Method for constructing cohesive force mixed damage model of adhesive interface of combined steel bridge deck

Abstract

The invention discloses a method for constructing a cohesive force mixed damage model of an adhesive interface of a combined steel bridge deck slab. The method comprises the following steps: step 1, determining an interface cohesive force mixed damage model and a damage judgment criterion; step 2, determining inversion parameters in a mixed damage mode; 3, constructing a numerical simulation response surface based on a neural network, and evaluating a training effect through a mean square error and a mean absolute error until convergence; 4, constructing a target function of full-field relative slip data based on a digital image correlation technology to evaluate the accuracy of a response surface; 5, optimizing and determining unknown parameters in the cohesion mixed damage model by adopting a genetic algorithm, and obtaining an optimal solution of the cohesion mixed damage model of the adhesive interface of the combined steel bridge deck; by means of the method, efficient inversion calculation of the cohesion mixed damage model capable of reflecting complex mechanical behaviors between the interfaces is achieved, and technical support is provided for combined steel bridge deck adhesive interface accumulated damage evolution and interface failure analysis and research.

Description

Technical Field

[0001] The present invention belongs to the technical field of bridge engineering, and particularly relates to a method for constructing a cohesive mixed damage model of the adhesive interface of a composite steel bridge deck. Background Art

[0002] Steel bridge decks are widely used in medium-span and long-span bridge structures due to their high load-bearing capacity, light self-weight, and high degree of industrialization. With the gradual in-depth research of scholars in various countries on the fatigue mechanism of steel bridge decks, it is found that: improving the local stiffness is an effective way to improve the fatigue resistance and extend the service life of steel bridge decks. At present, a common practice is to set a composite layer with high strength, high toughness, and lightweight characteristics on the top plate of the steel bridge deck to form a composite steel bridge deck. To ensure the coordinated work between the composite stress layer and the steel bridge deck, it is crucial to adopt a reliable composite connection method. Among them, the adhesive composite method has a large interfacial shear stiffness, high composite efficiency, does not introduce welding residual stress, and is non-destructive to the steel structure part, with significant technical advantages.

[0003] The cohesive force model has been widely used in simulating interfacial damage and failure behaviors, and its accuracy has also been verified by numerous experiments. When constructing a numerical analysis model of a composite steel bridge deck, it is crucial to select a constitutive model of cohesive force units that can accurately reflect the mechanical behavior of the toughened epoxy bonding interface. At present, the determination method of cohesive force parameters mainly adopts experimental and computational analysis verification. There is still a need to improve the trial calculation efficiency and accuracy in the process of determining cohesive force parameters, and it has a weak applicability to the cohesive mixed damage model with more unknown parameters, which cannot meet the technical requirements for studying the interfacial force transfer mechanism of composite steel bridge decks.

[0004] The present invention proposes a method for constructing a cohesive mixed damage model of the adhesive interface of a composite steel bridge deck. This method establishes a total cohesive failure equation under the mixed mode based on the BK criterion to calculate the complex stress behavior of the interface, uses the full interface relative slip database monitored by digital image correlation technology, obtains the neural network response surface through finite element numerical calculation, establishes the mapping relationship between the cohesive force parameters and the relative slip in combination with the neural network response surface, constructs the objective function optimization model required for the genetic algorithm, and efficiently inversely calculates the cohesive force model that can characterize the interfacial force transfer behavior in combination with the genetic algorithm, improving the inversion calculation accuracy and efficiency of the key parameters of the interfacial cohesive force model, realizing the inversion calculation of the cohesive mixed damage model with multiple unknown parameters, and improving the numerical calculation efficiency and simulation calculation accuracy of the adhesive interface of the composite steel bridge deck. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the defects of the prior art, such as low efficiency in determining the cohesion parameters, insufficient basis for trial calculation, and low applicability to multi-unknown parameter models, and to provide a method for constructing a cohesive mixed damage model of the adhesive interface of a composite steel bridge deck based on a neural network response surface and a genetic algorithm, aiming to provide a reliable calculation basis for accurately simulating the cohesive mixed damage model of the adhesive interface of a composite steel bridge deck, and supporting the accurate and efficient analysis of the force transmission mechanism research of the adhesive interface of a composite steel bridge deck.

[0006] The technical solution adopted to solve the above technical problem is: a method for constructing a cohesive mixed damage model of the adhesive interface of a composite steel bridge deck, including the following steps:

[0007] Step 1, determine the cohesive mixed damage model and damage judgment criterion: The cohesive mixed damage model includes cohesive damage models of three single failure modes in the normal direction, the first shear direction, and the second shear direction; use the quadratic nominal stress criterion as the damage initiation time judgment criterion, which is specifically expressed as the following formula

[0008]

[0009] In the formula, <t n > represents the cohesion that does not cause damage in the normal compression behavior, t s represents the cohesion in the first shear direction, t t represents the cohesion in the second shear direction, represents the maximum cohesion at the damage initiation time in the normal direction, represents the maximum cohesion at the damage initiation time in the first shear direction, represents the maximum cohesion at the damage initiation time in the second shear direction;

[0010] Use the BK criterion as the damage evolution stage judgment criterion, which is specifically expressed as the following formula

[0011]

[0012] In the formula, G IC represents the cohesive energy of a single normal mode, G IIC represents the cohesive energy of a single first shear mode, G shear represents the sum of the cohesive energies of the total shear modes, G T represents the total fracture energy, η is the BK criterion parameter, G C represents the critical fracture energy when the BK criterion judges failure;

[0013] Step 2, determine the inversion parameters in the mixed damage mode: By determining the relative displacement of the mixed mode at the start of damage The relative displacement of the mixed mode at damage failure The linear softening damage function D determines the unknown parameters to be determined in the mixed damage model during the damage evolution process. The unknown parameters include: the stiffness E in the normal mode nn , the stiffness E in the first shear mode ss or the stiffness E in the second shear mode tt , the maximum cohesive force at the onset of normal damage the maximum cohesive force at the onset of damage in the first shear direction or the maximum cohesive force at the onset of damage in the second shear direction the cohesive energy G in a single pure normal mode IC , the cohesive energy G in a single first shear mode IIC or the cohesive energy G in a single second shear mode IIIC , the BK criterion parameter η that affects the proportion of the mixed failure mode. Among them, the cohesive force model parameters in the first shear direction and the second shear direction of the toughened adhesive interface of quartz sand are the same;

[0014] Step 3: Construct a numerical simulation response surface based on a neural network: Establish a finite element model identical to the physical test conditions, use the Latin hypercube sampling method to explore the cohesive force model parameter space and automatically update the parameters of the cohesive force mixed damage model for numerical analysis, construct a response surface between the parameters of the cohesive force mixed damage model and the relative slip value of the finite element model, and evaluate the training effect through the mean square error and the mean absolute error during the neural network training until convergence, so as to obtain the relative slip prediction value s corresponding to the corresponding load through the response surface * ;

[0015] Step 4: Construct an objective function for the full-field relative slip data based on digital image correlation technology to evaluate the accuracy of the response surface: Use the slip database of the monitoring interface of digital image correlation technology in the physical test as the test observation value, and obtain the target function value through the root mean square of the difference between the relative slip prediction value s * corresponding to the corresponding load obtained by the neural network and the slip value s measured in the test, which is specifically expressed as the following formula

[0016]

[0017] In the formula, λ is the inversion parameter group, representing the combination of 7 parameters to be determined in step 2; N is the total number of measurement values under the displacements of all loading points, M is the total number of different loading point displacements, and n j is the number of reference points participating in the inversion calculation, s i,j is the slip value measured parallel to the physical experiment at the position of the i-th reference interface corresponding to the displacement of the j-th loading point, and s i,j * is the relative slip prediction value corresponding to the corresponding load obtained through the response surface at the position of the i-th reference interface corresponding to the displacement of the j-th loading point;

[0018] Step 5. Optimize and determine the unknown parameters in the cohesive force mixed damage model by using the genetic algorithm: Select the genetic algorithm as the optimization algorithm of the system, and control the automatic update of the cohesive force model parameters through the genetic variation of the algorithm. Finally, obtain the optimal solution of the parameter group to be determined, and substitute the optimal solution into the mixed damage model in Step 1 to establish the final damage mixed model form.

[0019] In Step 2 of the present invention, the mixed damage model, the quadratic nominal stress criterion and the BK criterion are used to obtain the relative displacement of the mixed mode at the beginning of damage. The relative displacement of the mixed mode at damage failure Specifically expressed as the following formula

[0020]

[0021] In the formula, is the relative displacement at the starting moment of the normal mode damage, is the relative displacement at the starting moment of the total shear mode damage, β is the mixed mode ratio, K is the stiffness parameter, is the actual relative displacement at the failure of the single first shear mode damage, is the actual relative displacement at the failure of the second shear mode damage, δ shear is the sum of the relative displacements in the actual first and second shear directions, δ n is the relative displacement of the actual single normal direction;

[0022] Then introduce the linear softening damage function D, specifically expressed as the following formula

[0023]

[0024] In the formula, refers to the maximum value of the effective displacement reached during the loading history;

[0025] Determine the damage evolution process through the linear softening damage function D defined based on energy or effective displacement. The value range of D is 0 to 1.

[0026] In Step 5 of the present invention, optimize the parameter configuration of the force transmission and damage evolution process at the adhesive combination interface. The steps to finally obtain the optimal solution are as follows;

[0027] Step 5.1. Initialize the population: Generate an initial population containing 100 random individuals. Each individual represents a cohesive force model, and 7 parameters represent 7 genes. The 7 genes constitute an individual;

[0028] Step 5.2, Fitness Evaluation and Selection: Calculate the fitness of each individual, evaluate its performance through the objective function in Step 4. The smaller the calculation result of the objective function, the better the fitness. The tournament selection method is used to select the intermediate population for generating the next generation;

[0029] Step 5.3, Crossover: Set a crossover probability of 0.5. Based on the intermediate population, the probability of exchanging gene segments between two individuals is set to 50% to generate new individuals and increase the diversity of the population;

[0030] Step 5.4, Mutation: Set a mutation probability of 0.0005, randomly modify some genes of the individuals in the intermediate population to introduce new mutations and explore different regions of the cohesive force model parameter space;

[0031] Step 5.5, Generate a New Population: Take the new individuals generated after the crossover and mutation operations as the new population, ensuring that the parameter values of all new individuals are within the preset range;

[0032] Step 5.6, Iterative Evolution: Repeat Steps 5.2 to 5.5 for 100 generations of evolution, and select the optimal individual in each generation;

[0033] Step 5.7, Termination Condition: Terminate when the evolution reaches 100 generations or terminate in advance according to the convergence of fitness;

[0034] Step 5.8, Select the individual with the highest fitness as the optimal solution.

[0035] In Step 3 of the present invention, the Latin hypercube sampling method is used to explore the cohesive force model parameter space, automatically update the cohesive force mixed damage model parameters, and conduct numerical analysis to construct a numerical model. The specific steps are as follows:

[0036] Step 3.1, Investigate relevant model tests to determine the specific ranges of the seven parameters of the cohesive force damage model;

[0037] Step 3.2, Set the following basic logical relationships for the parameters:

[0038] E ss > E nn

[0039]

[0040] G IIC > G IC

[0041] Step 3.3, Set the logical relationship between the parameters as the boundary condition for generating random numbers, and start generating random values of the parameters;

[0042] Step 3.4. Group the generated random parameters, import them into finite element software, and construct a numerical model.

[0043] After successfully constructing the numerical model in Step 3 of the present invention, the parameters in the generated cohesive force parameter space are incorporated into the numerical model one by one through writing Python scripts for automatic calculation to obtain excellent training results. A neural network is used to replace the heavy finite element calculation tasks in the subsequent parameter inversion process, which specifically includes the following steps:

[0044] Step 3.4.1. Use the TensorFlow neural network library in Python to construct a neural network and set it to have two hidden layers.

[0045] Step 3.4.2. Set the input layer: It consists of 8 nodes, namely 7 cohesive force parameters and the displacement values obtained by loading the finite element model under the control of the corresponding 7 parameters according to displacement.

[0046] Step 3.4.3. Set the output layer: It consists of 19 nodes, namely the load and the slip values at 18 monitoring and analysis positions related to digital image technology.

[0047] Step 3.4.4. Set the rectified linear unit activation function in the hidden layer.

[0048] Step 3.4.5. Normalize the calculation results and divide them into a training set, a test set, and a validation set; construct a response surface between the parameters of the cohesive force mixed damage model and the relative slip values of the finite element model.

[0049] The present invention has the following advantages compared with the prior art:

[0050] (1) At present, the interface cohesive force model with a single failure mode is commonly used to simulate the behavior of the adhesive interface of composite steel bridge decks. However, the actual composite interface is affected by the coupling of failure modes in 3 directions, namely the normal direction, the first shear direction, and the second shear direction. The cohesive force model with a single failure mode cannot accurately reflect the actual interface behavior and does not meet the requirements of complex interface analysis for interface structure anisotropy. In view of the above technical deficiencies, the present invention proposes an interface cohesive force mixed damage model considering the coupling effect of 3 - direction failure modes and establishes a failure criterion under the mixed damage mode, effectively improving the accuracy of the simulation calculation of the complex interface behavior of composite steel bridge decks.

[0051] (2) At present, the determination method of cohesion parameters mainly adopts tests and numerical calculations for verification, which can adapt to the cohesion models of single failure modes with fewer unknown parameters, but has weak applicability to the cohesion mixed damage models with more unknown parameters and complex coupling of failure modes. In addition, the trial calculation efficiency and accuracy in the existing process of determining cohesion parameters are still low, and cannot meet the precise requirements for the research on the interface force transfer mechanism of composite steel bridge decks. In view of the above technical deficiencies, the present invention proposes a method for constructing a cohesion mixed damage model based on neural network response surface and genetic algorithm, realizing the efficient calculation of the cohesion mixed damage model with multiple unknown parameters, and overcoming the technical barriers of low efficiency in determining the parameters of the cohesion mixed damage model and insufficient basis for trial calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is the construction flow chart of the present invention.

[0053] Figure 2 is a schematic diagram of the cohesion mixed damage mode model constructed by the present invention.

[0054] Figure 3 is the optimization convergence process diagram.

[0055] Figure 4 is the load-slip curve diagram of parallel tests and calculations using the cohesion mixed damage model. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] The present invention will be further described in detail below with reference to the drawings and embodiments, but the present invention is not limited to these embodiments.

[0057] Embodiment 1

[0058] In Figure 1 , the method for constructing the cohesion mixed damage model of the adhesive interface of the composite steel bridge deck involved in the present invention includes the following steps:

[0059] Step 1. Determine the cohesion mixed damage model of the interface and the damage judgment criterion: The cohesion mixed damage model is as shown in Figure 2 , and includes the cohesion damage models of three single failure modes: the normal direction (mode I, the letter subscript in this direction is n), the first shear direction (mode II, the letter subscript in this direction is s), and the second shear direction (mode III, the letter subscript in this direction is t); in the figure, t represents the cohesion (MPa);

[0060]

[0061] t n 、t s 、t t respectively represent the cohesion in the normal direction, the first shear direction, and the second shear direction;

[0062] They respectively represent the maximum cohesive force at the moment of damage initiation in the normal direction, the first shear direction, and the second shear direction;

[0063] <t n > represents the cohesive force at which no damage occurs in the normal compression behavior;

[0064] represents the cohesive force at the moment of damage initiation in the two shear directions;

[0065] δ n 、δ s 、δ t They respectively represent the relative displacements in the normal direction, the first shear direction, and the second shear direction;

[0066] δ shear is the sum of the relative displacements in the first shear direction and the second shear direction,

[0067] δ m is the mixed-mode relative displacement, with the subscript m indicating the mixed mode, and is calculated according to the following formula;

[0068]

[0069] is the relative displacement at the moment of damage initiation in the mixed mode;

[0070] They respectively represent the relative displacements at the moment of damage initiation in the normal mode, the first and second shear modes in the mixed mode;

[0071] They respectively represent the relative displacements at the moment of damage failure in the normal mode, the first and second shear modes in the mixed mode.

[0072] represents the relative displacement at the moment of damage failure in the two shear modes in the mixed mode;

[0073] The cohesive force mixed damage model is usually divided into two stages, damage initiation and damage evolution, in the process of simulating material failure. In the damage initiation stage, the quadratic nominal stress criterion is used as the damage initiation judgment criterion to describe the weakening of material stiffness, which is specifically expressed as the following formula

[0074]

[0075] The damage evolution stage describes the degradation process of the subsequent mechanical properties of the material after damage occurs. The BK criterion is used to judge damage failure, which is specifically expressed as the following formula

[0076]

[0077] In the formula, G IC represents the cohesive energy in a single normal mode, G IIC represents the cohesive energy in a single first shear mode, G shear represents the sum of the cohesive energies in the total shear mode, G T represents the total fracture energy, η is the BK criterion parameter, G C represents the critical fracture energy when the BK criterion judges failure; among them, the first and second shear direction cohesion model parameters of the toughened adhesive interface of quartz sand are the same, that is, the cohesive energy G IIC in a single first shear mode and the cohesive energy G IIIC in a single second shear mode are equal

[0078] Step 2. Determine the inversion parameters in the mixed damage mode: By determining the relative displacement of the mixed mode at the start of damage the relative displacement of the mixed mode at damage failure and the linear softening damage function D, determine the unknown parameters to be determined in the mixed damage model during the damage evolution process. The unknown parameters include: the stiffness E nn in the normal mode, the stiffness E ss in the first shear mode or the stiffness E tt in the second shear mode, the maximum cohesive force at the start of normal damage the maximum cohesive force at the start of damage in the first shear direction or the maximum cohesive force at the start of damage in the second shear direction the cohesive energy G IC in a single pure normal mode, the cohesive energy G IIC in a single first shear mode or the cohesive energy G IIIC in a single second shear mode, the BK criterion parameter η affecting the proportion of the mixed failure mode, among which, the first shear direction and the second shear direction cohesion model parameters of the toughened adhesive interface of quartz sand are the same; that is, the cohesive energy G IIC in a single first shear mode and the cohesive energy G IIIC in a single second shear mode are equal, the maximum cohesive force at the start of damage in the first shear direction or the maximum cohesive force at the start of damage in the second shear direction are equal, the stiffness E ss in the first shear mode or the stiffness E tt in the second shear mode are equal.

[0079] Among them, the relative displacement of the mixed mode at the start of damage and the relative displacement of the mixed mode at damage failure are obtained by using the mixed damage model, the quadratic nominal stress criterion and the BK criterion, and are specifically expressed as the following formula

[0080]

[0081] In the formula, is the relative displacement at the normal mode damage initiation time, is the relative displacement at the total shear mode damage initiation time, β is the mixed mode ratio, and K is the stiffness parameter. is the actual relative displacement at the failure of the single first shear mode damage, is the actual relative displacement at the failure of the second shear mode damage, δ shear is the sum of the relative displacements in the actual first and second shear directions, δ n is the relative displacement of the actual single normal direction;

[0082] Determine the mixed mode relative displacements corresponding to damage initiation and failure and After that, introduce the linear softening damage function D, which is specifically expressed as the following formula

[0083]

[0084] In the formula, refers to the maximum value of the effective displacement reached during the loading history.

[0085] Determine the damage evolution process through the linear softening damage function D defined based on energy or effective displacement. The value range of D is 0 to 1. When D is 0, it means the material is intact; when D is 1, it means the material is completely damaged.

[0086] Step 3: Construct a numerical simulation response surface based on a neural network: Establish a finite element model identical to the physical test conditions, use the Latin hypercube sampling method to explore the cohesive model parameter space and automatically update the cohesive mixed damage model parameters for numerical analysis, construct a response surface between the cohesive mixed damage model parameters and the relative slip value of the finite element model, and evaluate the training effect through the mean square error and mean absolute error during the neural network training until convergence, so as to obtain the relative slip prediction value s corresponding to the corresponding load through the response surface * ;

[0087] Specifically, use finite element software to construct a numerical model of a super high performance fiber reinforced cement based composite steel bridge deck bonded with a toughened silica sand adhesive layer that is consistent with the test conditions, apply cohesive elements to simulate the force transmission behavior and damage accumulation and evolution process of the interface, use the Latin hypercube sampling method to explore the cohesive model parameter space and automatically update the cohesive mixed damage model parameters for numerical analysis to construct a numerical model. This method can generate uniformly distributed and highly representative sample points in the multi-dimensional parameter space. The specific steps are as follows:

[0088] Step 3.1. Conduct research on relevant model tests to determine the specific ranges of the seven parameters of the cohesive force damage model;

[0089] Step 3.2. Set the basic logical relationships of the parameters as follows:

[0090] E ss >E nn

[0091]

[0092] G IIC >G IC

[0093] Step 3.3. Set the logical relationships between the parameters as the boundary conditions for generating random numbers, and start generating random values of the parameters;

[0094] Step 3.4. Group the generated random parameters, import them into finite element software, and construct a numerical model.

[0095] After successfully constructing the numerical model, the parameters in the generated cohesive force parameter space are incorporated into the numerical model one by one through writing Python scripts for automatic calculation to obtain excellent training results. Use a neural network to replace the heavy finite element calculation tasks in the subsequent parameter inversion process, which specifically includes the following steps:

[0096] Step 3.4.1. Use the TensorFlow neural network library in Python to construct a neural network and set it to have two hidden layers;

[0097] Step 3.4.2. Set the input layer: It consists of 8 nodes, namely 7 cohesive force parameters and the displacement values obtained by loading the finite element model under the control of the corresponding 7 parameters according to displacement;

[0098] Step 3.4.3. Set the output layer: It consists of 19 nodes, namely the load and the slip values at 18 monitoring and analysis positions related to digital image technology;

[0099] Step 3.4.4. Set the rectified linear unit activation function in the hidden layer; Selecting the rectified linear unit as the activation function can introduce non-linear changes, enhance the learning ability of the model, and its special function form can alleviate the training problem of gradient disappearance and improve the training effect.

[0100] Step 3.4.5. Normalize the calculation results and divide them into a training set, a test set, and a validation set; Construct a response surface between the parameters of the cohesive force mixed damage model and the relative slip values of the finite element model; among them, the training set is used to train the neural network, the test set is used to test the performance of the neural network, and the validation set is used to further verify the generalization ability of the network model.

[0101] Step 4: Construct an objective function for the full-field relative slip data based on digital image correlation technology to evaluate the accuracy of the response surface:

[0102] Use Python scripts to synchronize the data of the load and loading point collected by the static strain tester with the data of time and full-field displacement during the loading process obtained by digital image correlation technology, so as to obtain the slip relationship data between the loading point and the detection interface position of digital image correlation technology. Take the slip database of the monitoring interface of digital image correlation technology in the physical experiment as the experimental observation value, and obtain the relative slip prediction value s * corresponding to the measured slip value s in the experiment through the root mean square of the difference, and the objective function value is specifically expressed as the following formula

[0103]

[0104] In the formula, λ is the inversion parameter group, representing the 7 parameter combinations to be determined in Step 2; N is the total number of measured values under all loading point displacements, M is the total number of different loading point displacements, and n j is the number of reference points participating in the inversion calculation, s i,j is the slip value measured by parallel physical experiments at the position of the i-th reference interface corresponding to the j-th loading point displacement, s i,j * is the relative slip prediction value under the corresponding load obtained through the response surface of Step 3 at the position of the i-th reference interface corresponding to the j-th loading point displacement;

[0105] Step 5: Use the genetic algorithm to optimize and determine the unknown parameters in the cohesive force mixed damage model: Select the genetic algorithm as the optimization algorithm of the system, and control the automatic update of the cohesive force model parameters through the genetic variation of the algorithm. Finally, obtain the optimal solution of the parameter group to be determined, and substitute the optimal solution into the mixed damage model in Step 1 to establish the final damage mixed model form.

[0106] The steps to obtain the optimal solution for the above parameter configuration that optimizes the force transfer and damage evolution process of the adhesive combination interface are as follows;

[0107] Step 5.1: Initialize the population: Generate an initial population containing 100 random individuals, each individual represents a cohesive force model, and 7 parameters represent 7 genes, and 7 genes form an individual;

[0108] Step 5.2: Evaluate fitness and selection: Calculate the fitness of each individual, evaluate its performance through the objective function in Step 4, the smaller the calculation result of the objective function, the better the fitness, and use the tournament selection method to select the intermediate population for generating the next generation;

[0109] Step 5.3, Crossover: Set a crossover probability of 0.5. Based on the intermediate population, the probability of exchanging gene segments between two individuals is set to 50% to generate new individuals and increase the diversity of the population;

[0110] Step 5.4, Mutation: Set a mutation probability of 0.0005. Randomly modify some genes of the individuals in the intermediate population to introduce new mutations and explore different regions of the cohesive force model parameter space;

[0111] Step 5.5, Generate a new population: Use the new individuals generated after the crossover and mutation operations as the new population, ensuring that the parameter values of all new individuals are within the preset range;

[0112] Step 5.6, Iterative evolution: Repeat steps 5.2 to 5.5 for 100 generations of evolution, and select the optimal individual in each generation;

[0113] Step 5.7, Termination condition: The evolution terminates when it reaches 100 generations or is terminated in advance according to the fitness convergence situation;

[0114] Step 5.8, Select the individual with the highest fitness as the optimal solution.

[0115] Experiment 1

[0116] To verify the reliability of the method proposed in this application, an experiment on epoxy resin adhesive layer bonding steel - ultra - high performance fiber - reinforced cement - based composite slabs was carried out, and the parameters of the cohesive force mixed damage model were calculated according to the construction method of the cohesive force mixed damage model proposed in this application. The calculation results were compared with the parallel test results to test the reliability of the method proposed in this application.

[0117] Through the experiment on epoxy resin adhesive layer bonding steel - ultra - high performance fiber - reinforced cement - based composite slabs, experimental observation values were obtained. Based on the experimental observation values, the 5 - step technical solution of Example 1 was used to calculate the cohesive force mixed damage model. Among them, the convergence situation of the calculation results in step 5 is as Figure 3 . Figure 3 In it, the vertical axis represents the optimal fitness value in each generation, and the horizontal axis represents the number of genetic generations. Figure 3 It has good convergence efficiency. It can be seen that using the BK criterion to determine the total cohesive failure equation under the mixed damage mode in step two can well adapt to the test conditions of the composite steel bridge deck. The optimal parameters of the cohesive force mixed damage model calculated by using the technical solution of Example 1 are shown in Table 1.

[0118] Table 1 Optimal calculated values of the cohesive force mixed damage model

[0119]

[0120] Using the cohesive force mixed damage model obtained with the parameters in Table 1, finite element calculations were carried out to obtain the load-slip curve of the steel-ultra high performance fiber reinforced cementitious composite slab, as Figure 4 . From Figure 4 it can be seen that, compared with the parallel test results of the steel-ultra high performance fiber reinforced cementitious composite slab, during the process of the load increasing from 0 to the first peeling load of 134 kN, the errors between the two simulation results and the test results are controlled within 5%; after the first peeling occurs, the calculated results are slightly lower than the test results; the prediction of the interface damage behavior near the ultimate load is consistent with the test results; after reaching the ultimate load, the slip trend of the calculated results is consistent with the test results, both showing the slip platform phenomenon caused by the interface damage expansion. It can be seen that the calculated results are in good agreement with the test results, and the method for constructing the cohesive force mixed damage model of the adhesive interface of the composite steel bridge deck slab proposed in this application is reliable.

Claims

1. A method for constructing a mixed damage model of cohesive force at the adhesive interface of a composite steel bridge deck, characterized in that The following steps are involved: Step 1: Determine the interface cohesive mixed damage model and damage judgment criterion: The cohesive mixed damage model includes three single failure modes of cohesive damage models in the normal direction, the first shear direction, and the second shear direction; the quadratic nominal stress criterion is used as the criterion for judging the damage starting time, which is specifically expressed as follows: In the formula, <t n > represents the cohesion that does not cause damage in normal compression behavior, t s represents the cohesive force in the first shear direction, t t represents the cohesive force in the second shear direction, represents the maximum cohesive force at the beginning of normal damage, represents the maximum cohesive force at the beginning of damage in the first shear direction, It represents the maximum cohesive force at the beginning of damage in the second shear direction; The BK criterion is used as the criterion for determining the damage evolution stage, which is specifically expressed as follows: In the formula, G IC represents the cohesive energy of a single normal mode, G IIC represents the cohesive energy of the single first shear mode, G shear represents the sum of the total shear mode cohesive energy, G T represents the total fracture energy, η is the BK criterion parameter, G C It represents the critical fracture energy when judging failure by BK criterion; Step 2: Determine the inversion parameters under the mixed damage mode: by determining the mixed mode relative displacement at the onset of damage Mixed mode relative displacement at damage failure Linear softening damage function D is used to determine the unknown parameters of the hybrid damage model during the damage evolution process. The unknown parameters include: the stiffness E under the normal mode nn , the stiffness E in the first shear mode ss Or the stiffness E in the second shear mode tt , Maximum cohesion at the beginning of normal damage Maximum cohesion at the beginning of damage in the first shear direction Or the maximum cohesive force at the start of damage in the second shear direction Cohesive energy G of a single pure normal mode IC , single first shear mode cohesive energy G IIC Or single second shear mode cohesive energy G IIIC , the BK criterion parameter η that affects the proportion of mixed failure modes, where the cohesive model parameters of the quartz sand toughened adhesive interface in the first shear direction and the second shear direction are the same; Step 3: Construct a numerical simulation response surface based on a neural network: Establish a finite element model that is the same as the physical test condition, use the Latin hypercube sampling method to explore the cohesive force model parameter space and automatically update the cohesive force mixed damage model parameters for numerical analysis, and construct a response surface between the cohesive force mixed damage model parameters and the relative slip value of the finite element model. During the neural network training process, the training effect is evaluated by the mean square error and the mean absolute error until convergence, and the relative slip prediction value s under the corresponding load is obtained through the response surface. * ; Step 4: Construct an objective function of the full-field relative slip data based on digital image correlation technology to evaluate the accuracy of the response surface: The monitoring interface slip database of digital image correlation technology in physical experiments is used as the experimental observation value, and the relative slip prediction value s under the corresponding load is obtained through the neural network. * The objective function value is obtained by taking the root mean square of the difference between the slip value s measured by the test and the slip value s, which can be expressed as follows: Where λ is the inversion parameter group, representing the seven parameter combinations to be determined in step 2; N is the total number of measured values ​​under all loading point displacements, M is the total number of displacements at different loading points, and n is j is the number of reference points involved in the inversion calculation, s i,j is the slip value measured by the parallel physical experiment at the position of the i-th reference interface corresponding to the displacement of the j-th loading point, s i,j * is the predicted relative slip value under the corresponding load obtained by the response surface in step 3 at the i-th reference interface position corresponding to the j-th loading point displacement; Step 5. Use genetic algorithm to optimize and determine the unknown parameters in the cohesive force mixed damage model: Select genetic algorithm as the optimization algorithm of the system, control the automatic update of the cohesive force model parameters through the genetic variation of the algorithm, and finally obtain the optimal solution of the parameter group to be determined. Substitute the optimal solution into the mixed damage model of step 1 to establish the final damage mixed model form.

2. The method for constructing a mixed damage model of cohesive force at the adhesive interface of a composite steel bridge deck according to claim 1, characterized in that: In step 2, the mixed damage model, the quadratic nominal stress criterion and the BK criterion are used to obtain the mixed mode relative displacement at the beginning of damage. Mixed mode relative displacement at damage failure Specifically expressed as the following formula In the formula, is the relative displacement at the beginning of normal mode damage, is the relative displacement at the beginning of total shear mode damage, β is the mixed mode ratio, K is the stiffness parameter, is the actual relative displacement when the single first shear mode fails, is the actual relative displacement at the time of failure in the second shear mode, δ shear is the sum of the actual relative displacements in the first and second shear directions, δ n is the relative displacement in the actual single normal direction; Then the linear softening damage function D is introduced, which is specifically expressed as follows In the formula, Refers to the maximum effective displacement achieved during the loading history; The damage evolution process is determined by the linear softening damage function D defined based on energy or effective displacement, and the value of D ranges from 0 to 1.

3. The method for constructing a mixed damage model of cohesive force at the adhesive interface of a composite steel bridge deck according to claim 1 is characterized in that The step of optimizing the parameter configuration of the force transmission and damage evolution process of the adhesive composite interface in step 5 and finally obtaining the optimal solution is as follows; Step 5.1, Initialize the population: Generate an initial population of 100 random individuals, each individual represents a cohesion model, 7 parameters represent 7 genes, and 7 genes constitute an individual; Step 5.2, fitness evaluation and selection: Calculate the fitness of each individual, and evaluate its performance through the objective function of step 4. The smaller the result of the objective function calculation, the better the fitness. Use the tournament selection method to select the intermediate population for generating the next generation. Step 5.3, crossover: set the crossover probability to 0.5, and on the basis of the intermediate population, the probability of exchanging gene fragments between two individuals is set to 50% to generate new individuals and increase the diversity of the population; Step 5.4, mutation: set the mutation probability to 0.0005, randomly modify some genes of individuals in the intermediate population to introduce new mutations and explore different regions of the cohesion model parameter space; Step 5.5, generate a new population: take the new individuals generated after the crossover and mutation operations as the new population, and ensure that the parameter values ​​of all new individuals are within the preset range; Step 5.6, iterative evolution: repeat steps 5.2 to 5.5 for 100 generations of evolution, and select the best individual in each generation; Step 5.7, termination condition: the evolution terminates when it reaches 100 generations or terminates early according to the fitness convergence; Step 5.8: Select the individual with the highest fitness as the optimal solution.

4. The method for constructing a mixed damage model of cohesive force at the adhesive interface of a composite steel bridge deck according to claim 1 is characterized in that In step 3, the Latin hypercube sampling method is used to explore the cohesive force model parameter space and automatically update the cohesive force mixed damage model parameters to perform numerical analysis and construct a numerical model. The specific steps are as follows: Step 3.1, investigate relevant model tests to determine the specific ranges of the seven parameters of the cohesive damage model; Step 3.2: Set the basic logical relationship of parameters as follows: AND ss >And nn Step 3.3, set the logical relationship between the parameters as the boundary condition for generating random numbers, and start generating random values ​​of the parameters; Step 3.4: Group the generated random parameters, import them into finite element software, and construct a numerical model.

5. The method for constructing a mixed damage model of cohesive force at the adhesive interface of a composite steel bridge deck according to claim 4, characterized in that: After the numerical model is successfully constructed in step 3, the parameters in the generated cohesive force parameter space are incorporated into the numerical model one by one for automatic calculation by writing a Python script to obtain excellent training results. The neural network is used to replace the heavy finite element calculation tasks in the subsequent parameter inversion process. Specifically, the following steps are included: Step 3.4.1, use the tensorflow neural network library in Python to build a neural network and set it to have two hidden layers; Step 3.4.2, set the input layer: it consists of 8 nodes, which are 7 cohesion parameters and the displacement values ​​obtained after the finite element model is loaded according to the displacement under the control of the corresponding 7 parameters; Step 3.4.3, set the output layer: it consists of 19 nodes, which are the load and the slip value of 18 digital image correlation technology monitoring and analysis positions; Step 3.4.4, set the linear rectification function activation function in the hidden layer; Step 3.4.5: Normalize the calculation results and divide them into training set, test set and validation set; construct the response surface between the cohesive mixed damage model parameters and the relative slip value of the finite element model.

Images