Multiaxial fatigue damage assessment method and apparatus for steel bridge welds considering residual welding stress

Abstract

This invention provides a method and apparatus for multiaxial fatigue damage assessment of steel bridge welds considering residual welding stress, relating to the field of fatigue strength analysis and life prediction technology for steel bridge structures. The invention employs X-ray measurement of three-dimensional residual stress in the geometric transition region of the steel bridge weld, combined with infrared thermography and structured light scanning to acquire thermal distribution and surface morphology data, achieving spatial alignment of multi-source data based on a unified coordinate system. Secondly, random perturbations are applied to the residual stress to simulate stress fluctuations during service, constructing robust weighted risk factors. A dual-modal deep neural network is constructed, adaptively fusing infrared thermography and three-dimensional weld morphology information through an attention mechanism. The robust weighted risk factors are used as a supervisory signal to train a fatigue risk regression model, predicting risk factors at each location. Based on preset thresholds, low, medium, and high risk levels are defined, achieving accurate graded assessment of weld fatigue damage.

Description

Technical Field

[0001] This invention relates to the field of fatigue strength analysis and life prediction technology for steel bridge structures, specifically to a method and apparatus for assessing multiaxial fatigue damage of steel bridge welds considering welding residual stress. Background Technology

[0002] Currently, fatigue performance assessment of Q345 steel in bridge welded structures commonly employs methods such as nominal stress, hot spot stress, or local stress / strain. However, under multiaxial stress conditions, especially when complex residual stress distributions and geometric discontinuities exist at the weld toe and bevel of the weld, traditional methods often fail to accurately reflect the fatigue damage risk under actual service conditions. Particularly during the service of steel bridges, welding residual stress, as one of the key driving factors for fatigue crack initiation, exhibits strong spatial inhomogeneity and uncertainty in its spatial distribution.

[0003] Existing assessment methods typically treat residual stress as a deterministic input, neglecting the influence of measurement errors, local material heterogeneity, and stress relaxation after service loading on fatigue response. This assumption of a single-point deterministic value easily leads to large fluctuations and error sensitivity in the assessment results of hazard factors in actual structures, thus affecting the accurate prediction of fatigue life and the reliability of risk classification. Furthermore, due to the significant multiaxial stress state and dynamic load effects in the weld region, current fatigue damage assessment methods still have limitations in applicability in integrating multiaxial fatigue criteria and complex residual stress fields, making it difficult to cover the fatigue vulnerability identification needs under all working conditions.

[0004] Therefore, there is an urgent need for a multiaxial fatigue assessment method that can fully consider the spatial variation characteristics and uncertainties of welding residual stress to improve the accuracy of fatigue risk identification in the weld area of ​​steel bridges and the robustness of engineering applications.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method and apparatus for assessing multiaxial fatigue damage of steel bridge welds that takes into account welding residual stress, so as to solve the problems mentioned in the background art.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A multiaxial fatigue damage assessment method for steel bridge welds considering residual welding stress includes the following steps:

[0009] Step 1: The geometric transition region of the steel bridge weld area is measured with X-ray to obtain a three-dimensional residual stress, generate an infrared thermal image and a three-dimensional morphology image of the weld surface, and perform normalization processing. The geometric transition region includes the weld toe tip, the center point of the bevel bottom and the base material reference point.

[0010] Step 2: Apply a zero-mean Gaussian perturbation to the residual stress, construct a perturbation sample set through multiple rounds of Monte Carlo calculation, perform multiple rounds of hazard factor calculation on the perturbation sample using the Dang Van criterion, and construct a robust weighted hazard factor based on the hazard factor.

[0011] Step 3: Using the infrared thermal image and the three-dimensional morphology image of the weld as dual-modal inputs, the thermal distribution features and geometric morphology features of the weld are extracted by a deep convolutional neural network, and the features are fused by a shared attention mechanism. The robustness weight risk factor of the geometric transition region is used as a supervision signal to construct a nonlinear regression model of fatigue risk. The deep convolutional neural network is a dual-branch parallel convolutional neural network.

[0012] Step 4: Infrared thermal images and three-dimensional topographic images of the weld surface are obtained in the weld area by sliding sampling through an image window. These are used as inputs to the fatigue risk nonlinear regression model, and the risk factors at each window position are output to classify the fatigue damage in the weld area.

[0013] Furthermore, the method for measuring the three-dimensional residual stress in the geometric transition region of the steel bridge weld area using X-rays is as follows:

[0014] A spatial rectangular coordinate system is established with the intersection of the geometric center line of the weld section and the base metal as the origin O, the longitudinal direction of the weld as the x-axis, the direction perpendicular to the plate thickness as the y-axis, and the direction perpendicular to the weld cross section as the z-axis. The geometric transition region is defined as follows: the range within 0.5 mm of the geometric tip of the fusion line between the weld and the base metal is taken as the weld toe tip point; the range within 0.3 mm of the root of the weld, i.e., the lowest point where the fusion zone contacts the base metal, is taken as the bottom center point of the groove; and the base metal surface 30 mm away from the edge of the weld along the z-axis is taken as the base metal reference point.

[0015] The measurement point grid for the weld toe tip is divided as follows: within 0.5 mm from the weld toe tip, a 5×5 equally spaced grid is used, with a point spacing of 0.5 mm, for a total of 25 measurement points; the measurement point grid for the center point of the bevel bottom is divided as follows: centered on the lowest point where the fusion zone contacts the base material, a 3×3 equally spaced grid is used, with a point spacing of 0.3 mm, for a total of 9 measurement points; the measurement point grid for the base material reference point is divided as follows: in this area, a 2×3 equally spaced grid is used, with a point spacing of 5 mm, for a total of 6 measurement points; three-dimensional residual stress is measured at each measurement point using an X-ray diffractometer to obtain normal stress and shear stress in MPa. Diffraction data are collected at three offset angles of -20°, 0°, and +20° for each measurement point to construct the stress tensor matrix.

[0016]

[0017] In the formula, σ represents the stress tensor matrix, σ x σ represents the normal stress along the x-axis. y σ represents the normal stress along the y-axis. z τ represents the normal stress along the z-axis. xy τ represents the shear stress in the xy plane. yz τ represents the shear stress in the yz plane. xz This represents the shear stress in the xz plane.

[0018] Furthermore, the method for generating infrared thermal images and three-dimensional morphology images of the weld surface, and performing normalization processing, is as follows:

[0019] The preset heating temperature is T yr And 60≥T yr ≥40, after heating the steel bridge weld to the preset heating temperature, the spatial resolution of the high-resolution infrared thermal imager is set to 0.1mm. A vertical scan along the z-axis is performed at a distance of 0.3m from the weld surface, acquiring infrared radiation data in the 7.5-13μm band, generating a 640×512 pixel infrared thermal image. The coordinate axes correspond to the physical spatial location of the weld area, and the pixel values ​​correspond to the surface temperature. The temperature is normalized, and the output is an infrared thermal image in the form of a two-dimensional heat distribution matrix. The formula used for temperature normalization is:

[0020]

[0021] In the formula, T wd T represents the surface temperature. norm This indicates the normalized surface temperature;

[0022] A structured light 3D scanner was used to scan the weld surface. The scanner was moved perpendicular to the weld surface along the z-axis, covering the entire length of the weld. The scanning resolution was set to 0.1 mm / point to acquire point cloud data of the weld surface, which was then converted into a 640×512 pixel 3D weld topography image. The pixel coordinates corresponded one-to-one with the actual spatial coordinates of the weld. The x-axis and z-axis in the coordinate system corresponded to the row and column directions of the 3D weld topography image matrix, respectively. The coordinate system of the 3D weld topography image was constructed by dividing the image into a grid with a spatial resolution of 0.1 mm. The physical spatial coordinates of each pixel P(a, b) in the 3D weld topography image are: (x, z) = (0.1a, 0.1b). This coordinate system uses the same resolution and alignment as the infrared thermal image, ensuring that the pixel positions in the image correspond one-to-one. The height of each coordinate was normalized.

[0023] In the formula, H norm H represents the height after normalization, where H represents the weld surface height value. max H represents the maximum value of the surface height of all welds. min This represents the minimum value of the surface height of all weld seams.

[0024] Furthermore, the method for constructing a perturbation sample set by applying a zero-mean Gaussian perturbation to the residual stress and performing multiple rounds of Monte Carlo calculations is as follows:

[0025] A zero-mean Gaussian perturbation is introduced into the stress tensor of the residual stress at each measuring point to generate a perturbation tensor:

[0026] σ′=σ+Δσ

[0027] In the formula, σ′ represents the perturbation tensor, and Δσ represents the perturbation matrix, wherein the elements of the perturbation matrix are generated by the following method:

[0028] Generate a standard normal random variable ξ using the Mersenne-Twister algorithm. de ~N(0,1), generate perturbation matrix elements:

[0029]

[0030] In the formula, η represents the disturbance strength coefficient, which is taken as 5%-10% of the yield strength of the steel bridge material. σ represents the perturbation matrix element generated in the k-th simulation. ij Let N represent the elements in the stress tensor matrix, where i represents the row index, j represents the column index, d represents the row index, and e represents the column index. For each measurement point, N elements are generated repeatedly. srd Group perturbation tensors to form a perturbation sample set. Where, N srd It is a positive integer greater than 100.

[0031] Furthermore, the method of using the Dang Van criterion to calculate hazard factors for perturbed samples in multiple rounds, and constructing robust weighted hazard factors based on these hazard factors, is as follows:

[0032] For each perturbation tensor in the perturbation sample set, after applying a sinusoidal external load along the x-axis, N is calculated according to the Dang Van criterion. srd There are risk factors, where the risk factor obtained from the k-th simulation is denoted as . Sort in ascending order and take the first... Each value serves as a robustness weight risk factor.

[0033] Furthermore, using infrared thermal images and three-dimensional weld morphology images as dual-modal inputs, the method for extracting weld thermal distribution features and geometric morphology features respectively through a deep convolutional neural network is as follows:

[0034] Set a 64×64 pixel sliding window with a sliding step of 16 pixels. Slide 40 times along the x-axis and 32 times along the z-axis. Use zero-value filling to make the edge window complete. A total of 1280 windows are generated. For each window, extract the physical space coordinates corresponding to its center pixel and determine whether it is located in the geometric transition area of ​​the weld area. Only windows whose center point falls into the geometric transition area are retained as valid training samples.

[0035] For each valid sample, the nearest neighbor interpolation method is used to find the robustness weight risk factor corresponding to the nearest measurement point, based on the physical coordinates of the window center, and this factor is used as the supervision label value y of that window. ture The 64×64 pixel matrix cropped from the infrared thermal image is denoted as T. patch The 64×64 pixel matrix extracted from the 3D topography of the weld is denoted as H. patch ;

[0036] A dual-branch parallel convolutional neural network is used to extract features from both the infrared thermal image and the 3D topography image. The specific structure is as follows:

[0037] Input layer: Receives a 64×64×1 single-channel image; Convolutional layer of the first convolutional block: 32 3×3 convolutional kernels, "same" padding, ReLU activation function, batch normalization layer to standardize channel mean and variance, max pooling layer with 2×2 pooling kernel, stride 2, outputting a 32×32×32 feature map; Convolutional layer of the second convolutional block: 64 3×3 convolutional kernels, "same" padding, ReLU activation function, batch normalization layer to standardize channel mean and variance, max pooling layer with 2×2 pooling kernel, stride 2, outputting a 16×16×64 feature map; Flattening layer: Flattens the 3D feature map into a 16384-dimensional feature vector.

[0038] Furthermore, the shared attention mechanism fuses features, using the robustness weight risk factor of the geometric transition region as a supervisory signal to construct a nonlinear regression model for fatigue risk.

[0039] The shared attention mechanism concatenates the 16384-dimensional vector from the two branches into a 32768-dimensional vector. Fully connected layer 1 has 1024 neurons with ReLU activation; fully connected layer 2 has 512 neurons with ReLU activation; fully connected layer 3 has 2 neurons with Sigmoid activation; and the output hot weight α... T and shape weight α H , and α T +α H =1;

[0040] A regression model is constructed using a fully connected network and Dropout layers to output predicted risk factors. The steps for constructing the regression model using a fully connected network and Dropout layers are as follows:

[0041] Fully connected layer 1: 2048 neurons, ReLU activation function; Dropout layer: 25% dropout rate; Fully connected layer 2: 1024 neurons, ReLU activation function; Dropout layer: 25% dropout rate; Fully connected layer 3: 1 neuron, linear activation function;

[0042] The training learning rate is 1×10 -3 The optimizer is Adam, the training epochs are 200, Dropout is 0.25, and the batch size is 32, meaning 32 samples are randomly selected from the valid training samples. The main loss function in the current batch is:

[0043]

[0044] In the formula, L represents the main loss function. Let y represent the hazard factors predicted by the output for i samples. ture,iLet represent the supervision label value of the i-th sample, where i represents the index of the 32 randomly selected samples, i = 1, 2, ..., 32.

[0045] Furthermore, the method for classifying fatigue damage in the weld area by outputting the hazard factors at each window location is as follows:

[0046] For steel bridge welds requiring damage assessment, the hazard factors output by each sliding window are extracted, constructing a 40×32 two-dimensional risk matrix that corresponds one-to-one with the physical coordinates of the weld. Thresholds R1 and R2 for the hazard factors are pre-set, where R1>R2>0. For each hazard factor output by the sliding window... When satisfied When the area is classified as a low-risk damage area, and the following conditions are met... At that time, it was determined to be a medium-risk damage area, when the following conditions are met. When i′ is identified as a high-risk damage area, i′ represents the index of the sliding window, i′ = 1, 2, ..., 1280.

[0047] Additionally, a multiaxial fatigue damage assessment device for steel bridge welds considering welding residual stress is provided, characterized in that: the device is used to perform the aforementioned multiaxial fatigue damage assessment method for steel bridge welds considering welding residual stress, including:

[0048] The image generation module is used to measure the three-dimensional residual stress in the geometric transition area of ​​the steel bridge weld area using X-rays, generate an infrared thermal image and a three-dimensional morphology image of the weld surface, and perform normalization processing. The geometric transition area includes the weld toe tip, the center point of the bevel bottom, and the base material reference point.

[0049] The robust hazard factor module is used to apply zero-mean Gaussian perturbation to the residual stress. It constructs a set of perturbation samples through multiple rounds of Monte Carlo calculations, performs multiple rounds of hazard factor calculations on the perturbation samples using the Dang Van criterion, and constructs robust weighted hazard factors based on the hazard factors.

[0050] The fatigue risk regression module is used to extract the thermal distribution features and geometric features of the weld seam through a deep convolutional neural network, using infrared thermal imaging and weld seam three-dimensional morphology as dual-modal inputs. The features are fused by a shared attention mechanism, and the robustness weight risk factor of the geometric transition region is used as a supervision signal to construct a nonlinear regression model for fatigue risk. The deep convolutional neural network is a dual-branch parallel convolutional neural network.

[0051] The weld risk classification module is used to acquire infrared thermal images and three-dimensional morphology images of the weld surface in the weld area through image window sliding sampling. These images are used as inputs to the fatigue risk nonlinear regression model, and the module outputs the hazard factors at each window position to classify the fatigue damage in the weld area.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] This invention utilizes X-ray diffraction to measure the three-dimensional residual stress in the geometric transition region of steel bridge welds. Through Gaussian perturbation and multiple rounds of Monte Carlo simulation, it dynamically corrects the input parameters of the Dang Van criterion to calculate robustness weight risk factors, which are then used as supervision signals for a deep convolutional neural network. This constructs a nonlinear regression model for fatigue risk based on thermal distribution and geometric features, avoiding the fatigue damage prediction bias caused by the traditional Dang Van criterion's assumption of stable loads. It also overcomes the limitations of single stress or geometric data in fatigue identification. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0055] Figure 2 This is a fitting diagram of the disturbance, normal stress, and hazard factor in this invention;

[0056] Figure 3 This is a fitting graph of the D parameters and hazard factors in this invention;

[0057] Figure 4 This is a risk factor prediction and assessment chart for the present invention;

[0058] Figure 5 This is a schematic diagram illustrating the prediction accuracy of the present invention;

[0059] Figure 6 This is the topography weight output diagram of the present invention;

[0060] Figure 7 This is the thermal weight output diagram of the present invention;

[0061] Figure 8 This is a schematic diagram of the overall device structure of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0063] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0064] Example:

[0065] Please see Figures 1 to 7 The present invention provides a technical solution:

[0066] A multiaxial fatigue damage assessment method for steel bridge welds considering residual welding stress includes the following steps:

[0067] Step 1: The geometric transition region of the steel bridge weld area is measured with X-ray to obtain a three-dimensional residual stress, generate an infrared thermal image and a three-dimensional morphology image of the weld surface, and perform normalization processing. The geometric transition region includes the weld toe tip, the center point of the bevel bottom and the base material reference point.

[0068] A spatial rectangular coordinate system is established with the intersection of the weld cross-section geometric centerline and the base metal as the origin O, the longitudinal direction of the weld as the x-axis, the direction perpendicular to the plate thickness as the y-axis, and the direction perpendicular to the weld cross-section as the z-axis. This allows data from different sources, such as residual stress measured by X-rays, infrared thermography, and three-dimensional morphology scanning, to be aligned based on the same coordinate system, thus enabling fatigue damage assessment of steel structures. The geometric transition region is defined as the weld toe tip, defined as the area within 0.5 mm of the geometric tip of the weld-base metal fusion line. The weld toe is the geometric transition point between the weld and the base metal, exhibiting a significant abrupt shape change and being the most likely location for fatigue crack initiation. Studies show that the region within 0.5 mm of the weld toe tip is prone to stress concentration. The high number of residual stresses and large residual stress gradients significantly affect the fatigue life. The center point of the groove bottom is defined as 0.3mm from the lowest point of contact between the weld root and the base metal. Defects such as incomplete fusion and slag inclusions are prone to occur at the weld root during welding, and geometric features such as blunt edges and root gaps lead to stress concentration. The 0.3mm range can better capture the stress abrupt changes corresponding to minor geometric defects at the root. The base metal surface 30mm from the weld edge along the z-axis is used as the base metal reference point. This is because engineering experience generally considers the stress state of the base metal 20 to 50mm from the weld edge to be close to its original state, making it effective for comparing stress anomalies in the weld area.

[0069] The measuring point grid at the weld toe tip is divided as follows: within 0.5 mm from the weld toe tip, a 5×5 equally spaced grid is used, with a point spacing of 0.5 mm, for a total of 25 measuring points. The geometric change at the weld toe leads to a high stress concentration factor, which can reach 3-5. The 0.5 mm point spacing can capture 2-3 stress gradient changes within a single grid, avoiding stress characteristic distortion and thus avoiding an overly optimistic fatigue life assessment.

[0070] The measuring point grid at the center point of the bottom of the bevel is divided as follows: taking the lowest point of the fusion zone in contact with the base material as the center, a 3×3 equally spaced grid is used, with a point spacing of 0.3mm, for a total of 9 measuring points. The stress concentration factor at the root is about 2-3, and the stress gradient is lower than that at the weld toe. However, due to the possibility of non-fusion defects, a point spacing of 0.3mm is required to capture the stress change within a range of 0.5mm around the defect.

[0071] The measuring point grid of the base material reference point is divided as follows: a 2×3 equally spaced grid is used in this area, with a point spacing of 5mm and a total of 6 measuring points. The residual stress at 30mm from the weld has decayed to a very small value. The 5mm point spacing is sufficient to characterize the uniform stress field and reduce redundant data.

[0072] Three-dimensional residual stress measurements were performed at each measurement point using an X-ray diffractometer, such as RigakuSmartLab. The X-ray source was a Cu target Kα, with a spot size of 0.1 mm × 0.1 mm, a tube voltage of 40 kV, a tube current of 100 mA, and an exposure time of 30 s per angle. Normal and shear stresses were acquired in MPa. For each measurement point, nine diffraction patterns were collected at azimuth angles φ = 0°, 45°, and 90°, and at tilt angles ψ = -20°, 0°, and +20°. All diffraction patterns were imported into PDXL3 software, and the stress tensor matrix was output using the three-dimensional stress tensor mode of the stress calculation wizard.

[0073]

[0074] In the formula, σ represents the stress tensor matrix, σ x σ represents the normal stress along the x-axis. y σ represents the normal stress along the y-axis. z τ represents the normal stress along the z-axis. xy τ represents the shear stress in the xy plane. yz τ represents the shear stress in the yz plane. xz This represents the shear stress in the xz plane.

[0075] The preset heating temperature is T yr And 60≥T yr≥40℃, by stimulating the thermal response of the weld area through low-temperature heating, the internal stress state of the material is coupled with different temperatures, providing a basis for subsequent infrared thermal imaging detection based on temperature differences. This avoids damage to the microstructure of the steel bridge material caused by high temperatures. The recrystallization temperature of Q345 steel is approximately 600℃, and 40-60℃ is far below its phase transformation temperature, preventing a decrease in material strength or lattice distortion. Engineering experience shows that the thermal expansion effect of steel bridges is significant within this temperature range, facilitating the capture of minute temperature differences and thus avoiding a decrease in weld hardness or release of residual stress caused by high-temperature heating. After heating the steel bridge weld to the preset heating temperature, the spatial resolution of the high-resolution infrared thermal imager is set to 0.1mm, and a vertical scan is performed along the z-axis at a distance of 0.3m from the weld surface, collecting infrared radiation data in the 7.5-13μm band. According to reports, setting the spatial resolution of the high-resolution infrared thermal imager to 0.1 mm is to ensure a one-to-one correspondence between temperature data and subsequent three-dimensional morphology data in physical space. The diffraction effect of infrared light in the 7.5-13 μm band is weak, achieving sub-millimeter resolution at a distance of 0.3 m, meeting the requirements for detecting temperature anomalies in weld micro-defects, such as micro-cracks and stress concentration areas. A 640×512 pixel infrared thermal image is generated, with the coordinate axes corresponding to the physical spatial location of the weld area and the pixel values ​​corresponding to the surface temperature. Temperature is normalized to map the temperature data to the [0,1] interval, eliminating the influence of temperature dimensions and facilitating comparison of multiple sets of data and input to machine learning models. The output is an infrared thermal image in the form of a two-dimensional heat distribution matrix. The formula used for temperature normalization is:

[0076]

[0077] In the formula, T wd T represents the surface temperature. norm This indicates the normalized surface temperature;

[0078] A structured light 3D scanner was used to scan the weld surface. The scanner was moved perpendicular to the weld surface along the z-axis, covering the entire length of the weld. The scanning resolution was set to 0.1 mm / point. The weld reinforcement of Q345 steel bridge welds is typically ≤3 mm, while surface defects, such as undercut depth, can exceed 0.5 mm. A resolution of 0.1 mm can capture the size of these defects, meeting the accuracy requirements of engineering inspection. Point cloud data of the weld surface was acquired and converted into a 640×512 pixel 3D weld topography image. The pixel coordinates correspond one-to-one with the actual spatial coordinates of the weld, ensuring that the same physical location... Temperature and height data can be analyzed simultaneously. The x-axis and z-axis in the coordinate system correspond to the row and column directions of the weld 3D topography matrix, respectively. A grid is created with a spatial resolution of 0.1 mm to construct the coordinate system for the weld 3D topography. The physical spatial coordinates of each pixel P(a, b) in the weld 3D topography are: (x, z) = (0.1a, 0.1b). This coordinate system uses the same resolution and alignment as the infrared thermal imaging image, ensuring a one-to-one correspondence between pixel positions in space. The height of each coordinate is then normalized.

[0079] In the formula, H norm H represents the height after normalization, where H represents the weld surface height value. max H represents the maximum value of the surface height of all welds. min This represents the minimum value of the surface height of all welds. The overall height of different welds varies greatly. After normalization, the height data is unified to the [0,1] interval, eliminating the influence of temperature dimensions and facilitating comparison of multiple sets of data and input for machine learning models.

[0080] Step 2: Apply a zero-mean Gaussian perturbation to the residual stress, construct a perturbation sample set through multiple rounds of Monte Carlo calculation, perform multiple rounds of hazard factor calculation on the perturbation sample using the Dang Van criterion, and construct a robust weighted hazard factor based on the hazard factor.

[0081] The residual stress in steel bridge welds is affected by a triple factor: measurement error, material dispersion, and service disturbance (vehicle load fluctuation). Static simulations, based on historical data or experimental modeling, typically employ multiple measurements at a single point, failing to reproduce the non-uniformity of the three-dimensional stress field along the entire weld length. Traditional experiments only use constant-amplitude sinusoidal loading, lacking a stochastic process. By introducing a zero-mean Gaussian perturbation into the stress tensor of the residual stress at each measurement point, the aforementioned uncertainties can be transformed into stochastic corrections to the stress tensor, making fatigue risk assessment a probabilistic analysis and avoiding misjudgments due to neglecting non-systematic biases. The formula used to generate the perturbation tensor is:

[0082] σ′=σ+Δσ

[0083] In the formula, σ′ represents the perturbation tensor, and Δσ represents the perturbation matrix, wherein the elements of the perturbation matrix are generated by the following method:

[0084] By using MATLAB's randn function, which incorporates the Mersenne Twister algorithm—an industrial-grade pseudo-random number algorithm—simulation biases caused by repeated random numbers are avoided, thus generating a standard normal random variable ξ. de ~N(0,1), the formula used to generate the elements of the perturbation matrix is:

[0085]

[0086] In the formula, η represents the disturbance strength coefficient, which, based on engineering experience, is taken as 5%-10% of the yield strength of the steel bridge material. For example, the yield strength of Q345 steel bridge material is 345 MPa, and the value of η is in the range of 17.25-34.5 MPa. We set the value of η to 20 MPa. Let represent the elements of the perturbation matrix generated in the k-th simulation, d represent the row index of the perturbation matrix, and e represent the column index of the perturbation matrix. For example, if σ 22 =280MPa, ξ 22 =0.75, then Then, by synthesizing the perturbation tensor using σ′=σ+Δσ, N is repeatedly generated for each measurement point. srd Group perturbation tensors to form a perturbation sample set. Where, N srd It is a positive integer greater than 100.

[0087] For each perturbation tensor in the perturbation sample set, after applying a sinusoidal external load along the x-axis, N is calculated according to the Dang Van criterion. srd Based on the principles of mechanics of materials, 35% of the yield strength of Q345 steel is taken as the external load amplitude to ensure that the principal stress of the combined stress remains within the elastic range. The alternating stress caused by vehicle load is simulated, with an external load amplitude of 120 MPa. Statistics from the bridge dynamic weighing system show that the longitudinal stress amplitude borne by typical steel bridge deck welds is 80-150 MPa. 80 MPa corresponds to light vehicle traffic, and 150 MPa corresponds to heavy truck overload conditions. Taking the intermediate value of 120 MPa, it can cover 85% of conventional traffic loads. The load frequency is set to 0.5 Hz, corresponding to the first-order vertical vibration frequency of the bridge. The time history length is set to 0-4 s, corresponding to two complete cycles. The time step is 0.02 s, satisfying the Nyquist sampling theorem, thus capturing the load peak. The sinusoidal external load along the x-axis is represented as a three-dimensional stress tensor.

[0088]

[0089] In the formula, σext (t) represents the time-varying external stress tensor, σ amp Let f represent the amplitude of the external load, and f represent the frequency of the sinusoidal external load, taken as 0.5Hz. In engineering measured data, the first-order vertical vibration frequency of typical steel bridges is mostly in the range of 0.1-1.0Hz. Taking 0.5Hz is the middle value of this range, which can cover the typical working conditions of most steel bridges and is the typical frequency of the first-order vertical vibration of steel bridges. For example, at t=1s, sin(2πft)=0, and the three-dimensional stress tensor is 0. At t=0.5s, sin(2πft)=1, and the three-dimensional stress tensor is 120MPa. Among them, for the k-th perturbation tensor, the resultant stress at time t is:

[0090]

[0091] In the formula, σ represents the combined stress at time t. ′(k) This represents the k-th perturbation tensor; for example, When t = 0.5s, the combined stress is:

[0092]

[0093] right Solve the characteristic equation:

[0094] det(σ total -σ·I)=0

[0095] In the formula, σ total Represents the combined stress, and I represents the identity matrix, i.e. The three principal stresses σ1≥σ2≥σ3 are obtained. These three principal stresses reflect the most dangerous stress state after the residual stress is coupled with the external load.

[0096] Calculate the mean hydrostatic pressure:

[0097]

[0098] In the formula, P mean The average hydrostatic pressure is essentially a linear weighted average of the principal stresses, transforming the complex three-dimensional stress state into a single quantitative index. It characterizes the average volumetric stress state of a material element under a three-dimensional principal stress field. The larger the value of the term σ1+σ2+σ3, the greater the stress. mean The value increases accordingly. In practice, when the average hydrostatic pressure is greater than 0, it will promote dislocation movement and crack propagation. An increase in any principal stress will lead to an increase in hydrostatic pressure. All three are positively correlated with the average hydrostatic pressure.

[0099] Calculate the maximum shear stress:

[0100]

[0101] In the formula, τ max The maximum shear stress is directly determined by the difference between the maximum and minimum principal stresses. It reflects the maximum shear failure tendency that may occur inside the material. In the mechanics of materials, when an object is subjected to a complex stress state, shear stress will cause the material to slip or shear fracture. Its physical meaning is one of the critical indicators of the material's resistance to shear failure. The value of the maximum shear stress increases with the increase of the difference between σ1 and σ3, and is positively correlated with σ1 and negatively correlated with σ3.

[0102] For Q345 steel, fatigue testing was conducted to fit the hydrostatic pressure sensitivity coefficient, with a recommended standard value of 0.15-0.2. The hydrostatic pressure sensitivity coefficient was set at α′ = 0.18. Based on extensive testing, the recommended shear fatigue limit value τ was determined. DV =115MPa, calculate the Dang Van parameter:

[0103] D = τ max +α′·|P mean |

[0104] In the formula, D represents the Dang Van parameter;

[0105] Calculate risk factors:

[0106]

[0107] Where R DV The risk factor is represented by directly quantifying the fatigue failure risk of the structure by coupling shear stress and hydrostatic pressure effects. The risk factor obtained from the k-th simulation is expressed as follows: Sort them in ascending order. For example, after sorting the principal stresses of a certain instantaneous stress in ascending order, σ1 = 400 MPa, σ2 = 150 MPa, σ3 = 120 MPa, calculate the average hydrostatic pressure:

[0108]

[0109] Calculate the maximum shear stress:

[0110]

[0111] Calculate the Dang Van parameters:

[0112] D = τ max +α′·|P mean |=180.2MPa

[0113] Calculate risk factors:

[0114]

[0115] Then, for risk factors sorted in ascending order, take the first... Each value serves as a robustness weight risk factor. In the random disturbance simulation, 95% of the hazard factor values ​​are less than this value, and only 5% of the values ​​may be larger, thus providing a reasonable safety margin for structural fatigue design. Table 1 shows the hazard factor output table. By applying a zero-mean Gaussian perturbation to the stress tensor and then superimposing the principal stress of the combined stress obtained by the sinusoidal external load, the coupling effect of residual stress fluctuation and vehicle load is simulated.

[0116]

[0117] like Figure 2-Figure 3 As shown, the Dang Van parameter and the hazard factor R DV The fitting relationship shows a linear relationship, with the scatter points closely fitting the fitting line. The coupling effect of shear stress and hydrostatic pressure can be quantified by the Dang Van parameter. After applying a zero-mean Gaussian perturbation to the x-axis, as the normal stress on the x-axis increases, the principal stress of the combined stress and the hydrostatic pressure also increase, and the Dang Van parameter also increases accordingly. The simulated vehicle alternating load is closer to the real level, providing a data basis for the quantification and extraction of hazard factors.

[0118] Step 3: Using the infrared thermal image and the three-dimensional morphology image of the weld as dual-modal inputs, the thermal distribution features and geometric morphology features of the weld are extracted by a deep convolutional neural network, and the features are fused by a shared attention mechanism. The robustness weight risk factor of the geometric transition region is used as a supervision signal to construct a nonlinear regression model of fatigue risk. The deep convolutional neural network is a dual-branch parallel convolutional neural network.

[0119] A 64×64 pixel sliding window is set with a sliding step of 16 pixels. It slides 40 times along the x-axis and 32 times along the z-axis. Zero-value filling is used to make the edge window complete, generating a total of 1280 windows to capture the microscopic features of stress concentration. For each window, the physical space coordinates corresponding to its center pixel are extracted, and it is determined whether it is located in the geometric transition area of ​​the weld area. Only windows whose center point falls into the geometric transition area are retained as valid training samples. By using the sliding window, the continuous thermal image and topography image are segmented into local feature blocks, so that the model can focus on the subtle anomalies of high-risk areas such as weld toe and bevel, avoid global feature averaging from masking local defects, filter out most low-risk areas, and improve the correlation between samples and fatigue risk.

[0120] For each valid sample, the nearest neighbor interpolation method is used to find the robustness weight risk factor corresponding to the nearest measurement point, based on the physical coordinates of the window center, and this factor is used as the supervision label value y of that window. tureTo solve the spatial matching problem between image pixels and discrete measurement points, and to establish a mapping relationship between thermal distribution, morphology, and fatigue risk, the 64×64 pixel matrix extracted from the infrared thermal image is denoted as T. patch The 64×64 pixel matrix extracted from the 3D topography of the weld is denoted as H. patch A dual-branch parallel convolutional neural network is used to extract features from both the infrared thermal image and the 3D topography image. The two branches independently learn complementary features. By monitoring the temperature rise of the thermoelastic effect in the stress concentration area and geometric defects, the correlation with fatigue risk can be found. The specific structure is as follows:

[0121] Input layer: Receives a 64×64×1 single-channel image; First convolutional block: 32 3×3 convolutional kernels, "same" padding, ReLU activation function. The 3×3 kernels cover 3×3 pixels to capture micro-cracks and other small defects. Batch normalization layers are used to standardize the channel mean and variance. Max pooling layers use 2×2 pooling kernels with a stride of 2, outputting a 32×32×32 feature map; Second convolutional block: 64 3×3 convolutional kernels, "same" padding, ReLU activation function. Batch normalization layers are used to standardize the channel mean and variance. Max pooling layers use 2×2 pooling kernels with a stride of 2, outputting a 16×16×64 feature map; Flattening layer: Flattens the 3D feature map into a 16384-dimensional feature vector.

[0122] A shared attention mechanism is employed to adaptively allocate dual-modal feature weights. For example, when there is a significant temperature gradient in the thermal image, the model automatically increases α. T To enhance the contribution of heat distribution to fatigue risk, the 16384-dimensional vector of the two branches is concatenated to a 32768-dimensional vector. Fully connected layer 1 has 1024 neurons with ReLU activation; fully connected layer 2 has 512 neurons with ReLU activation; and fully connected layer 3 has 2 neurons with Sigmoid activation. The output heat weight α is used. T and shape weight α H , and α T +α H =1, using Sigmoid activation to ensure the weights are in the [0,1] interval, representing the importance ratio between thermal imaging and topographic features. For example, when topographic features dominate, α might be 1. H =0.7; when temperature anomaly characteristics dominate, α may be... T =0.6;

[0123] A regression model is constructed using a fully connected network and Dropout layers to output predicted risk factors. The steps for constructing the regression model using a fully connected network and Dropout layers are as follows:

[0124] The fully connected layer 1 has 2048 neurons and ReLU activation function; the dropout layer has a dropout rate of 25%; the fully connected layer 2 has 1024 neurons and ReLU activation function; the dropout layer has a dropout rate of 25%; and the fully connected layer 3 has 1 neuron and a linear activation function. The purpose of the dropout layer with a dropout rate of 25% is to randomly drop 25% of the neurons, thus mitigating overfitting caused by the limited sample size in steel bridge detection. The linear activation function allows the risk factor to be greater than 1.

[0125] The training learning rate is 1×10 -3 The optimizer is Adam, balancing convergence speed and accuracy. The training epochs are 200 to avoid overfitting. The batch size is 32 to reduce noise; 32 samples are randomly selected from the effective training samples. The main loss function in the current batch is:

[0126]

[0127] In the formula, L represents the main loss function. Let y represent the hazard factors predicted by the output for i samples. ture，i This represents the supervision label value of the i-th sample, where i represents the index of the 32 randomly selected samples, i = 1, 2, ..., 32. It directly measures the deviation between the predicted risk factor and the actual measurement, is suitable for regression tasks, and is not sensitive to extreme values, which is suitable for the average error requirement that needs to be considered in the fatigue risk assessment of steel bridges.

[0128] Step 4: In the weld area, infrared thermal images and three-dimensional morphology images of the weld surface are obtained by sliding sampling through image windows. These are used as inputs to the fatigue risk nonlinear regression model, and the risk factors at each window position are output to classify the fatigue damage in the weld area.

[0129] Use an electric heating plate to evenly heat the weld area to the heating temperature T. yr A thermal imager was used, with a spatial resolution of 0.1 mm, vertical scanning at a distance of 0.3 m from the weld surface, and a wavelength range of 7.5-13 μm, to generate a 640×512 pixel thermal image. Temperature was normalized. Using a structured light scanner with a scanning resolution of 0.1 mm / point and a translation speed of 5 mm / s along the z-axis, the entire length of the weld was covered, generating 640×512 pixel point cloud data. The origin was taken as the intersection of the weld profile centerline and the base material, ensuring that the x / z axis was consistent with the coordinate system of the thermal imaging image. The pixel physical coordinates were mapped as (x, zθ=(0.1a, 0.1b)).

[0130] A 64×64 pixel sliding window is set up with a sliding step of 16 pixels. It slides 40 times along the x-axis and 32 times along the z-axis, using zero-value padding to ensure the window edges are complete. A total of 1280 windows are generated. For each window, the physical space coordinates corresponding to its center pixel are extracted. Using the already constructed two-branch CNN model, a 64×64×1 matrix T is input. patch H patch The shared attention mechanism uses adaptive weight fusion to output the hot weight α. T and morphology α H The regression layer outputs the predicted hazard factors for each window, with pre-set hazard factor thresholds R1 and R2, where R1 > R2 > 0. For each sliding window, the output hazard factors... When satisfied When the area is classified as a low-risk damage area, and the following conditions are met... At that time, it was determined to be a medium-risk damage area, when the following conditions are met. When a region is identified as a high-risk damage area, i′ represents the index of the sliding window, i′=1, 2, …, 1280. In the Dang Van criterion, the shear fatigue limit value τ DV =115MPa, when the risk factor When the value exceeds 1, it indicates that the Dang Van parameter D under the current stress state has exceeded the fatigue limit of the material. At this point, the risk of fatigue crack initiation and propagation increases significantly. For example, if R1 is set to 1, when the following conditions are met... When an area is identified as a high-risk damage area, immediate local stress optimization or structural reinforcement is required. In actual steel bridge service, factors such as vehicle load and ambient temperature can cause stress amplitude fluctuations. When R² = 0.5, a 50% safety margin can be reserved to prevent low-risk areas from being misjudged as medium-risk due to accidental factors. Therefore, when the following conditions are met... When an area is identified as a medium-risk damage area, it needs to be included in the key monitoring targets; when the conditions are met... When the area is identified as a low-risk damage area, routine inspections can be performed to reduce maintenance costs. Table 2 shows the prediction results verification table of the dual-modal fatigue risk regression model, which is used to quantify the model's ability to predict fatigue risk in each area. After training with Monte Carlo Gaussian perturbation, the model collected 40 sets of window ID data under the geometric transition region at the center point of the window and summarized them into a table. The data were sorted by the predicted risk factors from smallest to largest. The overall error was less than 5%, indicating high accuracy and good predictive ability of the model for risk factors.

[0131]

[0132] like Figures 4-7 As shown, in Figure 4In the study, the predicted risk factors almost overlapped with the supervised label values ​​of the i-th sample, indicating sufficient learning of the heat-morphology-risk mapping relationship without overfitting. Figure 5 It can be seen that the overall error is relatively small, especially in high-risk areas, which not only meets the accuracy requirements but also has a higher ability to identify areas with high levels of fatigue damage. Figures 6-7 The values ​​of thermal weight and morphological weight change with the window ID. The color of the point in the figure represents the risk level of the corresponding window: red for risk level 1, green for risk level 2, and blue for risk level 3. Due to the large local temperature gradient caused by stress concentration, thermal imaging is more sensitive and the weight dominates between window IDs 30 and 40. On the other hand, morphological features such as abrupt changes in excess height and undercut have a more prominent structural induction effect on fatigue. Therefore, between window IDs 10 and 20, the weight of morphology dominates. This approach has both adaptive identification capability for complex defects and ensures prediction accuracy.

[0133] Please see Figure 8 The present invention further provides a multiaxial fatigue damage assessment apparatus for steel bridge welds considering welding residual stress, for performing the above-described multiaxial fatigue damage assessment method for steel bridge welds considering welding residual stress, comprising:

[0134] The image generation module is used to measure the three-dimensional residual stress in the geometric transition area of ​​the steel bridge weld area using X-rays, generate an infrared thermal image and a three-dimensional morphology image of the weld surface, and perform normalization processing. The geometric transition area includes the weld toe tip, the center point of the bevel bottom, and the base material reference point.

[0135] The robust hazard factor module is used to apply zero-mean Gaussian perturbation to the residual stress. It constructs a set of perturbation samples through multiple rounds of Monte Carlo calculations, performs multiple rounds of hazard factor calculations on the perturbation samples using the Dang Van criterion, and constructs robust weighted hazard factors based on the hazard factors.

[0136] The fatigue risk regression module is used to extract the thermal distribution features and geometric features of the weld seam through a deep convolutional neural network, using infrared thermal imaging and weld seam three-dimensional morphology as dual-modal inputs. The features are fused by a shared attention mechanism, and the robustness weight risk factor of the geometric transition region is used as a supervision signal to construct a nonlinear regression model for fatigue risk. The deep convolutional neural network is a dual-branch parallel convolutional neural network.

[0137] The weld risk classification module is used to acquire infrared thermal images and three-dimensional morphology images of the weld surface in the weld area through image window sliding sampling. These images are used as inputs to the fatigue risk nonlinear regression model, and the module outputs the hazard factors at each window position to classify the fatigue damage in the weld area.

[0138] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0139] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0140] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0141] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for assessing multiaxial fatigue damage of steel bridge welds considering residual welding stress, characterized in that, The specific steps include: Step 1: The geometric transition region of the steel bridge weld area is measured with X-ray to obtain a three-dimensional residual stress, generate an infrared thermal image and a three-dimensional morphology image of the weld surface, and perform normalization processing. The geometric transition region includes the weld toe tip, the center point of the bevel bottom and the base material reference point. Step 2: Apply a zero-mean Gaussian perturbation to the residual stress, construct a perturbation sample set through multiple rounds of Monte Carlo calculation, perform multiple rounds of hazard factor calculation on the perturbation sample using the Dang Van criterion, and construct a robust weighted hazard factor based on the hazard factor. Step 3: Using the infrared thermal image and the three-dimensional morphology image of the weld as dual-modal inputs, the thermal distribution features and geometric morphology features of the weld are extracted by a deep convolutional neural network, and the features are fused by a shared attention mechanism. The robustness weight risk factor of the geometric transition region is used as a supervision signal to construct a nonlinear regression model of fatigue risk. The deep convolutional neural network is a dual-branch parallel convolutional neural network. Step 4: Infrared thermal images and three-dimensional topographic images of the weld surface are obtained in the weld area by sliding sampling through an image window. These are used as inputs to the fatigue risk nonlinear regression model, and the risk factors at each window position are output to classify the fatigue damage in the weld area.

2. The multiaxial fatigue damage assessment method for steel bridge welds considering welding residual stress according to claim 1, characterized in that: The method for measuring the three-dimensional residual stress in the geometric transition region of the steel bridge weld area using X-rays is as follows: A spatial rectangular coordinate system is established with the intersection of the geometric center line of the weld section and the base metal as the origin O, the longitudinal direction of the weld as the x-axis, the direction perpendicular to the plate thickness as the y-axis, and the direction perpendicular to the weld cross section as the z-axis. The geometric transition region is defined as follows: the range within 0.5 mm of the geometric tip of the fusion line between the weld and the base metal is taken as the weld toe tip point; the range within 0.3 mm of the root of the weld, i.e., the lowest point where the fusion zone contacts the base metal, is taken as the bottom center point of the groove; and the base metal surface 30 mm away from the edge of the weld along the z-axis is taken as the base metal reference point. The measurement point grid for the weld toe tip is divided as follows: within 0.5 mm from the weld toe tip, a 5×5 equally spaced grid is used, with a point spacing of 0.5 mm, for a total of 25 measurement points; the measurement point grid for the center point of the bevel bottom is divided as follows: centered on the lowest point where the fusion zone contacts the base material, a 3×3 equally spaced grid is used, with a point spacing of 0.3 mm, for a total of 9 measurement points; the measurement point grid for the base material reference point is divided as follows: in this area, a 2×3 equally spaced grid is used, with a point spacing of 5 mm, for a total of 6 measurement points; three-dimensional residual stress is measured at each measurement point using an X-ray diffractometer to obtain normal stress and shear stress in MPa. Diffraction data are collected at three offset angles of -20°, 0°, and +20° for each measurement point to construct the stress tensor matrix. In the formula, Represents the stress tensor matrix. This represents the normal stress along the x-axis. This represents the normal stress along the y-axis. This represents the normal stress along the z-axis. This represents the shear stress in the xy plane. This represents the shear stress in the yz plane. This represents the shear stress in the xz plane.

3. The multiaxial fatigue damage assessment method for steel bridge welds considering welding residual stress according to claim 1, characterized in that: The method for generating infrared thermal images and three-dimensional morphology images of the weld surface, and then performing normalization processing, is as follows: Preset heating temperature is ,and After heating the steel bridge weld to a preset heating temperature, the spatial resolution of the high-resolution infrared thermal imager was set to 0.1 mm. A vertical scan along the z-axis was performed at a distance of 0.3 m from the weld surface, acquiring infrared radiation data in the 7.5-13 μm band, generating a 640×512 pixel infrared thermal image. The coordinate axes correspond to the physical spatial location of the weld area, and the pixel values ​​correspond to the surface temperature. The temperature was normalized, and the output was a two-dimensional thermal distribution matrix infrared thermal image. The formula used for temperature normalization is: Where, Indicates surface temperature. Indicates the normalized surface temperature; A structured light 3D scanner was used to scan the weld surface morphology. The scanner was moved perpendicular to the weld surface along the z-axis, covering the entire length of the weld. The scanning resolution was set to 0.1 mm / point to acquire point cloud data of the weld surface, which was then converted into a 640×512 pixel 3D weld morphology image. The pixel coordinates corresponded one-to-one with the actual spatial coordinates of the weld. The x-axis and z-axis in the coordinate system were set to correspond to the row and column directions of the 3D weld morphology image matrix, respectively. A grid was divided according to a spatial resolution of 0.1 mm to construct the coordinate system of the 3D weld morphology image. Each pixel in the 3D weld morphology image... The corresponding physical space coordinates are: Furthermore, this coordinate system uses the same resolution and alignment as the infrared thermal image, ensuring a one-to-one correspondence between pixel positions in the image in space, and normalizing the height of each coordinate. In the formula, Indicates the height of the normalization process. Indicates the weld surface height value. This represents the maximum value of the surface height of all welds. This represents the minimum value of the surface height of all weld seams.

4. The multiaxial fatigue damage assessment method for steel bridge welds considering welding residual stress according to claim 1, characterized in that: The method for constructing a perturbation sample set by applying a zero-mean Gaussian perturbation to the residual stress through multiple Monte Carlo rounds of calculation is as follows: A zero-mean Gaussian perturbation is introduced into the stress tensor of the residual stress at each measuring point to generate a perturbation tensor: In the formula, This represents the perturbation tensor. Let represent the perturbation matrix, where the elements of the perturbation matrix are generated using the following method: Standard normal random variables are generated using the Mersenne Twister algorithm. Generate perturbation matrix elements: Where, This represents the disturbance strength coefficient, which is 5%-10% of the yield strength of the steel bridge material. Indicates the first The perturbation matrix elements generated in the second simulation Represents the elements in the stress tensor matrix. This represents the row index of the stress tensor matrix. Represents the column index of the stress tensor matrix. Represents the row index of the perturbation matrix. The column index of the perturbation matrix is ​​generated repeatedly for each measurement point. Group perturbation tensors to form a perturbation sample set. ,in, It is a positive integer greater than 100.

5. The multiaxial fatigue damage assessment method for steel bridge welds considering welding residual stress according to claim 4, characterized in that: The method of using the Dang Van criterion to calculate hazard factors for perturbed samples in multiple rounds, and constructing robust weighted hazard factors based on these hazard factors, is as follows: For each perturbation tensor in the perturbation sample set, after applying a sinusoidal external load along the x-axis, the result is calculated according to the Dang Van criterion. There are risk factors, where the risk factor obtained from the k-th simulation is denoted as . Sort in ascending order and take the first... Each value serves as a robustness weight risk factor. .

6. The multiaxial fatigue damage assessment method for steel bridge welds considering welding residual stress according to claim 4, characterized in that: The method of extracting weld thermal distribution features and geometric morphology features using infrared thermal images and weld 3D topography images as dual-modal inputs and deep convolutional neural networks is as follows: A 64×64 pixel sliding window is set with a sliding step of 16 pixels. It slides 40 times along the x-axis and 32 times along the z-axis. Zero-value filling is used to make the edge window complete, generating a total of 1280 windows. For each window, the physical space coordinates corresponding to its center pixel are extracted, and it is determined whether it is located in the geometric transition area of ​​the weld seam. Only windows whose center point falls into the geometric transition area are retained as valid training samples. For each valid sample, the nearest neighbor interpolation method is used to find the robustness weight risk factor corresponding to the nearest measurement point, based on the physical coordinates of the window center, and this factor is used as the supervision label value of the window. The 64×64 pixel matrix cropped from the infrared thermal image is denoted as... The 64×64 pixel matrix extracted from the 3D topography of the weld is denoted as... ; A dual-branch parallel convolutional neural network is used to extract features from both the infrared thermal image and the 3D topography image. The specific structure is as follows: Input layer: Receives a 64×64×1 single-channel image; Convolutional layer of the first convolutional block: 32 3×3 convolutional kernels, "same" padding, ReLU activation function, batch normalization layer to standardize channel mean and variance, max pooling layer with 2×2 pooling kernel, stride 2, outputting a 32×32×32 feature map; Convolutional layer of the second convolutional block: 64 3×3 convolutional kernels, "same" padding, ReLU activation function, batch normalization layer to standardize channel mean and variance, max pooling layer with 2×2 pooling kernel, stride 2, outputting a 16×16×64 feature map; Flattening layer: Flattens the 3D feature map into a 16384-dimensional feature vector.

7. The multiaxial fatigue damage assessment method for steel bridge welds considering welding residual stress according to claim 6, characterized in that: The shared attention mechanism fuses features, and the robustness weight risk factor of the geometric transition region is used as a supervision signal to construct a nonlinear regression model for fatigue risk. The shared attention mechanism concatenates the 16384-dimensional vector from the two branches into a 32768-dimensional vector. Fully connected layer 1 has 1024 neurons with ReLU activation; fully connected layer 2 has 512 neurons with ReLU activation; and fully connected layer 3 has 2 neurons with Sigmoid activation, resulting in hot output weights. and shape weight ,and ; A regression model is constructed using a fully connected network and Dropout layers to output predicted risk factors. The steps for constructing the regression model using a fully connected network and Dropout layers are as follows: Fully connected layer 1: 2048 neurons, ReLU activation function; Dropout layer: 25% dropout rate; Fully connected layer 2: 1024 neurons, ReLU activation function; Dropout layer: 25% dropout rate; Fully connected layer 3: 1 neuron, linear activation function; Training learning rate The optimizer is Adam, the training epochs are 200, Dropout is 0.25, and the batch size is 32, meaning 32 samples are randomly selected from the valid training samples. The main loss function in the current batch is: In the formula, Represents the main loss function. express The output predicted risk factors for each sample Indicates the first The supervision label value of each sample, This represents the index of the 32 randomly selected samples. .

8. The multiaxial fatigue damage assessment method for steel bridge welds considering welding residual stress according to claim 7, characterized in that: The method for classifying fatigue damage in the weld area by outputting the hazard factors at each window location is as follows: For steel bridge welds requiring damage assessment, the hazard factors output by each sliding window are extracted to construct... A two-dimensional risk matrix with a one-to-one correspondence with the physical coordinates of the weld, and pre-set thresholds for risk factors. and ,in, For each sliding window output hazard factor When satisfied When the condition is met, it is determined to be a low-risk damage area. At that time, it was determined to be a medium-risk damage area, when the following conditions are met. At that time, it was identified as a high-risk damage area, among which, Indicates the index of the sliding window. .

9. A multiaxial fatigue damage assessment device for steel bridge welds considering residual welding stress, characterized in that: The apparatus is used to perform the multiaxial fatigue damage assessment method for steel bridge welds considering welding residual stress as described in any one of claims 1-8, including: The image generation module is used to measure the three-dimensional residual stress in the geometric transition area of ​​the steel bridge weld area using X-rays, generate an infrared thermal image and a three-dimensional morphology image of the weld surface, and perform normalization processing. The geometric transition area includes the weld toe tip, the center point of the bevel bottom, and the base material reference point. The robust hazard factor module is used to apply zero-mean Gaussian perturbation to the residual stress. It constructs a set of perturbation samples through multiple rounds of Monte Carlo calculations, performs multiple rounds of hazard factor calculations on the perturbation samples using the Dang Van criterion, and constructs robust weighted hazard factors based on the hazard factors. The fatigue risk regression module is used to extract the thermal distribution features and geometric features of the weld seam through a deep convolutional neural network, using infrared thermal imaging and weld seam three-dimensional morphology as dual-modal inputs. The features are fused by a shared attention mechanism, and the robustness weight risk factor of the geometric transition region is used as a supervision signal to construct a nonlinear regression model of fatigue risk. The deep convolutional neural network is a dual-branch parallel convolutional neural network. The weld risk classification module is used to acquire infrared thermal images and three-dimensional morphology images of the weld surface in the weld area through image window sliding sampling. These images are used as inputs to the fatigue risk nonlinear regression model, and the module outputs the hazard factors at each window position to classify the fatigue damage in the weld area.

Images