Real-time monitoring system for fatigue cracking of steel structure weld based on machine vision

Abstract

The present application relates to the technical field of intelligent scheduling, in particular to a steel structure weld fatigue cracking real-time monitoring system based on machine vision, in the present application, the image acquisition module obtains high stability image sequence under strong vibration environment; the displacement field reconstruction module uses frequency domain-space domain joint decomposition technology to break through the interference of environmental noise and material non-uniformity, synchronously outputs anti-noise deformation field and stress distribution map through double-channel parallel processing; the damage response module accurately locks the sub-pixel level crack initiation coordinate based on strain energy accumulation rate mutation, and dynamically predicts the yield risk area combined with the displacement gradient diffusion model; the early warning module intelligently triggers the hierarchical response according to the spatial coupling state of the crack position and the risk area, positions and maintains in the low risk area, automatically strengthens in the high risk area, and outputs the heat treatment scheme coordinate when there is no crack; the system solves the problems of false detection and single parameter monitoring lagging response caused by vibration noise, and realizes active protection from damage initiation to failure blocking.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, specifically to a real-time monitoring system for fatigue cracking of steel structure welds based on machine vision. Background Technology

[0002] Fatigue cracking of welds in steel structures is a major cause of structural failure in critical engineering projects. Traditional monitoring methods face multiple challenges: contact monitoring networks based on resistance strain gauges require dense sensor deployment, and their single-point measurement characteristics result in insufficient spatial resolution of the micro-deformation field in the heat-affected zone (HAZ) of the weld, and adhesion failure is prone to occur in large-gradient strain fields. Although acoustic emission technology can capture elastic waves of crack propagation, it is insensitive to the micron-level opening displacement during the subcritical crack initiation stage, and the signal propagation is modulated by the structural geometric boundaries, resulting in waveguide effects. Among emerging non-contact visual monitoring methods, digital image correlation technology can acquire full-field displacement, but under strong environmental vibrations (such as bridge traffic loads and factory machinery vibrations), the structural fundamental frequency resonant noise and the actual micro-deformation signal of the material are highly superimposed in the frequency domain, and conventional band filtering will simultaneously weaken key high-frequency damage components.

[0003] A more critical limitation lies in the significant material property gradient within the HAZ formed during welding—a transition from high-hardness quenched martensite at the fusion line to the ductile-plastic structure of the base material. Existing visual algorithms assume uniform isotropic material properties, leading to spurious strain concentrations in the displacement field reconstruction within the modulus abrupt change region. Furthermore, crack initiation and stress concentration evolution exhibit spatiotemporal asynchrony; single-parameter monitoring cannot construct a correlation model between the two, resulting in delayed maintenance decisions. These issues collectively contribute to the persistently high false negative rate of existing systems in early warning of microcracks, necessitating breakthroughs in vibration-resistant material adaptive analysis and multi-damage parameter fusion monitoring technologies. Summary of the Invention

[0004] The purpose of this invention is to provide a real-time monitoring system for fatigue cracking of steel structure welds based on machine vision, in order to solve the problems mentioned in the background art. The specific technologies include how to accurately separate the sub-pixel level micro-deformation signal of the weld area under strong environmental vibration, so as to solve the problem of false crack detection caused by inherent vibration noise masking the true deformation; and how to simultaneously establish a spatial correlation model between crack initiation coordinates and material yield risk zone, so as to solve the problem that traditional single-parameter monitoring cannot provide early warning of the risk of crack-stress concentration co-evolution.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a real-time monitoring system for fatigue cracking of steel structure welds based on machine vision, comprising an image acquisition module, a displacement field reconstruction module, a damage response module, and an early warning module, wherein:

[0006] The image acquisition module uses a rigid vibration-resistant bracket to fix a high-speed industrial camera to the side of the weld area. A laser speckle mark array is pre-placed on the weld surface to form a high-contrast optical feature grid, continuously capturing dynamic image sequences of the weld surface. With the help of adaptive exposure control and a wide-spectrum anti-interference filter, it ensures stable acquisition of the texture details required for sub-pixel displacement analysis under complex lighting and vibration environments. At the same time, the synchronous triggering device ensures the spatiotemporal coherence of the image sequence, providing a basic data source with both high spatial resolution and temporal consistency for displacement field reconstruction.

[0007] The extraction unit in the displacement field reconstruction module utilizes a joint frequency-spatial decomposition technique. It analyzes the inherent vibration characteristics of the structure using a pre-built elastic vibration response spectrum library and employs intrinsic mode decomposition to remove displacement components related to the fundamental frequency resonance. Simultaneously, it constructs spatial constraints using the elastic modulus gradient distribution function of the weld heat-affected zone. This function dynamically correlates with welding process parameters, accurately describing the gradual change in material properties from the weld center to the base material. High-frequency displacement components are then extracted using the displacement field continuity correction equation. This approach accurately separates environmental vibration noise and avoids displacement field distortion caused by material inhomogeneity, achieving dual isolation between material inhomogeneity and environmental vibration noise, and extracting high-frequency displacement components characterizing the true damage.

[0008] The first channel unit in the displacement field reconstruction module is used for physical constraint filtering of the first channel. It inputs high-frequency displacement components into the material continuity constraint model and dynamically adjusts the spatial smoothing coefficient according to the elastic modulus gradient distribution characteristics of the weld area. The high elastic modulus region at the center of the weld adopts a low smoothing strength to protect the micro-deformation abrupt characteristics, while the low elastic modulus region of the base material has an enhanced smoothing strength to suppress noise interference. The spatial consistency correction of the displacement field is achieved by solving the displacement equilibrium equations of Hooke's law, and the low-frequency vibration residual components are eliminated by coupling the fundamental frequency resonant band-stop filtering technology. Finally, the spatially distributed local deformation field with vibration noise elimination is output, generating a local deformation field with clear physical meaning. While retaining key damage characteristics, it achieves the coordinated filtering of vibration noise across the entire domain.

[0009] The second channel unit in the displacement field reconstruction module is used for second-channel stress concentration quantification. An edge-sensitive spatial convolution algorithm is employed to anisotropically process high-frequency displacement components. A composite operator combining Gaussian weighted smoothing and directional gradient detection accurately calculates the magnitude and principal direction of the displacement vector gradient, effectively identifying microscopic discontinuous abrupt regions in the displacement field. Based on the elastoplastic yield criterion, a mapping model between the displacement gradient magnitude and the material stress concentration coefficient is established. When the local gradient magnitude exceeds the material's yield critical threshold, it is automatically marked as a high-risk area of ​​stress concentration, generating a two-dimensional stress distribution map covering the entire weld area. This process transforms the microscopic displacement gradient into a visualized distribution map of full-field stress concentration, achieving quantitative early warning of yield risk under stress sensor-free conditions.

[0010] The damage response module calculates the cumulative rate of distorted strain energy node by node in the sub-pixel grid coordinate system of the weld, based on the plane strain energy density theory. A spatiotemporal accumulation monitoring window is constructed, and a wavelet transform modulus maxima detection algorithm is used to perform abrupt change feature analysis on the strain energy density sequence within a continuous monitoring period, identifying nonlinear jump points in the cumulative rate of strain energy. When the cumulative abrupt change index of a node exceeds a dynamically set threshold, it is marked as a potential crack initiation point. Then, a spatial clustering algorithm is used to remove isolated noise interference, outputting crack spatial coordinates with sub-pixel accuracy. This process captures subcritical crack initiation signals through the spatiotemporal accumulation characteristics of strain energy abrupt changes, significantly improving the early location accuracy and anti-false detection capability of microcracks.

[0011] The damage response module extracts the boundary displacement gradient magnitude of high-risk areas based on stress distribution maps, constructs a gradient diffusion index time-series analysis model, and determines the continuous diffusion characteristics of the yield risk zone by calculating the spatial expansion rate and directional consistency of gradient magnitudes between consecutive image frames. A gradient diffusion dynamics model is established using an analogy method based on the heat conduction equation to quantify the evolution trend of material yield risk and output the global risk zone distribution. This process achieves time-series prediction and spatial location of material yield risk by dynamically analyzing the diffusion path and intensity of stress concentration areas.

[0012] The early warning module maps crack coordinates and the material yield risk zone to a sub-pixel grid coordinate system. Based on a gradient diffusion index threshold, it constructs a spatial classification model of the risk zone, automatically dividing the material yield risk zone into low-risk and high-risk areas. A grid position matching algorithm calculates the positional relationship between crack coordinates and the polygonal boundary of the risk zone. When the crack coordinates are completely contained within the low-risk zone and the spatial overlap exceeds a set proportion, a local maintenance coordinate command is triggered. When the crack coordinates and the high-risk zone have geometric overlap and the overlapping area exceeds a set proportion of the total area of ​​the core zone, a structural strengthening device is automatically activated. When no crack coordinates are detected, coordinates for a preventative heat treatment scheme are generated based on the main diffusion direction and diffusion intensity of the material yield risk zone. The output includes a set of coordinateized process parameters containing the heat treatment temperature curve and its effective range, achieving proactive prevention before damage initiation. This process triggers a graded response mechanism based on the spatial coupling state of crack-stress risk, achieving a leap from passive alarm to proactive intervention in intelligent decision-making.

[0013] Compared with the prior art, the beneficial effects of the present invention are:

[0014] By employing frequency-spatial joint decomposition and material gradient adaptive filtering technology, the interference of environmental vibration noise and the non-uniform material characteristics of welding on displacement field reconstruction is resolved. An innovative dual-channel processing mechanism simultaneously outputs noise-resistant deformation field and stress concentration distribution map, significantly improving the accuracy of microcrack initiation location and the reliability of yield risk prediction. Combined with spatiotemporal analysis of strain energy mutation and gradient diffusion dynamics model, dynamic coupling assessment of crack-stress risk is achieved. Finally, relying on a sub-pixel-level spatial hierarchical decision-making mechanism, a precise response from local maintenance to structural reinforcement is triggered, and preventive heat treatment coordinates are output to actively block damage evolution, achieving a comprehensive leap from passive monitoring to active protection. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall modules of the present invention;

[0016] Figure 2 This is a schematic diagram of the displacement field reconstruction module unit of the present invention.

[0017] In the figure: 100, image acquisition module; 200, displacement field reconstruction module; 201, extraction unit; 202, first channel unit; 203, second channel unit; 300, damage response module; 400, early warning module. Detailed Implementation

[0018] To better understand the purpose, technical solution, and advantages of this application, the application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0019] Unless otherwise defined, the technical or scientific terms used in this application shall have the general meaning understood by one of ordinary skill in the art to which this application pertains. Words such as “a,” “an,” “an,” “the,” “the,” and “these” used in this application do not indicate quantitative limitation and may be singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include steps or modules (units) not listed, or may include other steps or modules (units) inherent to these processes, methods, products, or devices. Words such as “connected,” “linked,” and “coupled” used in this application are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. Normally, the character " / " indicates that the objects before and after it are in an "or" relationship. The terms "first," "second," "third," etc., used in this application are merely to distinguish similar objects and do not represent a specific order of objects.

[0020] Next, please refer to Figure 1 The present invention provides a technical solution: a real-time monitoring system for fatigue cracking of steel structure welds based on machine vision, including an image acquisition module 100, a displacement field reconstruction module 200, a damage response module 300, and an early warning module 400.

[0021] The image acquisition module 100 uses a rigid vibration-resistant bracket to fix a high-speed industrial camera to the side of the weld area. A laser speckle mark array is pre-placed on the weld surface to form a high-contrast optical feature grid. The module continuously captures dynamic image sequences of the weld surface at a sampling rate of no less than 200 frames per second. With the help of adaptive exposure control and a wide-spectrum anti-interference filter, it ensures stable acquisition of the texture details required for sub-pixel displacement analysis under complex lighting and vibration environments. At the same time, the synchronous triggering device ensures the spatiotemporal coherence of the image sequence, providing a basic data source with both high spatial resolution and temporal consistency for displacement field reconstruction.

[0022] Please see Figure 2The extraction unit 201 in the displacement field reconstruction module 200 establishes high-precision displacement field data through sub-pixel-level gridded displacement tracking technology. It calculates the instantaneous displacement vector of each grid node frame-by-frame using phase correlation analysis on continuously acquired weld dynamic image sequences, constructing displacement field data covering the entire weld area. Based on a pre-built elastic vibration response spectrum library, it analyzes the inherent vibration characteristics of the structure and extracts high-frequency displacement components through frequency-spatial joint constraint decomposition technology, specifically including:

[0023] Intrinsic mode decomposition (EMD) technology is used to remove components related to the fundamental frequency resonance of the structure. Furthermore, the gradient distribution function of the elastic modulus of the weld heat-affected zone (HAZ) is introduced as a spatial constraint. This function dynamically describes the gradual change in material properties from the weld center to the base metal based on welding process parameters. Current, voltage, welding speed, and interpass temperature control parameters during the welding process are collected as core input variables. The microstructure distribution from the weld center to the base metal is deduced using a welding metallurgical phase transformation model, and the cooling rate variation characteristics of the region are inverted using the heat conduction equation. Based on the mapping relationship between the microstructure phase content and the elastic modulus, a gradient function model is established that continuously decreases from the high elastic modulus region at the weld center to the low elastic modulus region in the base metal. During system operation, this function dynamically updates the material property change curve based on real-time input welding process parameters, ensuring that the applied spatial constraint accurately matches the actual material mechanical behavior of the weld HAZ. Finally, a high-frequency displacement component with continuous spatial distribution and stable time-varying characteristics is obtained. This high-frequency displacement component fully characterizes the microscopic deformation response of the material under service loads, providing a clearly defined physical source for subsequent dual-channel processing.

[0024] The first channel unit 202 in the displacement field reconstruction module 200 constructs a physical constraint filtering mechanism based on the constitutive relation of elasticity. It inputs the extracted high-frequency displacement components into a material continuity constraint model for processing. This material continuity constraint model dynamically adjusts the spatial smoothing coefficient according to the elastic modulus gradient distribution characteristics of the weld region, specifically including:

[0025] In the material continuity constraint model, the high elastic modulus region at the weld center adopts a low smoothing strength to retain subtle deformation characteristics, while the low elastic modulus region of the base material in the material continuity constraint model enhances the smoothing strength to suppress noise interference. Spatial consistency correction of displacement field data is achieved by solving the displacement equilibrium equations of Hooke's law, and low-frequency vibration residual components are eliminated by synchronously coupled fundamental frequency resonant band-stop filtering technology. After this physical-frequency dual constraint processing, a local deformation field with continuous spatial distribution and eliminated vibration noise is output. This local deformation field contains two-dimensional plane strain components and shear strain components, and its spatial resolution meets the sub-pixel level analysis requirements, fully depicting the true material deformation state of the weld area.

[0026] The second channel unit 203 in the displacement field reconstruction module 200 directly analyzes the spatial gradient distribution characteristics of the high-frequency displacement components and uses an edge-sensitive spatial convolution algorithm to perform anisotropic processing on the displacement field data, specifically including:

[0027] By calculating the magnitude and principal direction of the displacement vector gradient using a composite operator of Gaussian weighted smoothing and directional gradient detection, microscopic discontinuous abrupt regions in the displacement field are accurately identified. Based on the elastoplastic yield criterion, a mapping model between the displacement gradient magnitude and the material stress concentration coefficient is established. When the local gradient magnitude exceeds the material yield critical threshold, it is automatically marked as a high-risk area of ​​stress concentration. Finally, a two-dimensional stress distribution map covering the entire weld area is generated. The map clearly marks the spatial distribution characteristics from low-risk areas to high-risk areas. Its spatial positioning accuracy meets the requirements of micron-level damage identification, providing a visual decision-making basis for the detection of material yield precursors.

[0028] The displacement field reconstruction module 200 utilizes dual-channel technology with homogeneous high-frequency displacement components as the basic input data. Through differentiated processing paths, it reveals the dual physical nature of material deformation: the first channel extracts macroscopic deformation features through physical constraint filtering, solving the problem of reconstructing the real deformation field under vibration interference environment; the second channel locates microscopic stress concentration areas through gradient feature analysis, overcoming the problem that traditional methods cannot capture early damage; the output data of the two channels are strictly aligned in the spatial grid coordinate system, the deformation field data provides the basis for overall deformation quantification, and the stress distribution map locates local danger areas, forming a complementary analysis system of macroscopic deformation and microscopic damage, providing multi-dimensional criterion support for structural integrity assessment.

[0029] The damage response module 300 calculates the cumulative rate of distorted strain energy node-by-node in the sub-pixel grid coordinate system of the weld global domain based on local deformation field data, specifically including:

[0030] First, based on the plane strain energy density theory, the strain components are converted into instantaneous values ​​of distortion energy density. Second, a spatiotemporal cumulative monitoring window is constructed, and abrupt change feature analysis is performed on the strain energy density sequence of each grid node within a continuous monitoring period. The wavelet transform modulus maxima detection algorithm is used to identify the nonlinear jump points of the strain energy accumulation rate. When the node's cumulative mutation index exceeds a dynamically set threshold, it is marked as a potential crack initiation point, and isolated noise points are removed using a spatial clustering algorithm. Finally, crack coordinates with spatial location attributes are output. These coordinates are accurate to the sub-pixel level and include graded parameters for initiation intensity and propagation trend.

[0031] The damage response module 300 analyzes the diffusion persistence characteristics of high-frequency displacement gradients based on stress distribution maps, specifically including:

[0032] First, the displacement gradient amplitude of the high-risk zone boundary marked in the stress distribution map is extracted. Second, a gradient diffusion index time series analysis model is constructed. By calculating the spatial expansion rate and directional consistency of the gradient amplitude between consecutive image frames, the continuous diffusion characteristics of the yield risk zone are determined. A gradient diffusion dynamic model is established using the analogy method of the heat conduction equation. Finally, the material yield risk zone covering the entire weld area is output. The high-risk zone coordinates, main diffusion direction and risk level are marked in the material yield risk zone. The spatial positioning accuracy is strictly aligned with the displacement field grid.

[0033] The early warning module 400, based on the crack coordinates and material yield risk zone output by the damage response module 300, maps them to a sub-pixel grid coordinate system for spatial coupling analysis, specifically including:

[0034] First, a spatial classification model of the risk zone is established. Based on the gradient diffusion index threshold, the yield risk zone is divided into low-risk and high-risk zones. Gradient diffusion index data of the material's yield risk zone is obtained first. According to the preset gradient diffusion index threshold, the material's yield risk zone is divided into two spatial partitions: areas with a gradient diffusion index below the threshold are defined as low-risk transition zones, and areas with a gradient diffusion index exceeding the threshold are marked as high-risk core zones. This classification model achieves quantitative analysis of the spatial relationship between cracks and risk zones through grid coordinate mapping. Finally, the positional relationship between crack coordinate points and the polygonal boundary of the risk zone is calculated using a grid position matching algorithm.

[0035] First, the material yield risk zone is constructed as a geometric polygon boundary vector model in a sub-pixel grid coordinate system, and the crack coordinates are mapped to a set of grid points. A point set inclusion detection algorithm is used to determine the spatial relationship between crack points and the risk zone. When the crack point set is completely contained within the low-risk transition zone polygon and the proportion of its covered grid to the total number of grids in the risk zone exceeds a set proportion, the spatial overlap condition is satisfied and a local maintenance coordinate command is triggered. When the crack point set geometrically overlaps with the high-risk core zone polygon and the proportion of the overlapping area to the core zone grid area exceeds a set threshold, the start signal of the structural reinforcement device is activated. The local maintenance coordinate command includes the crack coordinate points and the associated grinding depth parameters. The execution of the structural reinforcement device is achieved by dynamically applying linear constraint forces in the overlapping area through a hydraulic mechanism.

[0036] When the crack detection status is null (i.e., the damage response module 300 does not output any crack coordinates), the diffusion principal direction vector and diffusion intensity scalar value output from the material yield risk zone are first extracted. Then, based on the spatial orientation of the diffusion principal direction, the axial distribution range of the heat treatment area in the weld sub-pixel grid coordinate system is determined. At the same time, the key parameters of the heat treatment temperature curve are dynamically set according to the diffusion intensity value. Finally, a decision output containing a spatially coordinated heat treatment scheme is generated. This scheme clearly marks the boundary line coordinates of the heat treatment area in the form of a grid coordinate set, and binds the corresponding heat treatment temperature curve and holding time parameters to each coordinate point, thereby forming a complete set of coordinated process parameters covering the risk diffusion area, realizing controllable intervention of material micro-damage.

[0037] As can be seen from the above description, the machine vision-based real-time monitoring system for fatigue cracking of steel structure welds provided in this embodiment has the following technical effects:

[0038] The image acquisition module 100 ensures a highly stable image source under strong interference environments; the displacement field reconstruction module 200 achieves a breakthrough by using frequency-spatial joint decomposition and material gradient adaptive filtering to simultaneously eliminate environmental vibration noise and welding non-uniformity interference, and the dual-channel technology outputs a physically meaningful anti-noise deformation field and a stress concentration distribution with micron-level precision; the damage response module 300 achieves sub-pixel-level precise location of crack initiation based on spatiotemporal analysis of strain energy mutation, and dynamically predicts the yield risk evolution path through a gradient diffusion dynamics model; the early warning module 400 intelligently triggers a coordinate-based hierarchical response mechanism based on the spatial coupling state between cracks and risk zones—locating and maintaining in low-risk zones, dynamically reinforcing in high-risk zones, and generating heat treatment coordinate schemes to actively repair material damage when there are no cracks. This system solves the problems of weak anti-interference ability, lagging single-parameter monitoring, and passive damage intervention in traditional methods, realizing full-cycle active protection of steel structure welds from micro-damage initiation to macroscopic failure prevention.

[0039] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A machine vision-based real-time monitoring system for fatigue cracking of steel structure welds, characterized in that, It includes an image acquisition module (100), a displacement field reconstruction module (200), a damage response module (300), and an early warning module (400), wherein: The image acquisition module (100) is used to continuously acquire dynamic image sequences of the weld surface; The displacement field reconstruction module (200) performs subpixel-level gridded displacement tracking on the dynamic image sequence of the weld surface, establishes displacement field data covering the weld area, extracts high-frequency displacement components through frequency domain-spatial domain joint constraint decomposition technology, and outputs two types of data channels simultaneously. The first channel generates the local deformation field after vibration interference is eliminated based on the elastic mechanical constitutive relation and material continuity constraint model, and the second channel directly analyzes the spatial distribution characteristics of the high-frequency displacement components to generate a stress concentration distribution map. The field reconstruction module (200) includes an extraction unit (201), and the extraction unit (201) extracts high-frequency displacement components through frequency domain-spatial domain joint constraint decomposition technology, specifically including: Based on the pre-built elasticity vibration response spectrum library, the inherent vibration characteristics of the structure are analyzed. The intrinsic mode decomposition technique is used to remove the fundamental frequency resonance related components. Spatial constraints are constructed through the elastic modulus gradient distribution function of the heat-affected zone of the weld. Then, the high-frequency displacement components are obtained through the displacement field continuity correction equation. The displacement field reconstruction module (200) includes a first channel unit (202), and the process of the first channel unit (202) generating a local deformation field specifically includes: A physical constraint filtering mechanism is constructed based on the constitutive relations of elasticity. The extracted high-frequency displacement components are input into the material continuity constraint model for processing. The material continuity constraint model dynamically adjusts the spatial smoothing coefficient according to the elastic modulus gradient distribution characteristics of the weld region. The high elastic modulus region at the weld center in the material continuity constraint model adopts a low smoothing strength, while the low elastic modulus region of the base material in the material continuity constraint model has an enhanced smoothing strength. The spatial consistency correction of the displacement field data is achieved by solving the displacement equilibrium equations of Hooke's law. The low-frequency vibration residual components are eliminated by synchronously coupled fundamental frequency resonant band-stop filtering technology, and the output is a spatially continuous and vibration noise-eliminated local deformation field. The displacement field reconstruction module (200) includes a second channel unit (203), and the process by which the second channel unit (203) generates a stress concentration distribution map specifically includes: The spatial gradient distribution characteristics of high-frequency displacement components are directly analyzed. An edge-sensitive spatial convolution algorithm is used to anisotropically process the displacement field data, including calculating the magnitude and principal direction of the displacement vector gradient through a composite operator of Gaussian weighted smoothing and directional gradient detection, and identifying microscopic discontinuous abrupt regions in the displacement field data. Based on the elastic-plastic mechanical yield criterion, a mapping model between the displacement gradient magnitude and the material stress concentration coefficient is established. When the local gradient magnitude exceeds the material yield critical threshold, it is marked as a high-risk area of ​​stress concentration, generating a two-dimensional stress distribution map covering the entire weld area. The damage response module (300) calculates the cumulative rate of distortion strain energy of the grid nodes based on local deformation field data, detects the crack initiation state through a spatiotemporal cumulative catastrophe algorithm to generate crack coordinates, and analyzes the diffusion persistence characteristics of high-frequency displacement gradient based on stress distribution map data to locate the material yield risk zone. The early warning module (400) maps the crack coordinates and the material yield risk zone to a sub-pixel grid coordinate system for spatial coupling analysis, wherein: When the crack coordinate is located in the low-risk zone of the material yield risk zone, a local maintenance coordinate command is triggered; when the crack coordinate overlaps with the high-risk zone of the material yield risk zone, the structural strengthening device is activated; when no crack coordinate is detected, the coordinates of the preventive heat treatment scheme are output.

2. The real-time monitoring system for fatigue cracking of steel structure welds based on machine vision according to claim 1, characterized in that, The elastic modulus gradient distribution function of the heat-affected zone of the weld dynamically describes the gradual change of material properties from the weld center to the base material based on the welding process parameters.

3. The machine vision-based real-time monitoring system for fatigue cracking of steel structure welds according to claim 1, characterized in that, The process of generating crack coordinates by the damage response module (300) specifically includes: Based on the local deformation field, the cumulative rate of distorted strain energy is calculated node by node in the global sub-pixel grid coordinate system of the weld. This includes converting strain components into instantaneous values ​​of distorted energy density according to the plane strain energy density theory; constructing a spatiotemporal cumulative monitoring window; performing abrupt change feature analysis on the strain energy density sequence of each grid node within a continuous monitoring period; using a wavelet transform modulus maxima detection algorithm to identify nonlinear jump points in the cumulative rate of strain energy; marking a node as a potential crack initiation point when the cumulative abrupt change index exceeds a dynamically set threshold; and removing isolated noise points using a spatial clustering algorithm to output crack coordinates with spatial location attributes.

4. The machine vision-based real-time monitoring system for fatigue cracking of steel structure welds according to claim 1, characterized in that, The process by which the damage response module (300) generates the material yield risk zone specifically includes: Based on the stress distribution map, the diffusion persistence characteristics of high-frequency displacement gradient are analyzed, including extracting the displacement gradient amplitude of the high-risk zone boundary marked in the stress distribution map; constructing a gradient diffusion index time series analysis model, and judging the persistence diffusion characteristics of the yield risk zone by calculating the spatial expansion rate and directional consistency of the gradient amplitude between consecutive image frames; and establishing a gradient diffusion dynamic model by using the heat conduction equation analogy method, outputting the material yield risk zone covering the entire weld area.

5. The machine vision-based real-time monitoring system for fatigue cracking of steel structure welds according to claim 1, characterized in that, The early warning module (400) maps the crack coordinates and the material yield risk zone to the sub-pixel grid coordinate system for spatial coupling analysis, including establishing a spatial classification model of the risk zone and dividing the material yield risk zone into low-risk zone and high-risk zone according to the gradient diffusion index threshold.

6. The real-time monitoring system for fatigue cracking of steel structure welds based on machine vision according to claim 1, characterized in that, The early warning module (400) calculates the positional relationship between the crack coordinate point and the polygonal boundary of the risk area through a grid position matching algorithm. When the crack coordinate is completely contained within the low-risk area and the spatial overlap exceeds a set ratio, a local maintenance coordinate command is triggered. When the crack coordinate has geometric overlap with the high-risk area and the area of ​​the overlap area exceeds a set ratio of the total area of ​​the core area, the structural reinforcement device is automatically activated.

7. The machine vision-based real-time monitoring system for fatigue cracking of steel structure welds according to claim 1, characterized in that, When the early warning module (400) does not detect the crack coordinates, it generates the coordinates of the preventive heat treatment scheme based on the main diffusion direction and diffusion intensity of the material yield risk zone, and outputs a set of coordinateized process parameters including the heat treatment temperature curve and the range of action.

Images