of the defect morphology and position characteristics to determine the characteristic parameter data set; S13. Apply the probability density function to the characteristic parameter data set to construct a statistical distribution model and judge the distribution characteristics; if the distribution characteristics meet the preset threshold range, generate a typical feature set based on the statistical distribution model; if not, return to the previous step to re-extract the feature parameters; S14. Perform feature clustering analysis based on the typical feature set, combined with the position characteristics and defect morphology, and use the K-means clustering algorithm to divide the feature categories to obtain classification results; S15. Based on the classification results, perform correlation analysis on each type of feature to determine the mapping relationship between the feature set and the failure record.

[0150] For example, when processing the material properties and historical failure data of a heterogeneous component, the system first automatically extracts relevant information through a database system. For example, attribute data such as a component's elastic modulus of 200 GPa and yield strength of 350 MPa are obtained from the material database. At the same time, failure cases are extracted from historical failure records, such as a component that suffered fatigue fracture after 5000 hours of operation, with the fracture location located in a stress concentration area at a connection. Next, image processing technology combined with deep learning algorithms is used to extract characteristic parameters of the defect morphology and location. Specifically, a convolutional neural network (CNN) is used to analyze the component surface defect image, identifying a crack length of 2.5 mm, a width of 0.3 mm, and a location coordinate of (x=10.2, y=5.7) cm. These parameters are then stored as structured data. Subsequently, a statistical distribution model was constructed. Based on the extracted defect data, assuming that the crack length conforms to the normal distribution, parameter estimation was performed using the SciPy library in Python. A distribution model with a mean of 2.0 mm and a standard deviation of 0.5 mm was calculated, and the goodness of fit of the model was verified (for example, through the Kolmogorov-Smirnov test, the p-value was 0.85, indicating that the model was credible). Finally, a typical feature set was generated. Based on the distribution model, 1,000 sets of defect feature data were generated using the Monte Carlo simulation method. Representative features were screened out, such as a feature subset with crack lengths ranging from 1.5 to 2.5 mm and locations concentrated in stress concentration areas, forming a typical data set that can be used for subsequent failure prediction.

[0151] S2. generating a virtual defect data set including morphological and positional features using numerical simulation based on the statistical distribution model and material properties;

[0152] In this embodiment, S2, based on the statistical distribution model and material properties, a virtual defect data set containing morphological and positional features is generated using numerical simulation, including: S21, using the statistical distribution and material properties, a pre-established model construction method is used to obtain initial defect distribution parameters and obtain a preliminary distribution framework; S22, based on the preliminary distribution framework, combined with numerical simulation technology, simulation calculations are performed on the morphological features and positional features to determine the geometric shape and spatial coordinates of the virtual defects; S23, the distribution of the virtual defects is iteratively optimized using simulation technology. If the calculation result does not meet the preset threshold of the material property, the defect distribution parameters are adjusted, the simulation data is re-acquired, and it is determined whether the distribution consistency is satisfied; S 24. Through the adjusted simulation data, for defect distribution and feature extraction, a virtual data set containing multiple morphological features is constructed to obtain a diverse set of defect samples; S25. Based on the diverse set of defect samples, combined with position features and attribute analysis, the random forest algorithm is used to classify the data set to determine the spatial distribution patterns of defects of different categories; S26. Based on the spatial distribution patterns after classification, the final virtual defect data set is generated for virtual defects and data sets to obtain a complete description of defect features; S27. If there is a deviation between the feature description of the final data set and the preset statistical distribution, the simulation parameters of the morphological features and position features are readjusted through simulation technology to determine whether the consistency requirements are met.

[0153] For example, the implementation method of using numerical simulation to generate a virtual defect data set containing morphological and positional characteristics based on a statistical distribution model and material properties can be implemented through the following logical chain. First, for the construction of the statistical distribution model, it is assumed that the size of the material defect follows a normal distribution with a mean of 5.0 mm and a standard deviation of 1.5 mm. The Monte Carlo method is used to generate 1,000 defect size data points. Random sampling is achieved by calling the random.normal function in the numpy library in Python. After the data set is generated, statistical analysis is performed, and the actual mean and standard deviation are calculated to verify the distribution characteristics. For example, the obtained mean is 5.02 mm and the standard deviation is 1.48 mm, which verifies that the results are in line with expectations. Secondly, considering the material properties, assuming the material is aluminum alloy, the defect morphology is primarily elliptical, with a major-to-minor-axis ratio of 2:1. Based on this property, a geometric algorithm is used to generate the defect morphology. Specifically, the size data point is used as the major axis, and the minor axis is 0.5 times the major axis. The area of ​​each defect is calculated. For example, when the major axis is 5.0 mm and the minor axis is 2.5 mm, the area is approximately 9.82 square millimeters. The morphological data is then stored as a two-dimensional coordinate point set. Next, regarding the defect location characteristics, assuming the material is a 100 mm x 100 mm plane and the location distribution follows a uniform distribution, the numpy.random.uniform function is used to generate 1000 defect center point coordinates, ensuring that the locations are random and non-overlapping. A distance check algorithm is used to regenerate the coordinates if the distance between two defect center points is less than 3.0 mm, ultimately resulting in a location dataset. Finally, numerical simulation technology is used to integrate the morphological and positional data. Through the API interface of finite element analysis software such as ANSYS, a virtual defect model is automatically generated. The material parameters are set to the elastic modulus of aluminum alloy 70GPa and Poisson's ratio 0.33. A virtual defect dataset containing morphological and positional characteristics is generated and output as a CSV format file, containing information such as defect number, size, morphological parameters, and position coordinates. The defect distribution map is drawn using data visualization tools such as Matplotlib to analyze whether the defect-dense areas meet expectations.

[0154] S3. Perform finite element analysis on the virtual defect data set, calculate stress concentration effects and expansion trends, and output a quantitative indicator of defect hazard;

[0155] In this embodiment, S3, finite element analysis is performed on the virtual defect data set, stress concentration effect and expansion trend are calculated, and a defect hazard quantitative index is output, including: S31, initial defect feature data is obtained by processing the virtual defect data set, and a calculation model is constructed using a finite element analysis method to obtain the stress distribution status of the defect area; S32, stress concentration effect is calculated based on the stress distribution status, local grid encryption is performed on the high stress area, and the key position of stress concentration is determined; for the key position of stress concentration, the defect expansion trend is simulated, and finite element iterative calculation is performed to obtain expansion path and speed change data; from the expansion path and speed change data, the expansion trend characteristics are extracted, and the potential risk level of the defect expansion is judged in combination with the preset threshold range; S33, if the potential risk level of the defect expansion exceeds the preset threshold, the expansion trend characteristics are weighted analyzed to obtain a preliminary value of the hazard quantitative index; S34, the preliminary value of the hazard quantitative index is calibrated in multiple dimensions, and the stress distribution and expansion trend data are integrated to determine the final defect hazard quantitative index.

[0156] For example, in a finite element analysis of a virtual defect dataset, the defect morphology was first simulated by constructing a three-dimensional model. The defect was assumed to be an elliptical crack 10 mm long, 2 mm wide, and 1 mm deep, located in the center of a steel plate measuring 100 mm × 50 mm × 5 mm. A mesh was generated using finite element software such as ANSYS. The mesh density was set to 0.1 mm near the defect to improve computational accuracy, and 1 mm in other areas to ensure a balance between computational efficiency and accuracy. Next, a stress concentration effect analysis was performed, applying a uniform tensile load of 100 MPa to both ends of the steel plate. By solving the linear elastic equation, the maximum stress at the defect tip was obtained to be 300 MPa, and the calculated stress concentration factor was 3.0, indicating that the defect significantly amplified local stress. Next, the crack growth trend was analyzed, and the crack growth rate was calculated using the Paris law (da / dN = C(ΔK)^m, where C = 2.0e-10 and m = 3). Assuming a stress intensity factor (ΔK) range of 20 MPa·m^0.5, the crack growth per cycle, da / dN, was estimated to be approximately 0.016 mm. The crack was predicted to reach 15 mm after 5000 cycles, threatening structural integrity. Finally, the defect criticality was quantified by integrating the stress concentration factor (SCF) and the crack growth rate. The criticality index was defined as the product of the SCF and the growth rate, which is 3.0 × 0.016 = 0.048. A threshold of 0.05 was set. Values ​​below the threshold indicate low criticality, but values ​​approaching the threshold warrant attention. This analysis forms a complete logical chain from modeling to criticality assessment. If data is insufficient, a material fatigue life database can be used to estimate the remaining life, combining defect size with loading conditions. This further aids decision-making and ensures the rigor and practicality of the analysis.

[0157] S4. If the hazard index exceeds a preset threshold, it is marked as a high-risk defect and a priority simulation target set is generated;

[0158] In this embodiment, S4, if the hazard index exceeds the preset threshold, it is marked as a high-risk defect and a priority simulation target set is generated, including: S41, by obtaining the hazard index data from the system, performing preliminary cleaning and formatting processing on it, and obtaining a normalized indicator data set; S42, if the value in the normalized indicator data set exceeds the preset threshold, it is marked as a potential high-risk defect, and a corresponding risk determination result is generated; S43, according to the risk determination result, a pre-established classification model is used to classify the potential high-risk defects and determine the specific defect category information; S44, for the determined defect category information, a priority evaluation mechanism is constructed, and a priority simulation defect subset is obtained by comparing the priority level data; S45, key features are extracted from the priority simulation defect subset, and the support vector machine algorithm is used to analyze the features to determine the direction of the applicable simulation strategy; S46, according to the simulation strategy direction obtained by analysis, a target set is constructed to generate the final priority simulation target set data; S47, by storing and indexing the priority simulation target set data, a structured data set for subsequent call is generated.

[0159] For example, hazardous indicator data is obtained from a monitoring system for construction machinery. This data may include indicators such as stress, strain, temperature, and vibration frequency of a component under different operating conditions. Due to the diverse data sources, missing values, outliers, and inconsistent formats may exist. Therefore, preliminary cleaning is required to remove obvious errors and missing data points. The data in different formats is then converted to a standardized format, such as standardizing stress data to megapascals (MPa) and strain data to microstrain units. This ultimately results in a standardized indicator dataset. Assuming a preset stress threshold of 250 MPa, if the stress value of a component in the standardized indicator dataset reaches 260 MPa under a specific operating condition, exceeding the preset threshold, it is marked as a potential high-risk defect. A corresponding risk assessment result is generated, recording detailed information such as the component's number, location, and the name and value of the indicator that exceeded the threshold. The pre-established classification model is trained on a large amount of known defect sample data. This sample covers various common defect types found in construction machinery components, such as cracks, wear, and deformation. Each defect type has corresponding characteristic parameters. Data marked as potentially high-risk defects is input into the classification model. The model matches and classifies the data based on similarities with known defect characteristics, determining the specific category of the potential high-risk defect. For example, it identifies it as a crack defect and records detailed characteristics such as crack length and direction. A priority assessment mechanism is established for each identified crack defect category. This mechanism prioritizes data based on factors such as crack length, location, growth rate, and the criticality of the engineering machinery component in which it is located. By comparing the priority data, a subset of defects is selected for priority simulation. For example, crack defects located in critical stress-bearing locations with longer lengths and faster growth rates are assigned higher priority and included in the priority defect subset. Key features are extracted from this priority crack defect subset, such as crack shape, size, and stress distribution. These features are analyzed using a support vector machine algorithm to determine the appropriate simulation strategy direction. For example, if the analysis reveals that the crack shape and stress distribution characteristics conform to a specific fatigue crack growth pattern, a simulation strategy based on fatigue analysis is appropriate. Based on the simulation strategy direction (i.e., the fatigue analysis direction) derived from this analysis, a target set is constructed. Collect various parameters and models related to fatigue analysis, such as material fatigue properties, crack growth models, and stress-strain cyclic relationships. This information is then integrated to generate a final set of prioritized simulation target data, including detailed information such as the required input parameters, boundary conditions, loading conditions, and expected simulation output results. The generated prioritized simulation target data is stored in a database or data file format, and corresponding indexes are established.

[0160] S5. Based on the high-risk defect set, generate artificial defects matching the virtual defects in the test sample using micromachining technology;

[0161] In this embodiment, S5, based on the high-risk defect set, uses micromachining technology to generate artificial defects that match the virtual defects in the test sample, including: S51, constructing a digital model of the virtual defect by extracting feature data from the high-risk defect set; S52, using micromachining technology to perform preliminary surface treatment on the test sample to obtain a sample matrix suitable for processing; S53, based on the digital model of the virtual defect, generating processing instructions that match the high-risk defect and determining the processing path; S54, if the processing instruction does not meet the preset accuracy threshold, the instruction parameters are adjusted through the information processing link to obtain an optimized processing plan; S55, based on the optimized processing plan, use micromachining technology to generate artificial defects on the test sample, and determine whether the generated results are consistent; S56, if the deviation between the generated artificial defect and the virtual defect feature exceeds the preset range, the source of the deviation is analyzed through the information processing link to determine the correction parameters; S57, use the correction parameters to calibrate the micromachining technology, regenerate artificial defects on the test sample, and obtain the final matching result.

[0162] For example, a specific implementation method for using micromachining technology to generate artificial defects in test samples that match virtual defects based on a set of high-risk defects can be fully automated through information technology. First, the system extracts data from a defect database for a set of high-risk defects. Assuming the set contains 100 defect types, the characteristic parameters of each defect include size (average length 5.2 microns, width 2.1 microns), depth (average 1.8 microns), and shape complexity (quantized value 0.75). These parameters are classified and modeled using a feature extraction algorithm (such as a deep learning-based convolutional neural network (CNN)). A matching score is calculated for each defect, and defects with a score above 0.85 are selected as priority processing targets, accounting for approximately 30% of the total number of defects. The system then inputs this selected virtual defect data into micromachining control software, automatically generating a machining path plan. Laser micromachining technology is used, with a laser power of 50 milliwatts and a pulse frequency of 10 kHz. The machining accuracy is controlled within 0.1 micron, ensuring that the dimensional error between the artificial and virtual defects is less than 5%. After processing is complete, the system automatically captures image data of the test sample using a high-resolution scanning electron microscope (SEM) with a resolution of 0.05 microns per pixel. Using image processing algorithms (such as the Sobel operator for edge detection), the system analyzes the actual parameters of the artificial defect, compares them with the virtual defect parameters, and calculates the deviation. If the deviation exceeds 3%, the system automatically adjusts the laser processing parameters (for example, increasing the power to 55 milliwatts) and reprocesses the part. Finally, the system stores all processing and analysis data in a cloud database, generating a defect report.

[0163] S6. Perform multimodal nondestructive testing on the test sample containing artificial defects. If the matching degree between the signal characteristics and the preset feature library is lower than the threshold, adjust the detection parameters and re-collect data;

[0164] In this embodiment, S6, multimodal nondestructive testing is performed on the test sample containing artificial defects. If the matching degree between the signal feature and the preset feature library is lower than the threshold, the detection parameters are adjusted and the data is re-collected, including: S61, the test sample is initially scanned by the multimodal detection technology to obtain a first set of signal feature data; S62, if the matching degree between the first set of signal feature data and the preset feature library is lower than the matching degree threshold, the parameter adjustment mechanism is triggered to generate adjusted detection parameters, and data is re-collected based on the adjusted parameters to obtain a second set of signal feature data; S63, the support vector machine algorithm is used to perform feature classification processing on the second set of signal feature data to determine whether there is any Abnormal signal patterns related to artificial defects; S64. Based on the classification processing results, if an abnormal signal pattern is detected, the corresponding defect feature template is extracted from the preset feature library to determine the specific defect type of the abnormal signal; S65. After obtaining the determined defect type, the test sample is locally enhanced scanned in combination with multiple signal sources of multimodal detection to obtain a third set of high-resolution signal feature data; S66. Through in-depth analysis of the third set of signal feature data, a convolutional neural network model is used to accurately locate the defect position and determine the distribution of defects in the test sample; S67. Based on the defect distribution, the corresponding defect recognition result is generated to complete a comprehensive evaluation of the test sample.

[0165] For example, in the specific implementation of multimodal nondestructive testing on a test sample containing an artificial defect, data is first collected from the sample using both ultrasonic and infrared thermal imaging modalities. Assuming the sample is a 10mm thick steel plate and the artificial defect is a circular hole with a diameter of 5mm, ultrasonic testing uses a 5MHz probe with a scanning step size of 0.5mm and a time window of 0.1ms for collecting echo signals. Simultaneously, infrared thermal imaging uses a camera with a resolution of 640x480, a heating power of 500W, and a heating time of 10s to record temperature change data. Next, signal features are extracted. The ultrasonic signal uses a fast Fourier transform (FFT) algorithm to calculate the dominant frequency component, resulting in a dominant frequency value of 4.8MHz. The infrared thermal imaging data uses a differential algorithm to extract the temperature gradient in the defect area, with a calculated maximum gradient of 2.5°C / mm. The extracted features are then matched against a pre-set feature library, which contains ultrasonic dominant frequencies ranging from 4.7 to 5.3 MHz and temperature gradients ranging from 2.0 to 3.0°C / mm. A Euclidean distance algorithm is used to calculate the matching degree, with a threshold of 0.8. The current matching degree is 0.6, which is below the threshold and triggers the parameter adjustment logic. The system automatically adjusts the detection parameters, increasing the ultrasonic frequency to 5.2 MHz, reducing the scanning step size to 0.3 mm, increasing the infrared heating power to 600 W, and extending the heating time to 12 seconds. The adjusted parameter combination is calculated using a pre-set optimization algorithm (based on gradient descent) to ensure a signal-to-noise ratio improvement of at least 10%. Data is then re-acquired, and the system automatically performs a new scan, recording the ultrasonic echo signal and infrared temperature field data. The feature extraction and matching process is repeated until the matching degree reaches 0.85, meeting the threshold requirement.

[0166] S7. Training a defect recognition model based on the optimized detection data. If the model accuracy does not reach a threshold, retroactively adjusting the generation parameters of the virtual defect dataset;

[0167] In this embodiment, S7, the defect recognition model is trained according to the optimized detection data. If the model accuracy does not reach the threshold, the generation parameters of the virtual defect data set are retroactively adjusted, including: S71, by processing the detection data, an initial defect recognition model is constructed, preliminary results in the training process are obtained, and whether the output of the model meets the expected standards; S72, if the recognition accuracy of the preliminary results does not reach the preset threshold, feature extraction is performed on the detection data, key influencing factors are determined, and the analysis results of the feature distribution are obtained; S73, according to the analysis results of the feature distribution, the generation parameters of the virtual defect data set are adjusted, the updated data set content is obtained, and the new training basis; S74, using the updated data set content, re-execute the training process of the defect recognition model, obtain a new recognition accuracy value, and determine whether it meets the preset standard; S75, if the new recognition accuracy value still does not reach the preset threshold, the adjustment direction of the generated parameters is analyzed through the backtracking mechanism to obtain the optimized parameter configuration; S76, based on the optimized parameter configuration, generate an improved virtual defect data set, obtain data content that is more in line with the actual scenario, and determine the final training input; S77, run the iterative training of the defect recognition model through the final training input, use the support vector machine algorithm for classification processing, and obtain the final model output result.

[0168] Exemplary implementation methods for training a defect recognition model based on optimized inspection data and retroactively adjusting the parameters for generating a virtual defect dataset based on whether the model accuracy reaches a threshold are as follows. First, a training set is constructed using the optimized inspection data. Assume that the data contains 10,000 images of industrial parts, including 3,000 defective samples and 7,000 normal samples, with an image resolution of 1024x1024 pixels. A convolutional neural network (CNN) model, ResNet-50, used in deep learning, is used for training. Training parameters are set to a learning rate of 0.001, a batch size of 32, and 50 iterations. The model is optimized using a cross-entropy loss function. During training, 80% of the data is used as the training set and 20% as the validation set. The accuracy on the validation set is calculated. Assuming a target accuracy threshold of 0.95, if the accuracy of the trained model on the validation set is 0.92, which does not reach the threshold, the retroactive mechanism is triggered. Next, the system automatically analyzed the model prediction results and found that the recognition rate for some tiny cracks was only 0.75. This may be because the crack feature distribution in the virtual defect dataset was not diverse enough. Therefore, the generation parameters of the virtual defect dataset were retroactively adjusted. The specific method was to call the generation algorithm, increase the generation ratio of crack defects from 20% to 30%, and adjust the crack width parameter from an average of 2.5 pixels to 3.5 pixels. The length distribution range was expanded from 10-50 pixels to 10-80 pixels. Random noise was introduced to enhance data diversity, and 5,000 virtual defect images were regenerated. Subsequently, the newly generated virtual dataset was merged with the original inspection data, the training process was repeated, and the model accuracy was re-evaluated. If it still did not meet the standard, the defect category misjudgment rate was further analyzed and the parameters were iteratively adjusted until the accuracy reached above 0.95.

[0169] S8. Integrate the updated defect data set, optimize the detection parameters and recognition model, and build a customized non-destructive testing standard and evaluation framework.

[0170] In this embodiment, S8, integrating the updated defect data set, optimizing the detection parameters and recognition model, and constructing a customized non-destructive testing standard and evaluation framework, includes: S81, obtaining the latest defect data set from the repository, completing the organization and classification of the initial data, and obtaining a structured defect data set; S82, based on the structured defect data set, using a preset feature extraction method, analyzing the data distribution characteristics, and determining the adjustment range of key detection parameters; S83, if the adjustment range exceeds the preset threshold, eliminating abnormal data points through the information filtering link, recalculating the parameter range, and obtaining the optimized detection parameter combination; S84, For the optimized detection parameter combination, call the pre-trained convolutional neural network model, input the structured defect data set, and output the preliminary defect recognition result; S85. According to the preliminary defect recognition result, analyze the recognition error distribution, adjust the internal weight of the model, and obtain the updated recognition model version; S86. Through the updated recognition model version, combined with the optimized detection parameter combination, generate a customized non-destructive testing specification draft, and judge the applicability of the specification draft; S87. If the applicability of the specification draft passes the test of the information verification link, integrate all processing results, build the final evaluation system framework, and output a complete detection standard.

[0171] For example, in the process of building a customized nondestructive testing standard and assessment framework, the updated defect dataset is first integrated through information technology. Assuming that the dataset contains 100,000 ultrasonic inspection images, of which cracks and pores account for 60% and 40% respectively, data preprocessing algorithms such as image enhancement and normalization are used to uniformly adjust the image resolution to 512x512 pixels. Data enhancement techniques (such as rotation and flipping) are then used to expand the dataset to 150,000 images to improve the diversity of model training. Next, when optimizing the inspection parameters, a genetic algorithm is used to automatically adjust the sensitivity parameters of the ultrasonic inspection equipment. The initial sensitivity is set to 80dB, and after 100 iterations, the optimal sensitivity is found to be 85.5dB. At the same time, the inspection frequency range is expanded from 1MHz to 5MHz. By analyzing the signal-to-noise ratio (SNR) of the defect echo signal, it is increased to 15.2, ensuring that the inspection accuracy is improved by approximately 10%. Subsequently, in the optimization of the recognition model, an improved convolutional neural network (CNN) architecture was selected, with the network layer number set to 18 and a training set to test set ratio of 8:2. The Adam optimizer was used with an initial learning rate of 0.001. After 50 epochs of training, the model achieved a defect recognition accuracy of 92.3% on the test set, a 5 percentage point improvement over the original model. Analysis showed that the model's crack recognition rate increased from 85% to 91%. Finally, when constructing a customized standard and evaluation framework, based on the above data and model results, a defect size threshold standard of 0.5mm was established as the minimum detectable crack length. An automated evaluation script was used to calculate the detection system's false alarm rate (3.2%) and false negative rate (2.8%). Based on business needs, the evaluation results were compared with industry standards (such as ISO 5817) to automatically generate inspection reports that meet specific industry requirements and ensure traceability of the inspection process.

[0172] Example 2

[0173] See also Figure 2 A non-destructive testing device for a different component of an engineering machinery, the device is based on the method described in one embodiment; the device comprises:

[0174] The acquisition module is used to obtain the material properties and historical failure data of heterogeneous components, extract defect morphology and location characteristic parameters, build a statistical distribution model and generate a typical feature set;

[0175] A simulation module, configured to generate a virtual defect data set including morphological and positional features using numerical simulation based on the statistical distribution model and material properties;

[0176] An indicator module is used to perform finite element analysis on the virtual defect data set, calculate stress concentration effects and expansion trends, and output a quantitative indicator of defect hazard;

[0177] A first determination module is configured to mark a defect as high-risk and generate a priority simulation target set if the hazard index exceeds a preset threshold;

[0178] a defect generation module, configured to generate artificial defects matching the virtual defects in a test sample using micromachining technology based on the high-risk defect set;

[0179] The second determination module is used to perform multimodal nondestructive testing on the test sample containing artificial defects. If the matching degree between the signal characteristics and the preset feature library is lower than the threshold, the detection parameters are adjusted and the data is re-collected;

[0180] a third determination module, configured to train a defect recognition model based on the optimized detection data, and retroactively adjust generation parameters of the virtual defect dataset if the model accuracy does not reach a threshold;

[0181] A building block is used to integrate updated defect datasets, optimize detection parameters and recognition models, and build customized nondestructive testing standards and evaluation frameworks.

[0182] In the above technical solution, in order to better utilize the method described in one of the embodiments, the present application proposes a non-destructive testing device for different components of engineering machinery. Each module corresponds to each step of the above method. The specific principles have been described above and will not be repeated here.

[0183] The above descriptions are only some embodiments of the present invention and do not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A non-destructive testing method for a foreign component of an engineering machinery, characterized in that: The method comprises: Obtain material properties and historical failure data of heterogeneous components, extract defect morphology and location characteristic parameters, build statistical distribution models and generate typical feature sets; generating a virtual defect data set including morphological and positional characteristics using numerical simulation according to the statistical distribution model and material properties; Performing finite element analysis on the virtual defect data set, calculating stress concentration effects and expansion trends, and outputting quantitative indicators of defect hazard; If the quantitative index of the defect's harmfulness exceeds a preset threshold, it is marked as a high-risk defect and a priority simulation target set is generated; Based on a set of high-risk defects, micromachining techniques are used to generate artificial defects in the test sample that match the virtual defects, including: By extracting characteristic data from a set of high-risk defects, a digital model of virtual defects is constructed; Micromachining technology is used to perform preliminary surface treatment on the test sample to obtain a sample matrix suitable for processing; Based on the digital model of virtual defects, generate processing instructions that match high-risk defects and determine the processing path; If the processing instruction does not meet the preset accuracy threshold, the instruction parameters are adjusted through the information processing link to obtain an optimized processing plan; According to the optimized processing plan, artificial defects are generated on the test sample using micromachining technology to determine whether the generated results are consistent; If the deviation between the generated artificial defect and the virtual defect characteristics exceeds the preset range, the source of the deviation is analyzed through the information processing link to determine the correction parameters; The micromachining technology is calibrated using the corrected parameters, and artificial defects are regenerated on the test sample to obtain the final matching results; Perform multimodal nondestructive testing on test samples containing artificial defects. If the matching degree between the signal characteristics and the preset feature library is lower than the threshold, adjust the detection parameters and re-collect data; Training a defect recognition model based on the optimized inspection data, and if the model accuracy does not reach a threshold, retroactively adjusting generation parameters of the virtual defect dataset; Integrate updated defect datasets, optimize detection parameters and recognition models, and build customized non-destructive testing standards and evaluation frameworks.

2. A non-destructive testing method for a foreign component of an engineering machinery as claimed in claim 1, characterized in that: Obtain material properties and historical failure data of heterogeneous components, extract defect morphology and location characteristic parameters, build statistical distribution models and generate typical feature sets, including: The initial data set is constructed by obtaining the material properties and historical failure records of heterogeneous components from the database; Based on the initial data set, preprocessing is performed and relevant parameters of defect morphology and location characteristics are extracted to determine the characteristic parameter data set; Based on the feature parameter data set, a probability density function is applied to construct a statistical distribution model to determine the distribution characteristics. If the distribution characteristics meet the preset threshold range, a typical feature set is generated based on the statistical distribution model. If not, the process returns to the previous step to re-extract the feature parameters. Based on the typical feature set, combined with the position features and defect morphology, feature clustering analysis is performed, and the K-means clustering algorithm is used to divide the feature categories to obtain the classification results; Based on the classification results, correlation analysis is performed on each type of features to determine the mapping relationship between the feature set and the failure record.

3. The non-destructive testing method for a foreign component of an engineering machinery according to claim 1, characterized in that: Based on the statistical distribution model and material properties, numerical simulation is used to generate a virtual defect dataset containing morphological and positional characteristics, including: Using the statistical distribution model and material properties, a pre-established model building method is used to obtain initial defect distribution parameters and a preliminary distribution framework; Based on the preliminary distribution framework and combined with numerical simulation technology, simulation calculations are performed on the morphological and positional characteristics to determine the geometric shape and spatial coordinates of the virtual defect; Use simulation technology to iteratively optimize the distribution of virtual defects. If the calculated results do not match the preset threshold of the material properties, adjust the defect distribution parameters, reacquire the simulation data, and determine whether the distribution consistency is met. Through the adjusted simulation data, a virtual data set containing various morphological features is constructed for defect distribution and feature extraction, thus obtaining a diverse set of defect samples. Based on a diverse set of defect samples, combined with location features and attribute analysis, the random forest algorithm is used to classify the data set and determine the spatial distribution patterns of different types of defects. Generate the final virtual defect dataset through the spatial distribution law after classification processing to obtain a complete defect feature description; If there is a deviation between the feature description of the final data set and the preset statistical distribution, the simulation parameters of the morphological and positional features are readjusted through simulation technology to determine whether the consistency requirements are met.

4. The nondestructive testing method for a foreign component of an engineering machinery according to claim 1, characterized in that: Finite element analysis is performed on the virtual defect data set to calculate the stress concentration effect and expansion trend, and output quantitative indicators of defect criticality, including: By processing the virtual defect data set, the initial defect characteristic data is obtained, and the calculation model is constructed using the finite element analysis method to obtain the stress distribution status of the defect area; Based on the stress distribution, the stress concentration effect is calculated, and local mesh encryption is performed on high-stress areas to determine the key locations of stress concentration. At these key locations of stress concentration, the defect expansion trend is simulated, and finite element iteration is used to obtain expansion path and velocity change data. From these expansion path and velocity change data, expansion trend characteristics are extracted and combined with a preset threshold range to determine the potential risk level of defect expansion. If the potential risk level of defect expansion exceeds the preset threshold, a weighted analysis is performed on the expansion trend characteristics to obtain a preliminary value of the quantitative indicator of harmfulness; The final defect criticality quantitative index is determined by performing multi-dimensional calibration on the preliminary value of the criticality quantitative index and integrating stress distribution and expansion trend data.

5. The non-destructive testing method for a foreign component of an engineering machinery according to claim 1, characterized in that: If the quantitative indicator of the defect's harmfulness exceeds a preset threshold, it is marked as a high-risk defect and a priority simulation target set is generated, including: By obtaining the quantitative indicator data of defect hazard from the system, it is preliminarily cleaned and formatted to obtain a standardized indicator data set; If the value in the normalized indicator data set exceeds the preset threshold, it will be marked as a potential high-risk defect and the corresponding risk determination result will be generated; Based on the risk assessment results, a pre-established classification model is used to classify potential high-risk defects and determine the specific defect category information; Based on the identified defect category information, a priority assessment mechanism is established to obtain a subset of defects for priority simulation by comparing priority data. Extract key features from the priority simulated defect subset, analyze the features using the support vector machine algorithm, and determine the direction of applicable simulation strategy; According to the simulation strategy direction obtained from the analysis, a target set is constructed to generate the final priority simulation target set data; By storing and indexing the priority simulation target set data, a structured data set that can be used for subsequent calls is generated.

6. The nondestructive testing method for a foreign component of an engineering machinery according to claim 1, characterized in that: Perform multimodal nondestructive testing on test samples containing artificial defects. If the matching degree between the signal characteristics and the preset feature library is lower than the threshold, adjust the detection parameters and re-collect data, including: Performing an initial scan on the test sample using multimodal detection technology to obtain a first set of signal characteristic data; If the matching degree between the first set of signal feature data and the preset feature library is lower than the matching degree threshold, the parameter adjustment mechanism is triggered to generate adjusted detection parameters, and data is collected again based on the adjusted parameters to obtain the second set of signal feature data; For the second set of signal feature data, the support vector machine algorithm is used to perform feature classification processing to determine whether there are abnormal signal patterns related to artificial defects; Based on the classification processing results, if an abnormal signal pattern is detected, the corresponding defect feature template is extracted from the preset feature library to determine the specific defect type of the abnormal signal; After determining the defect type, the test sample is scanned locally using a combination of multiple signal sources from multimodal testing to obtain a third set of high-resolution signal feature data. Through in-depth analysis of the third set of signal feature data, a convolutional neural network model is used to accurately locate the defect position and determine the distribution of defects in the test sample; According to the defect distribution, the corresponding defect identification results are generated to complete the comprehensive evaluation of the test samples.

7. The non-destructive testing method for a foreign component of an engineering machinery according to claim 1, characterized in that: The defect recognition model is trained based on the optimized detection data. If the model accuracy does not reach a threshold, the generation parameters of the virtual defect dataset are retroactively adjusted, including: By processing the inspection data, an initial defect recognition model is constructed, preliminary results are obtained during the training process, and it is determined whether the output of the model meets the expected standards; If the recognition accuracy of the preliminary results does not reach the preset threshold, feature extraction is performed on the detection data to determine the key influencing factors and obtain the analysis results of the feature distribution; Based on the analysis results of the feature distribution, the generation parameters of the virtual defect dataset are adjusted, the updated dataset content is obtained, and a new training basis is determined; Using the updated data set content, re-execute the defect recognition model training process to obtain a new recognition accuracy value and determine whether it meets the preset standards; If the new recognition accuracy value still does not reach the preset threshold, the adjustment direction of the generated parameters is analyzed through the backtracking mechanism to obtain the optimized parameter configuration; Generate an improved virtual defect dataset based on the optimized parameter configuration to obtain data content that is more in line with the actual scenario and determine the final training input; Through the final training input, the iterative training of the defect recognition model is run, and the support vector machine algorithm is used for classification processing to obtain the final model output result.

8. The non-destructive testing method for a foreign component of an engineering machinery according to claim 1, characterized in that: Integrate updated defect datasets, optimize inspection parameters and recognition models, and build customized non-destructive testing standards and assessment frameworks, including: By obtaining the latest defect data set from the repository, the initial data is sorted and classified to obtain a structured defect data set; Based on the structured defect data set, a preset feature extraction method is used to analyze the data distribution characteristics and determine the adjustment range of key detection parameters; If the adjustment range exceeds the preset threshold, the abnormal data points are eliminated through the information filtering link, the parameter range is recalculated, and the optimized detection parameter combination is obtained; For the optimized detection parameter combination, the pre-trained convolutional neural network model is called, the structured defect data set is input, and the preliminary defect recognition results are output; Based on the preliminary defect recognition results, the recognition error distribution is analyzed and the internal weights of the model are adjusted to obtain an updated recognition model version; Generate customized draft NDT specifications using the updated recognition model version and optimized test parameter combinations to determine their applicability. If the applicability of the draft specification passes the test in the information verification phase, all processing results will be integrated to build the final evaluation system framework and output complete testing standards.

9. A non-destructive testing device for different components of engineering machinery, characterized in that: The device is based on the method according to any one of claims 1 to 8; the device comprises: The acquisition module is used to obtain the material properties and historical failure data of the heterogeneous components, extract the defect morphology and location characteristic parameters, build a statistical distribution model and generate a typical feature set; A simulation module, configured to generate a virtual defect data set including morphological and positional features using numerical simulation based on the statistical distribution model and material properties; An indicator module is used to perform finite element analysis on the virtual defect data set, calculate stress concentration effects and expansion trends, and output a quantitative indicator of defect hazard; A first determination module is configured to mark a defect as a high-risk defect and generate a priority simulation target set if the defect hazard quantification index exceeds a preset threshold; a defect generation module, configured to generate artificial defects matching the virtual defects in the test sample using micromachining technology based on a set of high-risk defects; The second determination module is used to perform multimodal nondestructive testing on the test sample containing artificial defects. If the matching degree between the signal characteristics and the preset feature library is lower than the threshold, the detection parameters are adjusted and the data is re-collected; a third determination module, configured to train a defect recognition model based on the optimized detection data, and retroactively adjust generation parameters of the virtual defect dataset if the model accuracy does not reach a threshold; A building block is used to integrate updated defect datasets, optimize detection parameters and recognition models, and build customized nondestructive testing standards and evaluation frameworks.

Images