A method for rapid evaluation of fatigue life of steel trestle for extreme heavy load impact

Abstract

The application discloses a kind of steel trestle fatigue life rapid evaluation methods for extreme heavy load impact, it is related to bridge fatigue life evaluation technical field.Establish extreme impact load probability model;Multi-source bridge monitoring;The establishment and revision of finite element model;Fatigue damage field efficient simulation;Crack propagation multi-stage modeling and prediction;Multi-scale damage accumulation fusion analysis;Residual life evaluation and multi-dimensional risk classification.Through the combination of extreme impact load modeling, multi-source data fusion and physical constraint neural network technology, the stress response of the structure can be accurately identified and calculated under extreme heavy load conditions, and a three-stage crack propagation model is introduced to fully describe the entire process from micro-damage to structural fracture, significantly improving the efficiency and reliability of fatigue life evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bridge fatigue life assessment, in particular to a steel trestle fatigue life rapid assessment method for extreme heavy load impact. BACKGROUND

[0002] In the field of bridge structure fatigue research, traditional fatigue life prediction methods are mostly based on stress-life (S-N) curves or Miner linear cumulative criteria, which are difficult to accurately describe the nonlinear damage evolution characteristics under actual working conditions. With the popularity of heavy traffic transportation, temporary load-bearing structures such as steel trestles often suffer from the coupling effect of overloading vehicles and extreme impact loads during service, resulting in frequent problems such as structural component fatigue cracking, weld damage and local buckling. However, existing research still has obvious deficiencies in extreme impact load identification, multi-source monitoring data fusion and multi-stage crack propagation prediction.

[0003] In recent years, machine learning and deep learning technologies have shown strong modeling potential in the field of structural health monitoring, and some scholars have tried to introduce them into bridge fatigue life prediction to improve the fitting accuracy and prediction efficiency of the model. However, most current researches still remain at the level of single signal analysis or empirical regression, lacking physical mechanism constraints and limited model generalization ability. In addition, there is great uncertainty in bridge material parameters, support stiffness and connecting nodes in actual engineering, and the randomness of extreme impact events also makes it difficult for traditional methods to achieve rapid and stable life assessment.

[0004] Therefore, under the environment of extreme heavy load impact, how to comprehensively consider the random impact characteristics, material nonlinearity and multi-source fusion of monitoring data, and establish an evaluation method that not only conforms to the physical law but also can realize high-precision fatigue life rapid prediction, has become a technical problem to be solved in the field of bridge structure health monitoring and life management. SUMMARY

[0005] To solve the problems in the background art, the present application provides a steel trestle fatigue life rapid assessment method for extreme heavy load impact, which can accurately identify impact loads and calculate the stress response of the structure under extreme heavy load conditions by combining extreme impact load modeling, multi-source data fusion and physically constrained neural network technology, and simultaneously introduces a three-stage crack propagation model to comprehensively describe the whole process from micro-damage to structural fracture, significantly improving the efficiency and reliability of fatigue life assessment.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solution: a steel trestle fatigue life rapid assessment method for extreme heavy load impact, comprising the following steps:

[0007] Step 1: Establish an extreme impact load probability model;

[0008] The load vector is constructed by recording the axle load, vehicle speed and time required for the vehicle to pass through the steel trestle bridge in a set period of time, the total equivalent impact load of a single vehicle acting on the bridge is calculated based on the load vector, and then the extreme impact load set is screened and the probability density function thereof is fitted to obtain an extreme impact load-probability matrix;

[0009] Step two: multi-source bridge monitoring

[0010] According to the actual situation of the test site, representative monitoring vehicles are selected to pass through the bridge, strain sensors, displacement sensors and acceleration sensors are arranged at key positions of the bridge, and monitoring vectors are constructed by synchronously collecting monitoring signals The key positions include each support point, 1 / 4 span and mid-span position of the bridge

[0011] Step three: establishment and correction of the finite element model

[0012] A finite element model of the steel trestle bridge is established, the same load as the monitoring vehicle in step two is loaded, and the output vector composed of the strain, displacement and acceleration calculated at the key positions of the bridge is output by the finite element model Four sensitive parameters, i.e. elastic modulus of steel member, density of steel material, vertical stiffness of support and structural damping ratio, are selected as correction parameters of the finite element model, and then the relative sensitivity coefficients are calculated and the parameters meeting the requirements are selected as optimization variables, after that, weights are set for the three sensors and standardized, and on this basis, an objective function is constructed, the value of the optimization variables is selected to minimize the objective function, and a corrected finite element model is obtained, and then the extreme impact load-probability matrix obtained in step one is input into the corrected finite element model to obtain an output set

[0013] Step four: efficient simulation of fatigue damage field

[0014] The physical constraint equation of the fatigue damage field is constructed, and the output set of step three is used as the input variable of the PINN machine learning model The loss function of the PINN machine learning model includes the physical constraint equation condition and three boundary conditions of initial threshold, transition trigger and fracture criterion, and the model parameters are iteratively optimized to minimize the loss function for fatigue damage field training

[0015] Step five: multi-stage modeling and prediction of crack propagation

[0016] The crack propagation is divided into three stages of initiation, stable propagation and rapid propagation, the stress cycles required for each stage are calculated, the stress cycles of the three stages are accumulated to obtain the total cycle number of the bridge at different positions, and the minimum value is selected as the total cycle number of the whole bridge ​​

[0017] Step Six: Multi-scale damage accumulation and fusion analysis;

[0018] Extract the fatigue damage value and hot spot crack length from the output of step four, and combine them with the total number of cycles of the entire bridge from step five. Three damage indices were constructed: microscopic damage field, macroscopic crack evolution, and crack propagation stage. A unified damage index was obtained through weighted fusion.

[0019] Step 7: Remaining life assessment and multidimensional risk classification;

[0020] Based on a unified damage index and the number of cycles the bridge has already withstood, the remaining lifespan is calculated to determine the number of cycles it can withstand. This is then combined with the average load cycle rate to predict the remaining lifespan, and finally, a dimensionless term is constructed. Ultimately based on The risk level of the steel trestle bridge is classified based on the value and preset threshold.

[0021] Furthermore, in step one, the process of constructing the extreme impact load-probability matrix includes:

[0022] S1.1, Load Vector Represented as:

[0023]

[0024] In the formula, For the first vehicle number Axle load of each axle For the first The speed of the vehicle, For the first The time it takes for a vehicle to cross the bridge. , , For the number of vehicles, This refers to the number of axles.

[0025] S1.2, Total equivalent impact load of a single vehicle acting on the bridge Represented as:

[0026]

[0027] In the formula, For the first The influence coefficient of each axle on the total impact effect To account for the impact amplification factor due to vehicle speed, , For reference speed, take 80km / h;

[0028] S1.3 Setting a threshold Screening of extreme impact load sets , The fitted probability density function is:

[0029]

[0030] In the formula, Values ​​are taken for extreme impact loads. , , Given the fitted shape, size, and location parameters, the resulting extreme impact load-probability matrix is ​​expressed as:

[0031]

[0032] In the formula, This represents a specific extreme impact load value after statistical analysis. This represents the probability of an extreme impact load value occurring.

[0033] Furthermore, in step three, the finite element model correction process includes:

[0034] S3.1 Selecting the correction vector Includes the elastic modulus of steel components Density of steel materials Vertical stiffness of supports and structural damping ratio Four sensitivity parameters;

[0035] S3.2, Define the relative sensitivity coefficient for:

[0036]

[0037] In the formula, For the correction vector The first in One parameter, For the output vector The first in One parameter, express The parameter change value, express The parameter change value;

[0038] S3.3 Setting a threshold ,reserve The parameters are used as optimization variables;

[0039] S3.4. Assign weights to the three sensors and standardize them, setting:

[0040]

[0041] In the formula, , and These represent the weighting coefficients for the three sensors, respectively. , and These represent the covariance estimates for the three sensors, respectively.

[0042] S3.5, Construct the objective function as follows:

[0043]

[0044] In the formula, Represents the L2 norm, for The The parameters are obtained by iteratively solving the objective function, and the corrected parameters are then imported into the finite element model.

[0045] Furthermore, in step three, the output set Represented as:

[0046]

[0047] In the formula, For the first Node spatial coordinates , For the set of Number of applied stress cycles , For the set of Crack length , , , These are the maximum values ​​of the crack tip stress intensity factor. Minimum value and amplitude , For the set of Equivalent stress amplitude , .

[0048] Furthermore, in step four, the simulation process of the fatigue damage field includes:

[0049] S4.1 Construct the physical constraint equations for the fatigue damage field, expressed as:

[0050]

[0051] In the formula, Indicates fatigue damage value. Let be the material loss constant. The yield strength of the material. An index representing the degree to which stress amplitude affects damage rate. An index representing the degree of nonlinearity controlling damage evolution. Indicates the damage diffusion index;

[0052] In S4.2, the PINN machine learning model:

[0053] Input layer:

[0054]

[0055] Output layer:

[0056]

[0057] In the formula, For fracture toughness, This indicates the range of uncertainty in fatigue damage values. Indicates the length of the hot spot crack. This indicates the range of uncertainty regarding the length of the hot spot crack. The length of the crack microstructure. The initial length of the crack. This is the crack transition length. The critical crack length. This represents the effective stress intensity factor amplitude.

[0058] S4.3, The loss function is expressed as follows:

[0059]

[0060] In the formula, These are the physical constraint weighting coefficients. These are the initial threshold term weight coefficients. For the weighting coefficient of the transition trigger term, For the fracture criterion term, the weight coefficient is... This is the toughness proportionality coefficient. The minimum stress intensity factor amplitude required for a crack to begin stable propagation.

[0061] Furthermore, in step five, the stress cycle number for the three stages of crack propagation is expressed as follows:

[0062] Infancy stage:

[0063]

[0064]

[0065] In the formula, For cracks from the microstructure length Extended to initial length Required number of stress cycles, and These are the crack growth constant and the exponent, respectively. For Coffin–Manson parameter functions, For mapping index, For equivalent amplitude, The fatigue strength coefficient, The elastic modulus of the material. The fatigue ductility coefficient, The high-cycle fatigue index. Low-cycle fatigue index;

[0066] Stable expansion phase:

[0067]

[0068] In the formula, For the crack from the initial length Extended to transition length Required number of stress cycles, Fatigue damage value Influenced material constants Fatigue damage value The damage constant affected;

[0069] Rapid expansion phase:

[0070]

[0071] In the formula, For the crack from the transition length Extended to critical crack length Required number of stress cycles, The stress intensity factor under maximum load. For effective fracture toughness, This is a parameter representing the sensitivity of crack propagation to load. The correction parameter is used to prevent crack propagation from approaching the threshold. This is the effective crack propagation threshold.

[0072] Furthermore, in step six, the three damage indicators are respectively expressed as follows:

[0073] Damage indices of the microscopic damage field:

[0074]

[0075] Damage indicators of macroscopic crack evolution:

[0076]

[0077] Damage indicators during crack propagation:

[0078]

[0079] In the formula, Indicates hotspot area, Indicates the weight of the hotspot area. for Any point in the, For sensitivity parameters;

[0080] The unified damage index is then expressed as:

[0081]

[0082] In the formula, , , The weights of the three indicators are respectively, and they satisfy the following conditions: , This is the sensitivity weighting adjustment coefficient. .

[0083] Furthermore, in step seven, the remaining life assessment and multidimensional risk grading include:

[0084] S7.1, The number of cycles that the remaining lifetime can withstand is , This refers to the number of cycles that the system can withstand under life-limit conditions. Predict remaining lifespan based on the number of cycles already completed. , The average load cycle rate of the steel trestle bridge;

[0085] S7.2, Define a dimensionless term Calculated by the following formula:

[0086]

[0087]

[0088]

[0089]

[0090] In the formula, For residual stress, This represents the cumulative probability of an extreme shock occurring. For reference lifespan, To allow stress, , , These are the weights for the life term, residual stress term, and extreme probability, respectively.

[0091] S7.3, Preset Threshold , , ,according to , , and The steel trestle bridge is divided into four risk levels, from low to high, based on the degree of risk.

[0092] Compared with existing technologies, the beneficial effects of this invention are as follows: To solve the problems of low fatigue life prediction accuracy, difficulty in identifying extreme impact loads, and insufficient utilization of multi-source monitoring data in existing technologies, this invention combines extreme value statistics and multi-source data fusion technology to establish an extreme impact load identification model. By inverting and correcting finite element parameters through multi-source monitoring data, the calculation accuracy of structural stress response is improved. Furthermore, a fatigue damage field model is constructed based on a physical constraint neural network to achieve the prediction of damage evolution with the number of cycles. At the same time, physical constraint conditions of crack initiation, stable propagation, and rapid fracture are introduced to establish a full life prediction system. This effectively solves the problems of computational complexity and unstable prediction in traditional methods for fatigue life assessment under extreme loads, providing rapid, intelligent, and scientific technical support for the safety assessment and operation and maintenance decisions of steel trestle bridges. Attached Figure Description

[0093] Figure 1 This is a flowchart of the method of the present invention;

[0094] Figure 2 This is a finite element model diagram from the embodiment. Detailed Implementation

[0095] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0096] like Figure 1 As shown, a rapid fatigue life assessment method for steel trestle bridges subjected to extreme heavy-load impacts includes the following steps:

[0097] Step 1: Establish a probabilistic model for extreme impact loads;

[0098] Based on the demand, a time period is determined, and three data points are recorded for each vehicle passing over the steel trestle bridge during that time period: axle load, vehicle speed, and time required to cross the bridge. The load vector is then represented as:

[0099]

[0100] In the formula, For the first vehicle number Axle load of each axle For the first The speed of the vehicle, For the first The time it takes for a vehicle to cross the bridge. , , For the number of vehicles, This refers to the number of axles.

[0101] Then the first Total equivalent impact load of a vehicle on a bridge Represented as:

[0102]

[0103] In the formula, For the first The influence coefficient of each axle on the total impact effect The impact amplification factor, which takes into account the effect of vehicle speed, is calculated by the following formula:

[0104]

[0105] In the formula, For reference speed, 80 km / h is generally used.

[0106] The initial impact load set is then represented as:

[0107]

[0108] Select a threshold from the initial impact load set. Define the set of extreme impact loads Represented as:

[0109]

[0110] Fitting the set of extreme impact loads, its probability density function is:

[0111]

[0112] In the formula, Values ​​are taken for extreme impact loads. , , These are the fitted shape, size, and position parameters, respectively.

[0113] The resulting extreme impact load-probability matrix is ​​expressed as follows:

[0114]

[0115] In the formula, This represents a specific extreme impact load value after statistical analysis. This represents the probability of an extreme impact load value occurring.

[0116] Step 2: Multi-source bridge monitoring;

[0117] First, based on the actual conditions at the test site, a representative vehicle was selected as the monitoring vehicle. This vehicle traversed the entire bridge, and strain sensors, displacement sensors, and acceleration sensors were simultaneously deployed at key locations on the bridge. These key locations included each support point, the quarter-span, and the mid-span. At any given moment... Simultaneously acquire monitoring signals and define monitoring vectors for:

[0118]

[0119] In the formula, Represents strain data, Represents displacement data. Indicates acceleration data, subscript This indicates the number of a single type of sensor.

[0120] Step 3: Establishment and correction of the finite element model;

[0121] A finite element model of the steel trestle bridge was established based on the actual situation. First, the same load as the vehicle monitored in step two was added, and the strain, displacement, and acceleration calculated by the finite element model were obtained at key parts of the bridge, thus obtaining the output vector of the finite element model. Represented as:

[0122]

[0123] In the formula, , and These represent the strain, displacement, and acceleration data output by the finite element model, respectively.

[0124] Selecting the elastic modulus of steel components for steel trestle bridges Density of steel materials Vertical stiffness of supports and structural damping ratio The four sensitivity parameters are used as correction parameters for the finite element model to obtain the correction vector. Represented as:

[0125]

[0126] Then define the relative sensitivity coefficient. for:

[0127]

[0128] In the formula, For the correction vector The first in One parameter, For the output vector The first in One parameter, express The parameter change value, express The parameter change value.

[0129] Then set the threshold ,reserve The parameters are used as optimization variables.

[0130] To account for dimensions and confidence levels, weights are assigned to the three sensors and standardized, assuming:

[0131]

[0132] In the formula, , and These represent the weighting coefficients for the three sensors, respectively. , and These represent the covariance estimates for the three sensors, respectively.

[0133] The objective function is constructed as follows:

[0134]

[0135] In the formula, Represents the L2 norm, for The One parameter.

[0136] The optimal values ​​of the optimization variables are selected to minimize the objective function, and the corrected parameters are then imported into the finite element model to obtain the corrected finite element model. Finally, the extreme impact load-probability matrix obtained in step one is used... Input into the corrected finite element model, output set :

[0137]

[0138] In the formula, For the first Node spatial coordinates , For the set of Number of applied stress cycles , For the set of Crack length , , , These are the maximum values ​​of the crack tip stress intensity factor. Minimum value and amplitude , For the set of Equivalent stress amplitude , .

[0139] Step 4: Efficient simulation of fatigue damage field;

[0140] First, the physical constraint equations for the fatigue damage field are constructed, expressed as:

[0141]

[0142] In the formula, Indicates fatigue damage value. Let be the material loss constant. The yield strength of the material. An index representing the degree to which stress amplitude affects damage rate. An index representing the degree of nonlinearity controlling damage evolution. This indicates the damage diffusion index.

[0143] The PINN machine learning model based on physical constraint equations is used for efficient simulation of fatigue damage field, and the output set in step three is used to perform this simulation. The fatigue damage field is trained using these variables as input variables to the PINN machine learning model.

[0144] Input layer:

[0145]

[0146] Output layer:

[0147]

[0148] In the formula, For fracture toughness, This indicates the range of uncertainty in fatigue damage values. Indicates the length of the hot spot crack. This indicates the range of uncertainty regarding the length of the hot spot crack. The length of the crack microstructure. The initial length of the crack. This is the crack transition length. The critical crack length. This represents the effective stress intensity factor amplitude.

[0149] The loss function of the neural network is established, which includes the physical constraint equations and three boundary conditions: initial threshold, transition triggering, and breakage criterion. The loss function is expressed as follows:

[0150]

[0151] In the formula, These are the physical constraint weighting coefficients. These are the initial threshold term weight coefficients. For the weighting coefficient of the transition trigger term, For the fracture criterion term, the weight coefficient is... This is the toughness proportionality coefficient. The minimum stress intensity factor amplitude required for a crack to begin stable propagation.

[0152] By iteratively optimizing the neural network parameters to minimize the loss function and progressively solving the loss function, the output of the machine learning model is ensured to be consistent with the actual crack evolution law.

[0153] Step 5: Multi-stage modeling and prediction of crack propagation;

[0154] Crack propagation can be divided into three stages: initiation, stable propagation, and rapid propagation.

[0155] Infancy stage:

[0156]

[0157]

[0158] In the formula, For cracks from the microstructure length Extended to initial length Required number of stress cycles, and These are the crack growth constant and the exponent, respectively. For Coffin–Manson parameter functions, For mapping index, For equivalent amplitude, The fatigue strength coefficient, The elastic modulus of the material. The fatigue ductility coefficient, The high-cycle fatigue index. It is a low-cycle fatigue index.

[0159] Stable expansion phase:

[0160]

[0161] In the formula, For the crack from the initial length Extended to transition length Required number of stress cycles, Fatigue damage value Influenced material constants , Fatigue damage value The influence of damage constant, .

[0162] Rapid expansion phase:

[0163]

[0164] In the formula, For the crack from the transition length Extended to critical crack length Required number of stress cycles, The stress intensity factor under maximum load. For effective fracture toughness, , This is a parameter representing the sensitivity of crack propagation to load. The correction parameter is used to prevent crack propagation from approaching the threshold. To be the effective crack propagation threshold, .

[0165] The model uses crack length as the transition trigger condition between each stage, thus achieving cross-stage, multi-mechanism simulation of the entire crack life process. The total number of cycles required for the three stages of crack propagation can then be calculated using the following formula. , is represented as:

[0166]

[0167] The total number of cycles at different locations on the bridge was calculated. Then, select the minimum value. This is the total number of cycles for the entire bridge.

[0168] Step Six: Multi-scale damage accumulation and fusion analysis;

[0169] The microscopic structures obtained in step four With macro and the total number of loops of the entire bridge obtained in step five. The damage index is integrated into a unified damage index, in which:

[0170] The damage index of the micro-damage field is expressed as:

[0171]

[0172] The damage index of macroscopic crack evolution is expressed as:

[0173]

[0174] The damage index during the crack propagation stage is expressed as:

[0175]

[0176] In the formula, Indicates hotspot area, Indicates the weight of the hotspot area. for Any point in the, This refers to the sensitivity parameter.

[0177] The three damage indicators are then fused to obtain a unified damage indicator, which is expressed as:

[0178]

[0179] In the formula, , , The weights of the three indicators are respectively, and they satisfy the following conditions: , This is the sensitivity weighting adjustment coefficient. .

[0180] Step 7: Remaining life assessment and multidimensional risk classification;

[0181] Due to unified damage index This represents the proportion of fusion damage after a certain number of cycles have been endured, assuming the number of cycles has been endured is 1. The number of cycles that the lifespan limit condition can withstand is Then the number of cycles that the remaining lifespan can withstand is Converting to time yields the predicted remaining lifespan. , The average load cycle rate of the steel trestle bridge.

[0182] Define a dimensionless term Calculated by the following formula:

[0183]

[0184]

[0185]

[0186]

[0187] In the formula, For residual stress, This represents the cumulative probability of an extreme shock occurring. For reference lifespan, To allow stress, , , These represent the weights for the life term, residual stress term, and extreme probability, respectively.

[0188] According to the dimensionless term Combined with the selected , and Three thresholds, , , The risk classification of steel trestle bridges is divided into four risk levels from low to high: Level I (Safe), Level II (Attention), Level III (Warning), and Level IV (Emergency). Level I corresponds to... Level II corresponds Level III corresponds Level IV corresponds .

[0189] Example

[0190] This embodiment demonstrates the practical application of the method of the present invention to the earthwork steel trestle bridge of the Harbin Metropolitan Area North Ring Road, as shown below:

[0191] S1. Establish a probabilistic model for extreme impact loads:

[0192] By deploying a monitoring system, three types of data were recorded for 2,000 heavy-duty six-axle vehicles on the steel trestle bridge: axle load, vehicle speed, and time required to cross the bridge.

[0193] The obtained load vector is (partial):

[0194]

[0195] Calculated influence coefficient Impact amplification factor and total equivalent impact load (Partial) is shown in the table below:

[0196]

[0197] The initial impact load set is then:

[0198]

[0199] In the initial set of impact loads, a threshold is selected. Then the set of extreme impact loads is:

[0200]

[0201] By fitting the set of extreme impact loads, the extreme impact load-probability matrix is ​​obtained as follows:

[0202]

[0203] S2, Multi-source Bridge Monitoring:

[0204] Based on the actual conditions at the test site, a 45-ton "four-axle, eight-wheel" dump truck was selected as the monitoring vehicle. This vehicle traversed the entire bridge from the starting point, and strain sensors, displacement sensors, and acceleration sensors were deployed at each support point, quarter span, and mid-span of the bridge, with 50 of each type. The collected data are shown in the table below:

[0205]

[0206] S3. Establishment and correction of the finite element model:

[0207] A finite element model is established based on finite element analysis software, and the model is combined with... Figure 2 As shown. Adding the same load as the monitored vehicle in step S2, the strain, displacement, and acceleration magnitudes (partial) calculated by the finite element model at key parts of the bridge are shown in the table below:

[0208]

[0209] The sensitivity parameters of the steel trestle bridge are selected and represented as a correction vector:

[0210]

[0211] The calculated sensitivity parameters and relative sensitivity coefficients are shown in the table below:

[0212]

[0213] Set threshold ,reserve , To optimize the variables, we first calculated the covariance estimates for the three sensors, as shown in the table below (partially):

[0214]

[0215] After quality assessment, weighting coefficients were selected for the three types of sensors. , , ,but:

[0216]

[0217]

[0218]

[0219]

[0220] Construct the objective function:

[0221]

[0222] The objective function is solved using a sensitivity-guided Gauss-Newton iterative method, and the iteration is terminated according to the residual convergence criterion. After solving... Revised to 212000MPa The value is adjusted to 0.028, while other parameters remain unchanged. After obtaining the corrected finite element model, the extreme impact load-probability matrix obtained in step one is then used. Input into the model, output set (Partial) is:

[0223]

[0224] S4. High-efficiency simulation of fatigue damage field:

[0225] First, the physical neural network structure is invoked, and an input layer is created. Multiple physical variables are passed to the model via a unified tensor interface. Before entering the model, these variables are processed by a Normalization Layer and a Batch Format Wrapper to ensure that the different physical quantities have trainable consistency. Partial data of the model's input layer is shown in the table below:

[0226]

[0227] In addition, the fracture toughness input to the model for .

[0228] Within the model, a physical constraint function is constructed using `compute_physics_constraint`, followed by a physical constraint-driven multi-branch neural network structure. The main body uses the `build_pinn_backbone` method to build a multi-layer fully-connected encoder, and multi-head decoders are set up for prediction. The differentiability of the output is maintained by the tanh activation function. Multiple functional output branches are set after the backbone layer to simultaneously predict multiple key variables within the fatigue process.

[0229] The total loss function is established using the `calculate_loss` function, and the total loss consists of data terms and physical terms. Physical constraints are defined using the `compute_physics_constraint` function. Initial threshold conditions are defined using the `masked_mse` function. Transition trigger conditions are written into the loss term using the `relu_barrier` function. Fracture criterion conditions are constructed using the `fracture_barrier` function.

[0230] After establishing the loss function, the training phase begins. First, an adaptive gradient optimizer is used for initial parameter search, with the learning rate set to [value missing]. To obtain the decreasing trend of the physical residuals, a quasi-Newton optimizer is then used for global convergence, improving the coordination between the PDE residuals and crack constraints. The number of training iterations is set to be no less than 10,000 rounds.

[0231] After training, the results are output on all nodes through the batch inference interface. The model output results (partial) are shown in the table below:

[0232]

[0233] In addition, it also outputs , , , .

[0234] S5. Multi-stage modeling and prediction of crack propagation:

[0235] The following section discusses the three stages of crack propagation: initiation, stable propagation, and rapid propagation. , , Taking the data at that location as an example, calculate the total number of cycles required for the three stages of crack propagation.

[0236] Relevant parameters of the germination stage, equivalent amplitude Material elastic modulus fatigue strength coefficient fatigue ductility coefficient High-cycle fatigue index Low-cycle fatigue index Crack growth constant Crack growth index Mapping index Calculated:

[0237]

[0238]

[0239] During the stable expansion phase, calculations show that:

[0240]

[0241]

[0242] Relevant parameters during the rapid crack propagation phase, and sensitivity parameters for crack propagation. Correct parameters Calculated:

[0243]

[0244]

[0245] The total number of cycles required for the three stages of crack propagation is:

[0246]

[0247] Calculate all the different locations of the bridge Then, select the minimum value. This is the total number of cycles for the entire bridge.

[0248] S6. Multi-scale damage accumulation and fusion analysis:

[0249] The microscopic structures obtained in step four With macro and the total number of loops of the entire bridge obtained in step five. Integrate into a unified damage index, and select , , , The calculation results (partial) are shown in the table below:

[0250]

[0251] S7. Remaining life assessment and multidimensional risk stratification:

[0252] Based on the number of cycles the bridge has already withstood For example, the remaining lifespan is , Times / day, converted to time to obtain predicted remaining lifespan. sky.

[0253] Dimensionless term Calculated by the following formula:

[0254]

[0255]

[0256]

[0257] threshold , , In this embodiment, the bridge Value at and Between these levels, the risk level is Level II (attention).

[0258] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0259] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A rapid fatigue life assessment method for steel trestle bridges subjected to extreme heavy-load impacts, characterized in that: Includes the following steps: Step 1: Establish a probabilistic model for extreme impact loads; A load vector is constructed by recording three types of data: axle load, vehicle speed, and time required for vehicles to cross the steel trestle bridge within a set time period. Based on the load vector, the total equivalent impact load of a single vehicle on the bridge is calculated. Then, the set of extreme impact loads is selected and its probability density function is fitted to obtain the extreme impact load-probability matrix. Step 2: Multi-source bridge monitoring; Based on the actual conditions at the test site, representative monitoring vehicles were selected to cross the bridge. Strain sensors, displacement sensors, and acceleration sensors were deployed at key parts of the bridge to simultaneously collect monitoring signals and construct monitoring vectors. The key components include each support point, quarter span, and mid-span of the bridge; Step 3: Establishment and correction of the finite element model; A finite element model of the steel trestle bridge was established, and the same load as the vehicle monitored in step two was applied. The finite element model output a vector consisting of strain, displacement, and acceleration calculated at key parts of the bridge. Four sensitivity parameters—elastic modulus of steel components, steel density, support vertical stiffness, and structural damping ratio—were selected as correction parameters for the finite element model. The relative sensitivity coefficients were then calculated, and parameters meeting the requirements were selected and retained as optimization variables. Weights were then assigned to the three sensors and standardized. Based on this, an objective function was constructed, and the values ​​of the optimization variables were selected to minimize the objective function, resulting in the corrected finite element model. The extreme impact load-probability matrix obtained in step one was then input into the corrected finite element model to obtain the output set. ; Step 4: Efficient simulation of fatigue damage field; The physical constraint equations for the fatigue damage field are constructed, and the PINN machine learning model is used with the output set from step three. As input variables, the loss function of the PINN machine learning model includes physical constraint equation conditions and three boundary conditions: initial threshold, transition trigger, and fracture criterion. The model parameters are iteratively optimized to minimize the loss function, and fatigue damage field training is performed. Step 5: Multi-stage modeling and prediction of crack propagation; Crack propagation is divided into three stages: initiation, stable propagation, and rapid propagation. The required stress cycles for each stage are calculated separately. Using crack length as the transition trigger condition, the stress cycle counts of the three stages are summed to obtain the total number of cycles at different locations on the bridge. The minimum value is selected as the total number of cycles for the entire bridge. ; Step Six: Multi-scale damage accumulation and fusion analysis; Extract the fatigue damage value and hot spot crack length from the output of step four, and combine them with the total number of cycles of the entire bridge from step five. Three damage indices were constructed: microscopic damage field, macroscopic crack evolution, and crack propagation stage. A unified damage index was obtained through weighted fusion. Step 7: Remaining life assessment and multidimensional risk classification; Based on a unified damage index and the number of cycles the bridge has already withstood, the remaining lifespan is calculated to determine the number of cycles it can withstand. This is then combined with the average load cycle rate to predict the remaining lifespan, and finally, a dimensionless term is constructed. Ultimately based on The risk level of the steel trestle bridge is classified based on the value and preset threshold.

2. The method for rapid fatigue life assessment of steel trestle bridges under extreme heavy-load impact as described in claim 1, characterized in that: In step one, the process of constructing the extreme impact load-probability matrix includes: S1.1, Load Vector Represented as: In the formula, For the first vehicle number Axle load of each axle For the first The speed of the vehicle, For the first The time it takes for a vehicle to cross the bridge. , , For the number of vehicles, This refers to the number of axles. S1.2, Total equivalent impact load of a single vehicle acting on the bridge Represented as: In the formula, For the first The influence coefficient of each axle on the total impact effect To account for the impact amplification factor due to vehicle speed, , For reference speed, take 80km / h; S1.3 Setting a threshold Screening of extreme impact load sets , The fitted probability density function is: In the formula, Values ​​are taken for extreme impact loads. , , Given the fitted shape, size, and location parameters, the resulting extreme impact load-probability matrix is ​​expressed as: In the formula, This represents a specific extreme impact load value after statistical analysis. This represents the probability of an extreme impact load value occurring.

3. The method for rapid fatigue life assessment of steel trestle bridges under extreme heavy-load impact as described in claim 2, characterized in that: In step three, the correction process of the finite element model includes: S3.1 Selecting the correction vector Includes the elastic modulus of steel components Density of steel materials Vertical stiffness of supports and structural damping ratio Four sensitivity parameters; S3.2, Define the relative sensitivity coefficient for: In the formula, For the correction vector The first in One parameter, For the output vector The first in One parameter, express The parameter change value, express The parameter change value; S3.3 Setting a threshold ,reserve The parameters are used as optimization variables; S3.

4. Assign weights to the three sensors and standardize them, setting: In the formula, , and These represent the weighting coefficients for the three sensors, respectively. , and These represent the covariance estimates for the three sensors, respectively. S3.5, Construct the objective function as follows: In the formula, Represents the L2 norm, for The The parameters are obtained by iteratively solving the objective function, and the corrected parameters are then imported into the finite element model.

4. The rapid fatigue life assessment method for steel trestle bridges subjected to extreme heavy-load impacts according to claim 3, characterized in that: In step three, the output set is... Represented as: In the formula, For the first Node spatial coordinates , For the set of Number of applied stress cycles , For the set of Crack length , , , These are the maximum values ​​of the crack tip stress intensity factor. Minimum value and amplitude , For the set of Equivalent stress amplitude , .

5. The method for rapid fatigue life assessment of steel trestle bridges under extreme heavy-load impact as described in claim 4, characterized in that: In step four, the simulation process of the fatigue damage field includes: S4.1 Construct the physical constraint equations for the fatigue damage field, expressed as: In the formula, Indicates fatigue damage value. Let be the material loss constant. The yield strength of the material. An index representing the degree to which stress amplitude affects damage rate. An index representing the degree of nonlinearity controlling damage evolution. Indicates the damage diffusion index; In S4.2, the PINN machine learning model: Input layer: Output layer: In the formula, For fracture toughness, This indicates the range of uncertainty in fatigue damage values. Indicates the length of the hot spot crack. This indicates the range of uncertainty regarding the length of the hot spot crack. The length of the crack microstructure. The initial length of the crack. This is the crack transition length. The critical crack length. This represents the effective stress intensity factor amplitude. S4.3, The loss function is expressed as follows: In the formula, These are the physical constraint weighting coefficients. These are the initial threshold term weight coefficients. For the weighting coefficient of the transition trigger term, For the fracture criterion term, the weight coefficient is... This is the toughness proportionality coefficient. The minimum stress intensity factor amplitude required for a crack to begin stable propagation.

6. The method for rapid fatigue life assessment of steel trestle bridges under extreme heavy-load impact as described in claim 5, characterized in that: In step five, the stress cycle number for the three stages of crack propagation is expressed as follows: Infancy stage: In the formula, For cracks from the microstructure length Extended to initial length Required number of stress cycles, and These are the crack growth constant and the exponent, respectively. For Coffin–Manson parameter functions, For mapping index, For equivalent amplitude, The fatigue strength coefficient, The elastic modulus of the material. The fatigue ductility coefficient, The high-cycle fatigue index. Low-cycle fatigue index; Stable expansion phase: In the formula, For the crack from the initial length Extended to transition length Required number of stress cycles, Fatigue damage value Influenced material constants Fatigue damage value The damage constant affected; Rapid expansion phase: In the formula, For the crack from the transition length Extended to critical crack length Required number of stress cycles, The stress intensity factor under maximum load. For effective fracture toughness, This is a parameter representing the sensitivity of crack propagation to load. The correction parameter is used to prevent crack propagation from approaching the threshold. This is the effective crack propagation threshold.

7. The method for rapid fatigue life assessment of steel trestle bridges under extreme heavy-load impact as described in claim 6, characterized in that: In step six, the three damage indicators are expressed as follows: Damage indices of the microscopic damage field: Damage indicators of macroscopic crack evolution: Damage indicators during crack propagation: In the formula, Indicates hotspot area, Indicates the weight of the hotspot area. for Any point in the, For sensitivity parameters; The unified damage index is then expressed as: In the formula, , , The weights of the three indicators are respectively, and they satisfy the following conditions: , This is the sensitivity weighting adjustment coefficient. .

8. The method for rapid fatigue life assessment of steel trestle bridges under extreme heavy-load impact as described in claim 7, characterized in that: Step seven, the remaining life assessment and multidimensional risk grading, includes: S7.1, The number of cycles that the remaining lifetime can withstand is , This refers to the number of cycles that the system can withstand under life-limit conditions. Predict remaining lifespan based on the number of cycles already completed. , The average load cycle rate of the steel trestle bridge; S7.2, Define a dimensionless term Calculated by the following formula: In the formula, For residual stress, This represents the cumulative probability of an extreme shock occurring. For reference lifespan, To allow stress, , , These are the weights for the life term, residual stress term, and extreme probability, respectively. S7.3, Preset Threshold , , ,according to , , and The steel trestle bridge is divided into four risk levels, from low to high, based on the degree of risk.

Images