Subframe weld fatigue crack identification method and system

Abstract

The application provides a subframe weld fatigue crack identification method and system, which comprises the following steps: according to the target risk area determined according to the stress distribution cloud picture, the space position of the target risk area is used to dynamically adjust the interval, angle and excitation power of the preset excitation coil array to generate a directional eddy current excitation field focused on the target risk area; a dynamic baseline signal library is constructed, and the least squares support vector machine is used for adaptive calibration to output a target fusion signal; the feature parameters related to the crack in the target fusion signal and the subframe weld are decoupled through Bayesian inference, and the ultrasonic reflection signal corresponding to the target risk area is collected through an ultrasonic sensor; the crack reflection amplitude feature in the ultrasonic reflection signal is detected, and the crack reflection amplitude feature and the feature parameters are analyzed in correlation, so that the corresponding crack risk level is determined according to the analysis result. The application can effectively improve the crack identification efficiency.

Description

Technical Field

[0001] This invention relates to the field of automotive technology, and in particular to a method and system for identifying fatigue cracks in subframe welds. Background Technology

[0002] As the automotive industry moves towards lightweight design, aluminum alloys are widely used in subframe manufacturing due to their lightweight advantages. As a core load-bearing component, the subframe's weld seams are susceptible to fatigue cracks due to factors such as load cycles and residual welding stress. Failure to detect these cracks in a timely manner can lead to structural failure and safety accidents. Therefore, accurate and efficient identification and detection of fatigue cracks in subframe weld seams is a critical requirement for ensuring vehicle safety and has significant application value.

[0003] Non-destructive testing (NDT) is the mainstream method for inspecting subframe welds. Traditional eddy current testing (EDT) technology is widely used due to its advantages such as non-contact operation, fast response, and compatibility with conductive materials. Corresponding identification methods and systems are relatively mature. Its core principle is to induce eddy currents by generating an alternating magnetic field through an excitation coil, and then use the eddy current distortion signal caused by defects to identify and locate cracks. In practical applications, to simplify equipment and signal processing, this technology typically uses a finite number of discrete frequencies as excitation signals.

[0004] Furthermore, traditional array eddy current detection technology has significant drawbacks: limited by the eddy current skin effect, discrete frequency excitation can only cover the shallow layer of the material, failing to comprehensively cover different depths of the weld. For common characteristics of weld fatigue cracks, such as microscopic, deep, and irregular morphology, this technology struggles to capture clear eddy current distortion signals, easily leading to missed or false detections, and thus failing to meet high-precision detection requirements. Therefore, there is an urgent need to develop a novel method and system for identifying fatigue cracks in subframe welds to overcome existing technological bottlenecks. Summary of the Invention

[0005] Based on this, the purpose of the present invention is to provide a method and system for identifying fatigue cracks in subframe welds, so as to solve the problem that the existing technology cannot capture clear eddy current distortion signals, which leads to the easy occurrence of missed detections and false detections.

[0006] The first aspect of the present invention proposes: A method for identifying fatigue cracks in subframe welds, wherein the method includes: The stress distribution cloud map of the subframe weld under preset working conditions is collected by finite element simulation. The corresponding target risk area is determined according to the stress distribution cloud map. Simultaneously, the spacing, angle and excitation power of the preset excitation coil array are dynamically adjusted according to the spatial position of the target risk area to generate a directional eddy current excitation field focused on the target risk area. The microscopic eddy current distortion signal and macroscopic eddy current response signal generated by the subframe weld under the directional eddy current excitation field are collected to construct a corresponding dynamic baseline signal library. Simultaneously, the least squares support vector machine is used for adaptive calibration to output the corresponding target fusion signal. The characteristic parameters related to the cracks in the subframe weld are decoupled from the target fusion signal by Bayesian inference, and the ultrasonic reflection signals generated in the target risk area are collected simultaneously by an ultrasonic sensor. The crack reflection amplitude characteristics in the ultrasonic reflection signal are detected, and the crack reflection amplitude characteristics are simultaneously correlated with the characteristic parameters to determine the corresponding crack hazard level based on the analysis results.

[0007] The beneficial effects of this invention are as follows: This solution accurately locates the target risk area of ​​the subframe weld seam through finite element simulation, dynamically adjusts the excitation coil array parameters to generate a directional eddy current excitation field, and achieves focused signal acquisition; it constructs a dynamic baseline library by combining microscopic and macroscopic eddy current signals and performs adaptive calibration using least squares support vector machine, and combines Bayesian inference to decouple crack characteristic parameters with ultrasonic signal collaborative analysis. This not only solves the pain point of fuzzy eddy current distortion signal capture in existing technologies, but also significantly improves the sensitivity, accuracy, and anti-interference ability of crack detection through multi-dimensional signal fusion and cross-validation, greatly reduces the missed detection and false detection rates, and can accurately determine the crack hazard level, providing a reliable guarantee for the safety of the subframe structure.

[0008] Furthermore, the step of acquiring stress distribution cloud maps of the subframe welds under preset working conditions through finite element simulation, and determining the corresponding target risk areas based on the stress distribution cloud maps, includes: The residual stress distribution data of the subframe weld was detected by X-ray diffraction and simultaneously incorporated into the weld finite element model as the initial stress field. Based on the dynamic working condition sequence of the actual service of the subframe, the corresponding load was applied sequentially through the weld finite element model to generate a stress distribution cloud map covering the entire working condition cycle. Based on the stress distribution cloud map, the stress data of each weld unit in the subframe weld is time-series tracked to detect the evolution trajectory of the stress peak, and the slope and fluctuation coefficient of the evolution trajectory are calculated simultaneously to screen out stress-sensitive areas accordingly. The target risk area is determined based on the stress-sensitive area.

[0009] Furthermore, the step of determining the target risk area based on the stress-sensitive area includes: The type, size, and location of inherent welding defects in the stress-sensitive region are detected to create a coupling risk model between welding defects and stress. The corresponding coupling risk coefficient is calculated by combining time-series stress data, and the defect-stress coupling sensitive region is screened according to the magnitude of the coupling risk coefficient. The magnetic memory signal of the defect-stress coupling sensitive area is collected by a magnetic memory sensor, and the fatigue damage accumulation degree of the subframe weld is determined according to the magnetic memory signal to identify the corresponding risk incubation area. The influence weight of potential crack propagation in the risk incubation area relative to the overall load-bearing performance of the subframe is analyzed, and the crack propagation path and key node corresponding to the highest influence weight are detected. At the same time, the area covering the crack propagation path and the key node is set as the target risk area.

[0010] Furthermore, the step of decoupling the characteristic parameters related to the cracks in the subframe weld using Bayesian inference includes: The target fused signal is adaptively decomposed using an ensemble empirical mode decomposition algorithm to obtain several intrinsic mode functions. The instantaneous frequency, instantaneous amplitude, and kurtosis value of each intrinsic mode function are calculated simultaneously to create a corresponding multidimensional feature matrix. The multidimensional feature matrix is ​​converted into a corresponding decoupled feature set, and the decoupled feature set is simultaneously divided into a crack-sensitive feature layer and an interference feature layer to perform layered decoupling of crack feature components and crack interference components, and the corresponding initial crack feature parameters are output synchronously. The initial crack feature parameters are optimized and corrected to generate the corresponding feature parameters.

[0011] Furthermore, the step of optimizing and correcting the initial crack feature parameters to generate the corresponding feature parameters includes: The attention-enhanced convolutional neural network constructs a corresponding temperature compensation factor based on the actual temperature monitoring data of the weld, and performs feature enhancement processing on the initial crack feature parameters based on the temperature compensation factor, and outputs the corresponding crack feature vector simultaneously. A virtual simulation scenario is constructed based on the preset working conditions. The crack feature vector is simultaneously substituted into each of the virtual simulation scenarios for simulation verification. The fluctuation range of the feature parameters under each working condition is calculated by the interval estimation algorithm. If the fluctuation range is detected to meet the preset range requirements, the crack feature vector is converted into the corresponding target parameter, and the target parameter is simultaneously set as the feature parameter.

[0012] Furthermore, the step of performing correlation analysis between the crack reflection amplitude characteristics and the characteristic parameters to determine the corresponding crack hazard level based on the analysis results includes: The crack reflection amplitude feature is mapped to the feature parameter as a corresponding modal feature vector, and a sparse coding algorithm is simultaneously used to sparsely represent the modal feature vector to output the corresponding sparse feature coefficients. Based on the sparse characteristic coefficients, the mechanical parameters of the weld material are introduced as physical constraints. The Lagrange multiplier method is used to solve the feature correlation objective function that satisfies the constraints, so as to output the corresponding feature correlation degree. The feature correlation degree is used as the core evidence source, and the weld fatigue cycle number and actual working condition load level are introduced as auxiliary evidence sources to output the corresponding comprehensive confidence vector. The crack risk level is determined according to the comprehensive confidence vector.

[0013] Furthermore, the step of determining the crack hazard level based on the comprehensive confidence vector includes: Based on the comprehensive confidence vector, a corresponding multi-dimensional judgment benchmark space is constructed. Simultaneously, the comprehensive confidence vector is cross-domain mapped in the multi-dimensional judgment benchmark space through a domain adversarial training algorithm to output a normalized confidence vector. Based on the normalized confidence vector, the full life cycle monitoring data of the subframe weld is synchronously input to create a corresponding grade mapping model; The normalized confidence vector is input into the grade mapping model to output the corresponding initial crack risk level. Simultaneously, a reinforcement learning algorithm is used to dynamically correct the crack risk level to output the corresponding crack risk level.

[0014] The second aspect of the present invention proposes: A fatigue crack identification system for subframe welds, wherein the system comprises: The acquisition module is used to acquire stress distribution cloud maps of the subframe welds under preset working conditions through finite element simulation, so as to determine the corresponding target risk area based on the stress distribution cloud map, and simultaneously dynamically adjust the spacing, angle and excitation power of the preset excitation coil array according to the spatial position of the target risk area to generate a directional eddy current excitation field focused on the target risk area. The construction module is used to collect the microscopic eddy current distortion signal and macroscopic eddy current response signal generated by the subframe weld under the directional eddy current excitation field, so as to construct the corresponding dynamic baseline signal library, and simultaneously perform adaptive calibration through least squares support vector machine to output the corresponding target fusion signal; The processing module is used to decouple the target fusion signal from the crack in the subframe weld through Bayesian inference, and simultaneously collect the ultrasonic reflection signal generated in the target risk area through an ultrasonic sensor. The detection module is used to detect the crack reflection amplitude characteristics in the ultrasonic reflection signal, and simultaneously perform correlation analysis between the crack reflection amplitude characteristics and the characteristic parameters to determine the corresponding crack hazard level based on the analysis results.

[0015] Furthermore, the acquisition module is specifically used for: The residual stress distribution data of the subframe weld was detected by X-ray diffraction and simultaneously incorporated into the weld finite element model as the initial stress field. Based on the dynamic working condition sequence of the actual service of the subframe, the corresponding load was applied sequentially through the weld finite element model to generate a stress distribution cloud map covering the entire working condition cycle. Based on the stress distribution cloud map, the stress data of each weld unit in the subframe weld is time-series tracked to detect the evolution trajectory of the stress peak, and the slope and fluctuation coefficient of the evolution trajectory are calculated simultaneously to screen out stress-sensitive areas accordingly. The target risk area is determined based on the stress-sensitive area.

[0016] Furthermore, the acquisition module is specifically used for: The type, size, and location of inherent welding defects in the stress-sensitive region are detected to create a coupling risk model between welding defects and stress. The corresponding coupling risk coefficient is calculated by combining time-series stress data, and the defect-stress coupling sensitive region is screened according to the magnitude of the coupling risk coefficient. The magnetic memory signal of the defect-stress coupling sensitive area is collected by a magnetic memory sensor, and the fatigue damage accumulation degree of the subframe weld is determined according to the magnetic memory signal to identify the corresponding risk incubation area. The influence weight of potential crack propagation in the risk incubation area relative to the overall load-bearing performance of the subframe is analyzed, and the crack propagation path and key node corresponding to the highest influence weight are detected. At the same time, the area covering the crack propagation path and the key node is set as the target risk area.

[0017] Furthermore, the processing module is specifically used for: The target fused signal is adaptively decomposed using an ensemble empirical mode decomposition algorithm to obtain several intrinsic mode functions. The instantaneous frequency, instantaneous amplitude, and kurtosis value of each intrinsic mode function are calculated simultaneously to create a corresponding multidimensional feature matrix. The multidimensional feature matrix is ​​converted into a corresponding decoupled feature set, and the decoupled feature set is simultaneously divided into a crack-sensitive feature layer and an interference feature layer to perform layered decoupling of crack feature components and crack interference components, and the corresponding initial crack feature parameters are output synchronously. The initial crack feature parameters are optimized and corrected to generate the corresponding feature parameters.

[0018] Furthermore, the processing module is specifically used for: The attention-enhanced convolutional neural network constructs a corresponding temperature compensation factor based on the actual temperature monitoring data of the weld, and performs feature enhancement processing on the initial crack feature parameters based on the temperature compensation factor, and outputs the corresponding crack feature vector simultaneously. A virtual simulation scenario is constructed based on the preset working conditions. The crack feature vector is simultaneously substituted into each of the virtual simulation scenarios for simulation verification. The fluctuation range of the feature parameters under each working condition is calculated by the interval estimation algorithm. If the fluctuation range is detected to meet the preset range requirements, the crack feature vector is converted into the corresponding target parameter, and the target parameter is simultaneously set as the feature parameter.

[0019] Furthermore, the detection module is specifically used for: The crack reflection amplitude feature is mapped to the feature parameter as a corresponding modal feature vector, and a sparse coding algorithm is simultaneously used to sparsely represent the modal feature vector to output the corresponding sparse feature coefficients. Based on the sparse characteristic coefficients, the mechanical parameters of the weld material are introduced as physical constraints. The Lagrange multiplier method is used to solve the feature correlation objective function that satisfies the constraints, so as to output the corresponding feature correlation degree. The feature correlation degree is used as the core evidence source, and the weld fatigue cycle number and actual working condition load level are introduced as auxiliary evidence sources to output the corresponding comprehensive confidence vector. The crack risk level is determined according to the comprehensive confidence vector.

[0020] Furthermore, the detection module is specifically used for: Based on the comprehensive confidence vector, a corresponding multi-dimensional judgment benchmark space is constructed. Simultaneously, the comprehensive confidence vector is cross-domain mapped in the multi-dimensional judgment benchmark space through a domain adversarial training algorithm to output a normalized confidence vector. Based on the normalized confidence vector, the full life cycle monitoring data of the subframe weld is synchronously input to create a corresponding grade mapping model; The normalized confidence vector is input into the grade mapping model to output the corresponding initial crack risk level. Simultaneously, a reinforcement learning algorithm is used to dynamically correct the crack risk level to output the corresponding crack risk level.

[0021] The third aspect of the present invention proposes: A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the fatigue crack identification method for subframe welds as described above.

[0022] The fourth aspect of the present invention proposes: A readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the fatigue crack identification method for subframe welds as described above.

[0023] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0024] Figure 1 A flowchart of the fatigue crack identification method for subframe welds provided in the first embodiment of the present invention; Figure 2 The structural block diagram of the fatigue crack identification system for subframe welds provided in the third embodiment of the present invention is shown.

[0025] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0026] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0027] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0029] Please see Figure 1 The image shows a method for identifying fatigue cracks in subframe welds provided in the first embodiment of the present invention. This method can accurately identify cracks in subframe welds and simultaneously output the corresponding crack level, thereby improving the identification efficiency.

[0030] Specifically, this embodiment provides: A method for identifying fatigue cracks in subframe welds, wherein the method includes: Step S10: The stress distribution cloud map of the subframe weld under preset working conditions is collected by finite element simulation. The corresponding target risk area is determined according to the stress distribution cloud map. Simultaneously, the spacing, angle and excitation power of the preset excitation coil array are dynamically adjusted according to the spatial position of the target risk area to generate a directional eddy current excitation field focused on the target risk area. It should be noted that, firstly, addressing the problems of traditional inspection methods such as "full-area scanning, low efficiency, and easy omission of high-risk areas," finite element simulation is used to simulate the stress distribution of the subframe under preset working conditions (such as full-load bumps, emergency braking, and steering limit conditions), generating a stress distribution cloud map. Specifically, stress concentration areas are high-incidence areas of fatigue cracks, thereby identifying target risk areas and defining precise ranges for subsequent inspections. Simultaneously, based on the spatial location of the target risk area (such as weld corners, lap joints, and other complex structural parts), the spacing, angle, and excitation power of the preset excitation coil array are dynamically adjusted to generate a directional eddy current excitation field. Directional excitation can concentrate eddy current energy in the target area, enhance the disturbance effect of cracks on the eddy current signal, avoid structural interference in non-risk areas, and improve the signal-to-noise ratio.

[0031] Step S20: Collect the microscopic eddy current distortion signal and macroscopic eddy current response signal generated by the subframe weld under the directional eddy current excitation field to construct the corresponding dynamic baseline signal library, and simultaneously perform adaptive calibration through least squares support vector machine to output the corresponding target fusion signal; It should be noted that, secondly, the microscopic eddy current distortion signal (reflecting the microscopic morphology of cracks, such as width and depth) and macroscopic eddy current response signal (reflecting the macroscopic distribution of cracks, such as length and extension direction) of the subframe weld seam under directional eddy current excitation field are collected to construct a dynamic baseline signal library. Specifically, the baseline library contains standard signals under crack-free conditions to provide a reference for crack signal identification. Since eddy current signals are easily affected by temperature, material inhomogeneity, etc., least squares support vector machine is used for adaptive calibration: the algorithm learns the mapping relationship between interference factors and signal deviation, dynamically corrects the collected signal, and outputs the target fusion signal after eliminating interference to ensure the accuracy of the signal.

[0032] Step S30: Decouple the target fusion signal from the crack in the subframe weld by Bayesian inference, and simultaneously collect the ultrasonic reflection signal generated in the target risk area by an ultrasonic sensor. It should be noted that the target fusion signal contains multiple components such as cracks, structural noise, and excitation interference, which need to be decoupled using Bayesian inference. Bayesian inference has powerful probabilistic modeling capabilities and can separate crack-related feature parameters (such as eddy current distortion amplitude and phase shift) based on prior knowledge (such as the correlation between cracks and eddy current signals). At the same time, to improve feature reliability, an ultrasonic sensor is introduced to collect ultrasonic reflection signals from the target risk area. Specifically, the ultrasonic signal is sensitive to the depth and length of the crack and complements the eddy current signal to achieve multimodal verification.

[0033] Step S40: Detect the crack reflection amplitude characteristics in the ultrasonic reflection signal, and simultaneously perform correlation analysis between the crack reflection amplitude characteristics and the characteristic parameters to determine the corresponding crack hazard level based on the analysis results.

[0034] It should be noted that, finally, the crack reflection amplitude characteristics in the ultrasonic reflection signal (the reflection amplitude is positively correlated with the crack size) are detected, and correlation analysis is performed with the crack characteristic parameters obtained by Bayesian inference: by calculating the correlation strength between the two, the risk of misjudgment by a single signal is eliminated; based on the analysis results, the crack hazard level (such as slight, moderate, or severe) is determined, providing a basis for decision-making on the maintenance and replacement of the subframe, thus completing the entire fatigue crack identification process.

[0035] In addition, it should be noted that the second embodiment of the present invention provides:

[0036] Furthermore, the step of acquiring stress distribution cloud maps of the subframe welds under preset working conditions through finite element simulation, and determining the corresponding target risk areas based on the stress distribution cloud maps, includes: The residual stress distribution data of the subframe weld was detected by X-ray diffraction and simultaneously incorporated into the weld finite element model as the initial stress field. Based on the dynamic working condition sequence of the actual service of the subframe, the corresponding load was applied sequentially through the weld finite element model to generate a stress distribution cloud map covering the entire working condition cycle. Based on the stress distribution cloud map, the stress data of each weld unit in the subframe weld is time-series tracked to detect the evolution trajectory of the stress peak, and the slope and fluctuation coefficient of the evolution trajectory are calculated simultaneously to screen out stress-sensitive areas accordingly. The target risk area is determined based on the stress-sensitive area.

[0037] It should be noted that, firstly, the residual stress generated during the welding process is a significant factor in the initiation of fatigue cracks. Traditional finite element simulations do not consider residual stress, which can easily lead to misjudgment of risk areas. Therefore, X-ray diffraction is used to detect the welding residual stress distribution data of the subframe welds, and this data is incorporated into the weld finite element model as the initial stress field. Simultaneously, based on the dynamic operating condition sequence of the subframe in actual service (such as urban road driving, high-speed driving, and mountain road bumps), the finite element model is loaded with corresponding loads (such as vertical loads, horizontal loads, and torque) in sequence to generate a stress distribution cloud map covering the entire operating condition cycle. Specifically, the full-condition time-series loading can simulate the stress evolution process of the weld in actual use, ensuring the authenticity of the stress distribution cloud map.

[0038] Secondly, based on the stress distribution cloud map under all working conditions, the stress data of each weld unit in the subframe weld is tracked over time: by tracking the stress changes of each unit under different working conditions, the evolution trajectory of the stress peak is detected (such as the stress peak of a certain unit continuously increasing under bumpy working conditions); the slope of the evolution trajectory (reflecting the stress growth rate) and the fluctuation coefficient (reflecting stress stability) are calculated, and stress-sensitive areas with fast stress growth and large fluctuations are screened out. Specifically, these areas are the most likely places for fatigue cracks to initiate, providing a core basis for determining the target risk area.

[0039] Finally, the selected stress-sensitive areas are used directly as the basis for the target risk areas, ensuring that subsequent detection focuses on the core areas with high stress and high crack risk, avoiding waste of detection resources and improving identification efficiency and accuracy.

[0040] Furthermore, the step of determining the target risk area based on the stress-sensitive area includes: The type, size, and location of inherent welding defects in the stress-sensitive region are detected to create a coupling risk model between welding defects and stress. The corresponding coupling risk coefficient is calculated by combining time-series stress data, and the defect-stress coupling sensitive region is screened according to the magnitude of the coupling risk coefficient. The magnetic memory signal of the defect-stress coupling sensitive area is collected by a magnetic memory sensor, and the fatigue damage accumulation degree of the subframe weld is determined according to the magnetic memory signal to identify the corresponding risk incubation area. The influence weight of potential crack propagation in the risk incubation area relative to the overall load-bearing performance of the subframe is analyzed, and the crack propagation path and key node corresponding to the highest influence weight are detected. At the same time, the area covering the crack propagation path and the key node is set as the target risk area.

[0041] It should be noted that, firstly, if inherent welding defects (such as porosity, slag inclusions, and incomplete penetration) exist in stress-sensitive areas, they will further exacerbate stress concentration and accelerate crack initiation. Therefore, it is necessary to detect the type, size, and location of inherent welding defects in stress-sensitive areas and construct a coupling risk model of welding defects and stress. Combined with time-series stress data, the coupling risk coefficient is calculated (coefficient = defect influence weight × stress peak value; the larger the defect size and the higher the stress, the larger the coefficient). Based on the coupling risk coefficient, defect-stress coupling sensitive areas are screened out. Specifically, the crack initiation risk in these areas is much higher than that in purely stress-sensitive areas, making them more critical potential risk areas.

[0042] Secondly, the initiation of fatigue cracks is directly related to the accumulation of fatigue damage. The degree of damage cannot be determined solely by stress and defects. Therefore, magnetic memory sensors are used to collect magnetic memory signals from defect-stress coupling sensitive areas. During the accumulation of fatigue damage in metallic materials, the magnetic domain structure changes, leading to abnormal magnetic memory signals. By analyzing the peak value, gradient, and other characteristics of the magnetic memory signals, the degree of fatigue damage accumulation in the subframe welds (such as mild, moderate, and severe damage) can be determined, identifying high-risk areas with severe damage accumulation. Specifically, these areas are close to the critical state of crack initiation and require focused detection.

[0043] Finally, the impact of crack propagation in different risk incubation areas on the overall load-bearing performance of the subframe varies. Therefore, the influence weight of potential crack propagation in risk incubation areas on the overall load-bearing performance is analyzed: the load-bearing capacity changes after crack propagation in different areas are simulated by finite element simulation, and the crack propagation path and key nodes (such as the node where the crack propagates to the main load-bearing weld) corresponding to the highest influence weight are detected; the area covering the crack propagation path and key nodes is set as the target risk area. Specifically, crack propagation in this type of area will directly threaten the load-bearing safety of the subframe and is the core area that needs to be accurately detected.

[0044] Furthermore, the step of decoupling the characteristic parameters related to the cracks in the subframe weld using Bayesian inference includes: The target fused signal is adaptively decomposed using an ensemble empirical mode decomposition algorithm to obtain several intrinsic mode functions. The instantaneous frequency, instantaneous amplitude, and kurtosis value of each intrinsic mode function are calculated simultaneously to create a corresponding multidimensional feature matrix. The multidimensional feature matrix is ​​converted into a corresponding decoupled feature set, and the decoupled feature set is simultaneously divided into a crack-sensitive feature layer and an interference feature layer to perform layered decoupling of crack feature components and crack interference components, and the corresponding initial crack feature parameters are output synchronously. The initial crack feature parameters are optimized and corrected to generate the corresponding feature parameters.

[0045] It should be noted that, firstly, the target fusion signal is a complex nonlinear signal containing cracks and interference (temperature, material inhomogeneity), and direct feature extraction is easily affected by interference. Therefore, an ensemble empirical mode decomposition algorithm is used to adaptively decompose the target fusion signal. This algorithm can decompose the complex signal into several intrinsic mode functions (IMFs), each IMF corresponding to signal components in different frequency ranges, realizing frequency stratification of the signal. The instantaneous frequency (reflecting signal frequency changes), instantaneous amplitude (reflecting signal intensity changes), and kurtosis value (reflecting signal impact characteristics; cracks will cause the kurtosis value to increase) of each IMF are calculated to construct a multidimensional feature matrix. Specifically, the multidimensional feature matrix comprehensively characterizes the time-domain and frequency-domain features of the target fusion signal, providing rich data support for subsequent crack feature separation.

[0046] Secondly, the multidimensional feature matrix is ​​converted into a decoupled feature set. Based on prior knowledge (such as the frequency range of crack features and kurtosis threshold), the decoupled feature set is divided into a crack-sensitive feature layer (such as IMF features in a specific frequency range and high kurtosis value features) and an interference feature layer (such as low-frequency IMF features caused by temperature). The two feature layers are decoupled hierarchically through Bayesian inference: the posterior probability of each feature belonging to crack is calculated using a Bayesian probability model, crack-sensitive features with high posterior probability are retained, interference features are eliminated, and the initial crack feature parameters are output. Specifically, hierarchical decoupling can effectively separate crack features from interference components and improve the reliability of the initial feature parameters.

[0047] Finally, the initial crack feature parameters are optimized and corrected: outliers (such as feature mutations caused by sensor noise) are removed, feature fluctuations are corrected through smoothing, and the final crack-related feature parameters are generated to provide accurate feature input for subsequent crack level determination.

[0048] Furthermore, the step of optimizing and correcting the initial crack feature parameters to generate the corresponding feature parameters includes: The attention-enhanced convolutional neural network constructs a corresponding temperature compensation factor based on the actual temperature monitoring data of the weld, and performs feature enhancement processing on the initial crack feature parameters based on the temperature compensation factor, and outputs the corresponding crack feature vector simultaneously. A virtual simulation scenario is constructed based on the preset working conditions. The crack feature vector is simultaneously substituted into each of the virtual simulation scenarios for simulation verification. The fluctuation range of the feature parameters under each working condition is calculated by the interval estimation algorithm. If the fluctuation range is detected to meet the preset range requirements, the crack feature vector is converted into the corresponding target parameter, and the target parameter is simultaneously set as the feature parameter.

[0049] It should be noted that, firstly, temperature changes cause variations in the electrical conductivity and magnetic permeability of the weld material, which in turn interferes with the eddy current signal and affects the accuracy of crack feature parameters. Therefore, a temperature compensation factor is constructed based on a convolutional neural network (CNN) enhanced by an attention mechanism, combined with actual temperature monitoring data of the weld. The attention mechanism can focus on the correlation region between temperature and feature parameters, and the CNN dynamically generates the temperature compensation factor by learning the mapping relationship between temperature changes and feature deviations. The temperature compensation factor is used to perform feature enhancement processing on the initial crack feature parameters (such as correcting feature amplitude deviations caused by temperature), and outputs a crack feature vector that eliminates temperature interference.

[0050] Secondly, under different operating conditions (such as different load magnitudes and driving speeds), the stress state of the weld is different, and the crack characteristic parameters may fluctuate. Therefore, virtual simulation scenarios (such as combinations of different loads and temperatures) are constructed based on preset operating conditions. The crack characteristic vector is substituted into each virtual simulation scenario for simulation verification: the changes of characteristic parameters under different operating conditions are simulated, and the fluctuation range of characteristic parameters under each operating condition is calculated through the interval estimation algorithm. Specifically, the fluctuation range reflects the stability of characteristic parameters under actual service conditions.

[0051] Finally, a preset range requirement is set (e.g., the amplitude deviation of the fluctuation range is ≤5%). If the fluctuation range is found to meet the requirement, it indicates that the crack feature vector is stable and reliable under different working conditions. It is then converted into the corresponding target parameter and set as the final crack feature parameter. If the requirement is not met, the temperature compensation factor or feature enhancement strategy is readjusted until the fluctuation range meets the requirement, ensuring the engineering practicality of the feature parameter.

[0052] Furthermore, the step of performing correlation analysis between the crack reflection amplitude characteristics and the characteristic parameters to determine the corresponding crack hazard level based on the analysis results includes: The crack reflection amplitude feature is mapped to the feature parameter as a corresponding modal feature vector, and a sparse coding algorithm is simultaneously used to sparsely represent the modal feature vector to output the corresponding sparse feature coefficients. Based on the sparse characteristic coefficients, the mechanical parameters of the weld material are introduced as physical constraints. The Lagrange multiplier method is used to solve the feature correlation objective function that satisfies the constraints, so as to output the corresponding feature correlation degree. The feature correlation degree is used as the core evidence source, and the weld fatigue cycle number and actual working condition load level are introduced as auxiliary evidence sources to output the corresponding comprehensive confidence vector. The crack risk level is determined according to the comprehensive confidence vector.

[0053] It should be noted that, firstly, the crack reflection amplitude characteristics (ultrasonic signal) and crack characteristic parameters (eddy current signal) are mapped to a modal feature vector of a unified dimension. Specifically, the unified mapping enables the collaborative analysis of multi-modal signals and avoids the one-sidedness of a single signal. Secondly, a sparse coding algorithm is used to sparsely represent the modal feature vector: this algorithm can eliminate redundant features, retain the key features that best characterize the crack state, and output sparse feature coefficients, thereby improving the efficiency of feature analysis.

[0054] Secondly, sparse feature coefficients only reflect the correlation of signal features and do not consider the physical properties of weld materials, which can easily lead to judgment bias. Therefore, mechanical parameters of weld materials (such as yield strength and fatigue limit) are introduced as physical constraints. The feature correlation objective function that satisfies the constraints is solved by the Lagrange multiplier method (objective function = feature correlation strength + mechanical parameter constraint penalty term), and the feature correlation degree is output. Specifically, the feature correlation measure quantifies the degree of matching between eddy current crack features and ultrasonic crack features. The higher the correlation degree, the more reliable the crack identification result.

[0055] Finally, determining the crack level solely based on feature correlation is insufficient and requires integration with actual service information of the weld. Therefore, feature correlation is used as the core evidence source, while weld fatigue cycle count (reflecting the degree of damage accumulation) and actual working load level (reflecting the stress magnitude) are introduced as auxiliary evidence sources. Multi-source evidence is integrated through DS evidence theory to output a comprehensive confidence vector (each element of the vector corresponds to the confidence level of different crack levels). The crack hazard level is determined based on the comprehensive confidence vector (e.g., the level with the highest confidence is the final level), achieving multi-dimensional and comprehensive crack level determination.

[0056] Furthermore, the step of determining the crack hazard level based on the comprehensive confidence vector includes: Based on the comprehensive confidence vector, a corresponding multi-dimensional judgment benchmark space is constructed. Simultaneously, the comprehensive confidence vector is cross-domain mapped in the multi-dimensional judgment benchmark space through a domain adversarial training algorithm to output a normalized confidence vector. Based on the normalized confidence vector, the full life cycle monitoring data of the subframe weld is synchronously input to create a corresponding grade mapping model; The normalized confidence vector is input into the grade mapping model to output the corresponding initial crack risk level. Simultaneously, a reinforcement learning algorithm is used to dynamically correct the crack risk level to output the corresponding crack risk level.

[0057] It should be noted that, firstly, the weld materials and welding processes of different batches of subframes may differ, leading to different distributions of the comprehensive confidence vector. Direct judgment is prone to bias. Therefore, a multi-dimensional judgment benchmark space is constructed based on the comprehensive confidence vector (the benchmark space includes standard confidence distributions under different materials and processes). The comprehensive confidence vector is then mapped across the benchmark space using a domain adversarial training algorithm. This algorithm can eliminate the distribution differences between different batches of subframes, map the current confidence vector to the standard benchmark space, and output a normalized confidence vector. Specifically, normalization ensures the uniformity of the judgment standard and improves the versatility of the method.

[0058] Secondly, the normalized confidence vector needs to be combined with the full life cycle data of the weld (such as historical inspection records, maintenance records, and service life) to more accurately determine the level. Therefore, the normalized confidence vector and the full life cycle monitoring data are input synchronously to create a level mapping model: the model establishes an accurate mapping relationship by learning the correlation between "confidence vector - actual crack level" in historical data. Specifically, the introduction of full life cycle data can make up for the shortcomings of single signal features and improve the accuracy of the model's judgment.

[0059] Finally, the normalized confidence vector is input into the grade mapping model to output the initial crack hazard level. Due to the uncertainty of actual service conditions (such as sudden overload), a reinforcement learning algorithm is used for dynamic correction: the reward function is "minimizing the deviation between the judgment result and the actual crack state". The grade judgment result is dynamically adjusted through iterative learning, and finally the accurate crack hazard level is output. Specifically, dynamic correction can eliminate the judgment deviation caused by the uncertainty of the service conditions, ensure the reliability of the grade judgment, and provide accurate decision-making basis for the safe operation and maintenance of the subframe.

[0060] Please see Figure 2 The third embodiment of the present invention provides: A fatigue crack identification system for subframe welds, wherein the system comprises: The acquisition module is used to acquire stress distribution cloud maps of the subframe welds under preset working conditions through finite element simulation, so as to determine the corresponding target risk area based on the stress distribution cloud map, and simultaneously dynamically adjust the spacing, angle and excitation power of the preset excitation coil array according to the spatial position of the target risk area to generate a directional eddy current excitation field focused on the target risk area. The construction module is used to collect the microscopic eddy current distortion signal and macroscopic eddy current response signal generated by the subframe weld under the directional eddy current excitation field, so as to construct the corresponding dynamic baseline signal library, and simultaneously perform adaptive calibration through least squares support vector machine to output the corresponding target fusion signal; The processing module is used to decouple the target fusion signal from the crack in the subframe weld through Bayesian inference, and simultaneously collect the ultrasonic reflection signal generated in the target risk area through an ultrasonic sensor. The detection module is used to detect the crack reflection amplitude characteristics in the ultrasonic reflection signal, and simultaneously perform correlation analysis between the crack reflection amplitude characteristics and the characteristic parameters to determine the corresponding crack hazard level based on the analysis results.

[0061] Furthermore, the acquisition module is specifically used for: The residual stress distribution data of the subframe weld was detected by X-ray diffraction and simultaneously incorporated into the weld finite element model as the initial stress field. Based on the dynamic working condition sequence of the actual service of the subframe, the corresponding load was applied sequentially through the weld finite element model to generate a stress distribution cloud map covering the entire working condition cycle. Based on the stress distribution cloud map, the stress data of each weld unit in the subframe weld is time-series tracked to detect the evolution trajectory of the stress peak, and the slope and fluctuation coefficient of the evolution trajectory are calculated simultaneously to screen out stress-sensitive areas accordingly. The target risk area is determined based on the stress-sensitive area.

[0062] Furthermore, the acquisition module is specifically used for: The type, size, and location of inherent welding defects in the stress-sensitive region are detected to create a coupling risk model between welding defects and stress. The corresponding coupling risk coefficient is calculated by combining time-series stress data, and the defect-stress coupling sensitive region is screened according to the magnitude of the coupling risk coefficient. The magnetic memory signal of the defect-stress coupling sensitive area is collected by a magnetic memory sensor, and the fatigue damage accumulation degree of the subframe weld is determined according to the magnetic memory signal to identify the corresponding risk incubation area. The influence weight of potential crack propagation in the risk incubation area relative to the overall load-bearing performance of the subframe is analyzed, and the crack propagation path and key node corresponding to the highest influence weight are detected. At the same time, the area covering the crack propagation path and the key node is set as the target risk area.

[0063] Furthermore, the processing module is specifically used for: The target fused signal is adaptively decomposed using an ensemble empirical mode decomposition algorithm to obtain several intrinsic mode functions. The instantaneous frequency, instantaneous amplitude, and kurtosis value of each intrinsic mode function are calculated simultaneously to create a corresponding multidimensional feature matrix. The multidimensional feature matrix is ​​converted into a corresponding decoupled feature set, and the decoupled feature set is simultaneously divided into a crack-sensitive feature layer and an interference feature layer to perform layered decoupling of crack feature components and crack interference components, and the corresponding initial crack feature parameters are output synchronously. The initial crack feature parameters are optimized and corrected to generate the corresponding feature parameters.

[0064] Furthermore, the processing module is specifically used for: The attention-enhanced convolutional neural network constructs a corresponding temperature compensation factor based on the actual temperature monitoring data of the weld, and performs feature enhancement processing on the initial crack feature parameters based on the temperature compensation factor, and outputs the corresponding crack feature vector simultaneously. A virtual simulation scenario is constructed based on the preset working conditions. The crack feature vector is simultaneously substituted into each of the virtual simulation scenarios for simulation verification. The fluctuation range of the feature parameters under each working condition is calculated by the interval estimation algorithm. If the fluctuation range is detected to meet the preset range requirements, the crack feature vector is converted into the corresponding target parameter, and the target parameter is simultaneously set as the feature parameter.

[0065] Furthermore, the detection module is specifically used for: The crack reflection amplitude feature is mapped to the feature parameter as a corresponding modal feature vector, and a sparse coding algorithm is simultaneously used to sparsely represent the modal feature vector to output the corresponding sparse feature coefficients. Based on the sparse characteristic coefficients, the mechanical parameters of the weld material are introduced as physical constraints. The Lagrange multiplier method is used to solve the feature correlation objective function that satisfies the constraints, so as to output the corresponding feature correlation degree. The feature correlation degree is used as the core evidence source, and the weld fatigue cycle number and actual working condition load level are introduced as auxiliary evidence sources to output the corresponding comprehensive confidence vector. The crack risk level is determined according to the comprehensive confidence vector.

[0066] Furthermore, the detection module is specifically used for: Based on the comprehensive confidence vector, a corresponding multi-dimensional judgment benchmark space is constructed. Simultaneously, the comprehensive confidence vector is cross-domain mapped in the multi-dimensional judgment benchmark space through a domain adversarial training algorithm to output a normalized confidence vector. Based on the normalized confidence vector, the full life cycle monitoring data of the subframe weld is synchronously input to create a corresponding grade mapping model; The normalized confidence vector is input into the grade mapping model to output the corresponding initial crack risk level. Simultaneously, a reinforcement learning algorithm is used to dynamically correct the crack risk level to output the corresponding crack risk level.

[0067] The fourth embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the fatigue crack identification method for subframe welds as described above.

[0068] The fifth embodiment of the present invention provides a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the fatigue crack identification method for subframe welds as described above.

[0069] In summary, the fatigue crack identification method and system for subframe welds provided in the above embodiments of the present invention can accurately identify cracks in the welds and simultaneously output the corresponding crack level, thereby improving the crack identification efficiency.

[0070] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0071] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0072] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0073] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0074] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0075] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for identifying fatigue cracks in subframe welds, characterized in that, The method includes: The stress distribution cloud map of the subframe weld under preset working conditions is collected by finite element simulation. The corresponding target risk area is determined according to the stress distribution cloud map. Simultaneously, the spacing, angle and excitation power of the preset excitation coil array are dynamically adjusted according to the spatial position of the target risk area to generate a directional eddy current excitation field focused on the target risk area. The microscopic eddy current distortion signal and macroscopic eddy current response signal generated by the subframe weld under the directional eddy current excitation field are collected to construct a corresponding dynamic baseline signal library. Simultaneously, the least squares support vector machine is used for adaptive calibration to output the corresponding target fusion signal. The characteristic parameters related to the cracks in the subframe weld are decoupled from the target fusion signal by Bayesian inference, and the ultrasonic reflection signals generated in the target risk area are collected simultaneously by an ultrasonic sensor. The crack reflection amplitude characteristics in the ultrasonic reflection signal are detected, and the crack reflection amplitude characteristics are simultaneously correlated with the characteristic parameters to determine the corresponding crack hazard level based on the analysis results.

2. The method for identifying fatigue cracks in subframe welds according to claim 1, characterized in that, The step of acquiring stress distribution cloud maps of subframe welds under preset working conditions through finite element simulation, and determining the corresponding target risk areas based on the stress distribution cloud maps, includes: The residual stress distribution data of the subframe weld was detected by X-ray diffraction and simultaneously incorporated into the weld finite element model as the initial stress field. Based on the dynamic working condition sequence of the actual service of the subframe, the corresponding load was applied sequentially through the weld finite element model to generate a stress distribution cloud map covering the entire working condition cycle. Based on the stress distribution cloud map, the stress data of each weld unit in the subframe weld is time-series tracked to detect the evolution trajectory of the stress peak, and the slope and fluctuation coefficient of the evolution trajectory are calculated simultaneously to screen out stress-sensitive areas accordingly. The target risk area is determined based on the stress-sensitive area.

3. The method for identifying fatigue cracks in subframe welds according to claim 2, characterized in that, The step of determining the target risk area based on the stress-sensitive area includes: The type, size, and location of inherent welding defects in the stress-sensitive region are detected to create a coupling risk model between welding defects and stress. The corresponding coupling risk coefficient is calculated by combining time-series stress data, and the defect-stress coupling sensitive region is screened according to the magnitude of the coupling risk coefficient. The magnetic memory signal of the defect-stress coupling sensitive area is collected by a magnetic memory sensor, and the fatigue damage accumulation degree of the subframe weld is determined according to the magnetic memory signal to identify the corresponding risk incubation area. The influence weight of potential crack propagation in the risk incubation area relative to the overall load-bearing performance of the subframe is analyzed, and the crack propagation path and key node corresponding to the highest influence weight are detected. At the same time, the area covering the crack propagation path and the key node is set as the target risk area.

4. The method for identifying fatigue cracks in subframe welds according to claim 1, characterized in that, The step of decoupling the characteristic parameters related to the cracks in the subframe weld using Bayesian inference includes: The target fused signal is adaptively decomposed using an ensemble empirical mode decomposition algorithm to obtain several intrinsic mode functions. The instantaneous frequency, instantaneous amplitude, and kurtosis value of each intrinsic mode function are calculated simultaneously to create a corresponding multidimensional feature matrix. The multidimensional feature matrix is ​​converted into a corresponding decoupled feature set, and the decoupled feature set is simultaneously divided into a crack sensitive feature layer and an interference feature layer to perform layered decoupling of crack feature components and crack interference components, and the corresponding initial crack feature parameters are output synchronously. The initial crack feature parameters are optimized and corrected to generate the corresponding feature parameters.

5. The method for identifying fatigue cracks in subframe welds according to claim 4, characterized in that, The step of optimizing and correcting the initial crack feature parameters to generate the corresponding feature parameters includes: The attention-enhanced convolutional neural network constructs a corresponding temperature compensation factor based on the actual temperature monitoring data of the weld, and performs feature enhancement processing on the initial crack feature parameters based on the temperature compensation factor, and outputs the corresponding crack feature vector simultaneously. A virtual simulation scenario is constructed based on the preset working conditions. The crack feature vector is simultaneously substituted into each of the virtual simulation scenarios for simulation verification. The fluctuation range of the feature parameters under each working condition is calculated by the interval estimation algorithm. If the fluctuation range is detected to meet the preset range requirements, the crack feature vector is converted into the corresponding target parameter, and the target parameter is simultaneously set as the feature parameter.

6. The method for identifying fatigue cracks in subframe welds according to claim 1, characterized in that, The step of performing correlation analysis between the crack reflection amplitude characteristics and the characteristic parameters, and determining the corresponding crack hazard level based on the analysis results, includes: The crack reflection amplitude feature is mapped to the feature parameter as a corresponding modal feature vector, and a sparse coding algorithm is simultaneously used to sparsely represent the modal feature vector to output the corresponding sparse feature coefficients. Based on the sparse characteristic coefficients, the mechanical parameters of the weld material are introduced as physical constraints. The Lagrange multiplier method is used to solve the feature correlation objective function that satisfies the constraints, so as to output the corresponding feature correlation degree. The feature correlation degree is used as the core evidence source, and the weld fatigue cycle number and actual working condition load level are introduced as auxiliary evidence sources to output the corresponding comprehensive confidence vector. The crack risk level is determined according to the comprehensive confidence vector.

7. The method for identifying fatigue cracks in subframe welds according to claim 6, characterized in that, The step of determining the crack risk level based on the comprehensive confidence vector includes: Based on the comprehensive confidence vector, a corresponding multi-dimensional judgment benchmark space is constructed. Simultaneously, the comprehensive confidence vector is cross-domain mapped in the multi-dimensional judgment benchmark space through a domain adversarial training algorithm to output a normalized confidence vector. Based on the normalized confidence vector, the full life cycle monitoring data of the subframe weld is synchronously input to create a corresponding grade mapping model; The normalized confidence vector is input into the grade mapping model to output the corresponding initial crack risk level. Simultaneously, a reinforcement learning algorithm is used to dynamically correct the crack risk level to output the corresponding crack risk level.

8. A fatigue crack identification system for subframe welds, characterized in that, The system includes: The acquisition module is used to acquire stress distribution cloud maps of the subframe welds under preset working conditions through finite element simulation, so as to determine the corresponding target risk area based on the stress distribution cloud map, and simultaneously dynamically adjust the spacing, angle and excitation power of the preset excitation coil array according to the spatial position of the target risk area to generate a directional eddy current excitation field focused on the target risk area. The construction module is used to collect the microscopic eddy current distortion signal and macroscopic eddy current response signal generated by the subframe weld under the directional eddy current excitation field, so as to construct the corresponding dynamic baseline signal library, and simultaneously perform adaptive calibration through least squares support vector machine to output the corresponding target fusion signal; The processing module is used to decouple the target fusion signal from the crack in the subframe weld through Bayesian inference, and simultaneously collect the ultrasonic reflection signal generated in the target risk area through an ultrasonic sensor. The detection module is used to detect the crack reflection amplitude characteristics in the ultrasonic reflection signal, and simultaneously perform correlation analysis between the crack reflection amplitude characteristics and the characteristic parameters to determine the corresponding crack hazard level based on the analysis results.

9. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the fatigue crack identification method for subframe welds as described in any one of claims 1 to 7.

10. A readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the fatigue crack identification method for subframe welds as described in any one of claims 1 to 7.

Images