Time-varying vulnerability and residual seismic capacity assessment method for existing steel columns based on machine learning

Abstract

The method for assessing the time-varying vulnerability and residual seismic capacity of existing steel columns based on machine learning includes the following steps: Step 1: construct a physical test model; Step 2: establish a three-dimensional refined finite element model of steel frame columns; Step 3: determine the component vulnerability function, sort out the process of constructing the time-varying vulnerability model of existing corroded steel frame columns; Step 4: divide the damage and destruction levels of existing corroded steel frame columns and determine the corresponding quantitative limits. Step 5: construct a parameter data set of the time-varying vulnerability function of existing corroded steel frame columns, and establish a time-varying vulnerability curve of existing multi-age steel frame columns under offshore atmospheric environment; Step 6: obtain the analysis results of the residual seismic capacity; Step 7: establish an optimal assessment model for the residual seismic capacity of earthquake-damaged existing multi-age steel frame columns; Step 8: realize a rapid and reasonable assessment of its residual seismic capacity. The present invention provides a technical basis for engineering and technical personnel engaged in earthquake damage early warning, assessment and reinforcement of existing engineering structures.

Description

Technical Field

[0001] The present invention belongs to the technical field of seismic capacity assessment, and specifically relates to a method for assessing the time-varying vulnerability and residual seismic capacity of existing steel columns based on machine learning. Background Art

[0002] Steel structure, as a structure with metal materials as the main load-bearing material, has the disadvantage of poor corrosion resistance. Many steel structure buildings that have been in corrosive media environments for a long time (the "four major corrosive media" of ocean, atmosphere, soil, and chemical industry) are usually difficult to avoid rust through protection, structural measures and maintenance systems, which leads to durability problems. At the same time, my country is a country with many earthquakes, and a large number of steel structures in corrosive media environments are also in high seismic fortification intensity areas. Therefore, the evaluation of the seismic performance of steel structures under the coupling of environmental erosion and seismic loads is an urgent problem to be solved.

[0003] In recent years, with the continuous deepening of the concept of performance-based earthquake engineering (PBEE), the probability of structural collapse and casualties in earthquakes has been significantly reduced, but a large number of structures that have not collapsed have suffered serious damage and are difficult to repair or have no repair value, forming a "standing ruins" phenomenon. For this reason, researchers have proposed the concept of "seismic recoverability", which characterizes the ability of cities or structures to resist earthquake disaster disturbances and recover quickly after earthquakes. It is also considered to be the core of the next generation of performance-based earthquake engineering.

[0004] The component vulnerability model is a key part of the structural seismic recoverability assessment, which characterizes the conditional probability of a component reaching or exceeding a specific damage state (DS) under a given engineering demand parameter (EDP). It quantifies the seismic performance of components; at the same time, by defining the repair methods and cost information corresponding to different damage states, it can predict the reinforcement plan and repair cost of earthquake-damaged structures in the later stage, providing (basic) support for the evaluation of structural seismic toughness (seismic recoverability assessment). However, existing studies ignore the influence of factors such as construction history, service environment and age on the seismic capacity of existing structural components, and cannot achieve scientific quantification of the actual seismic capacity of structural components.

[0005] The prediction of the residual seismic capacity of earthquake-damaged components and structures is also an important part of the seismic recoverability assessment. It is an important basis for the restoration and reconstruction of urban or structural functions after disasters, and is of great significance to improving my country's disaster prevention and mitigation capabilities and scientific decision-making level. At present, earthquake-prone countries such as the United States, Japan, Greece, and China have carried out a large number of studies on the seismic capacity assessment methods of earthquake-damaged buildings. However, the existing residual seismic capacity assessment is too theoretical, and the correlation theory between earthquake damage characteristics and residual seismic capacity is missing. The uncertainty of the structure itself and durability damage is not considered. The assessment is time-consuming and relies on on-site investigations by professionals. These problems restrict the effective implementation of post-earthquake performance evaluation and safety appraisal work.

[0006] Therefore, there is an urgent need for a new technology that takes into account the service environment and age, and can scientifically and efficiently continuously evaluate and predict the vulnerability of steel frame columns throughout their life cycle and their residual seismic capacity after an earthquake. This technology has broken through many problems faced by traditional steel structure seismic performance assessment. Summary of the invention

[0007] In order to overcome the above problems existing in the prior art, the purpose of the present invention is to provide a method for evaluating the time-varying vulnerability and residual seismic capacity of existing steel columns based on machine learning. The evaluation method can improve the intelligence and reliability of "revealing the evolution law of the seismic performance of existing steel structures before an earthquake - evaluating the residual seismic capacity of damaged steel structures after an earthquake", provide theoretical support for the effective implementation of seismic performance evaluation, earthquake damage prediction and post-earthquake safety appraisal of existing steel structures in earthquake-prone areas, and provide a technical basis for engineering and technical personnel engaged in earthquake damage early warning, assessment and reinforcement of existing engineering structures.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is:

[0009] The method for assessing time-varying fragility and residual seismic capacity of existing steel columns based on machine learning includes the following steps;

[0010] Step 1: Construct a physical test model and carry out low-cycle reciprocating loading tests on corroded steel frame columns in offshore atmospheric environment to reveal the seismic damage mechanism and seismic performance degradation law of existing corroded steel frame columns in offshore atmospheric environment;

[0011] Step 2: Based on the physical test model in step 1, a three-dimensional refined finite element model of the steel frame column considering corrosion damage and ductile metal damage criteria is established, and the accuracy and effectiveness of the established numerical model are verified by comparing the test results;

[0012] Step 3: Determine the component vulnerability function and sort out the process of constructing the time-varying vulnerability model of the existing corroded steel frame columns;

[0013] Step 4: Select the engineering requirement parameter EDP index that can characterize the seismic performance of existing corroded steel frame columns; classify the damage levels of existing corroded steel frame columns and determine the corresponding quantitative limits.

[0014] Step 5: Combining the physical test data in step 1 with the numerical simulation analysis data of the three-dimensional refined finite element model of the steel frame column in step 2, and following the component vulnerability function establishment method in steps 3 and 4, a parameter data set of the time-varying vulnerability function of the existing corroded steel frame columns is constructed, and a time-varying vulnerability curve of the existing multi-age steel frame columns under the offshore atmospheric environment is established;

[0015] Step 6: Based on the three-dimensional refined finite element model of the steel frame column established in step 2 and the damage level determined in step 4, the evolution law of the residual seismic capacity of the existing corroded steel frame columns under different damage levels, different service ages, and different design parameters is revealed, and the residual seismic capacity analysis results are obtained;

[0016] Step 7: Based on the residual seismic capacity analysis results of step 6, a BP neural network model optimized by particle swarm optimization is used to build a nonlinear mapping relationship between the design parameters, service age, damage degree and residual stiffness, residual bearing capacity and residual energy dissipation of existing corroded steel frame columns, and establish an optimal evaluation model for the residual seismic capacity of earthquake-damaged existing multi-age steel frame columns;

[0017] Step 8: Combining the fragility curve in step 5 with the optimal assessment model of residual seismic capacity in step 7, the residual seismic capacity of existing multi-age steel frame columns under specific horizontal displacement in the offshore atmospheric environment is given in a probabilistic sense, so as to achieve a rapid and reasonable assessment of its residual seismic capacity.

[0018] Furthermore, in step 1, referring to existing specifications and engineering examples, steel frame column specimens are designed. The specimens are all made of hot-rolled H-shaped steel. According to the Artificial Atmosphere Corrosion Test-Salt Spray Test (GB / T 10125-2012), the ZHT / W2300 climate environment simulation system is used to conduct salt spray accelerated corrosion on all specimens. Then, low-cycle reciprocating loading tests are conducted on the corroded specimens.

[0019] Through the tensile performance test of corroded steel, the least square method was used to perform regression analysis on the test results, and the functional relationship between the mechanical performance index of corroded steel and its corrosion degree (weight loss rate) in offshore atmospheric environment was obtained.

[0020]

[0021] Where: f y0 、f u0 , δ0, and E0 are the yield strength, ultimate strength, elongation and elastic modulus of the steel before corrosion, respectively; and f y 、fu , δ, and E are the corresponding mechanical properties of steel after corrosion; η is the weight loss rate (less than 14%);

[0022] Through the test failure morphology, hysteresis curve, skeleton curve, stiffness degradation, ductility and energy dissipation analysis of corroded steel frame columns, the influence of corrosion degree on the seismic performance of existing steel frame columns is revealed. It shows that with the increase of corrosion degree, the bearing capacity, deformation capacity and energy dissipation capacity of existing steel frame columns gradually decrease.

[0023] Furthermore, in step 2, a three-dimensional refined finite element model considering corrosion damage and ductile metal damage criteria was established, and a numerical simulation of the test steel frame column was performed. The corrosion damage was divided into uniform corrosion and pitting corrosion to comprehensively consider the different mechanisms of their influence on the performance of steel components:

[0024] (1) Uniform corrosion is reflected in the reduction of plate thickness, that is, the reduction of component cross-sectional dimensions;

[0025] (2) Pitting is manifested in the reduction of mechanical properties of materials, that is, the functional relationship between the yield strength, ultimate strength, elastic modulus, and elongation of steel and the weight loss rate is introduced;

[0026]

[0027] Where: f y0 、f u0 , δ0, and E0 are the yield strength, ultimate strength, elongation and elastic modulus of the steel before corrosion, respectively; and f y 、f u , δ, and E are the corresponding mechanical properties of steel after corrosion; η is the weight loss rate (less than 14%);

[0028] Among them, the maximum residual thickness is used to determine the mechanical performance index;

[0029] In addition, in order to accurately reflect the bearing capacity and stiffness degradation characteristics of steel components and structures caused by the accumulation of plastic strain under large deformation, the ductile metal damage criterion (Ductile Damage) based on stress triaxiality is added to the material properties. The damage criterion is defined as follows:

[0030]

[0031]

[0032] Where: is the damage initiation strain; ε pl is the equivalent plastic strain; η ave is stress triaxiality; σ m is the hydrostatic stress; eis the Mises yield stress; ε1, ε2 and ε3 are three principal strains. The fracture damage initiation strain can be 0.1-0.3, the steel damage evolution path adopts a linear model, and the fracture displacement is 1.5-2.5mm;

[0033] All parts of the model adopt C3D8R solid elements, and the grid is divided by structured meshing technology. The maximum unit size is about 15mm and the minimum unit size is 3mm. Binding constraints (Tie mode) are used to simulate the welding effect between components. The boundary conditions and loading conditions of the model are consistent with the experiment. At the same time, in order to accurately simulate the local buckling phenomenon at the bottom of the steel frame column, an initial defect based on the first-order buckling mode is introduced.

[0034] The step three is specifically as follows:

[0035] The component vulnerability function represents the conditional probability of a component reaching or exceeding a specific damage state (DS) under a given engineering demand parameter (EDP). In this study, it is assumed that the component vulnerability follows a log-normal distribution, and the corresponding probability density function and cumulative distribution function are:

[0036]

[0037] Where: P(DS≥ds i |EDP=x) is the damage level (destruction state) reached or exceeded by the component ds i The probability under the condition of Φ[·] is the standard normal cumulative distribution function; x is the index parameter of the engineering demand parameter EDP; θ i and β i is the median and logarithmic standard deviation of the probability distribution of parameter x;

[0038] Among them, the median value θ i and the logarithmic standard deviation β i It can be calculated as follows:

[0039]

[0040] Where: n is the number of samples of statistical data; x j is the jth engineering requirement parameter value of the component; β r,i and β u,i are the logarithmic standard deviations of the vulnerability function considering intrinsic uncertainty and subjective uncertainty, respectively;

[0041] The establishment of the fragility curve of existing steel frame columns mainly includes the following contents: (1) selecting the EDP index that can characterize the seismic performance of components; (2) classifying the damage and destruction levels of components; (3) establishing the fragility function; (4) eliminating outliers according to the Peirce criterion; (5) determining the median value θ under each damage level. i and the logarithmic standard deviation β i , and then construct the fragility curve; (6) Use the Lilliefors method to test the goodness of fit of the fragility curve. Using the Lilliefors test method, the formula is as follows:

[0042] D=max|P(x j )-S n (x j )| (9)

[0043] S n (x j )=(j-0.5) / n (10)

[0044] D crit =0.895 / (n 0.5 -0.01+0.85n -0.5 ) (11)

[0045] In the formula, P and S n are the fitted vulnerability function value and the sample cumulative distribution function value respectively; D is the maximum absolute difference between the two functions, D crit is the upper limit of D at the 5% significance level, which can be determined according to formula (6). crit , the goodness-of-fit test passes; otherwise, the test fails.

[0046] Furthermore, in step 4,

[0047] The damage status of steel frame columns is divided into five levels: basically intact, slightly damaged, moderately damaged, severely damaged and collapsed;

[0048] The displacement angle is selected as the seismic performance index of steel frame members, and the theoretical characterization of each damage state is given based on the moment-rotation skeleton curve of steel frame column members.

[0049] Furthermore, in step five,

[0050] The finite element analysis method based on the verification of corroded steel frame columns is used to construct a vulnerability function parameter data set, and then the vulnerability curve of existing steel frame columns in the offshore atmospheric environment is established.

[0051] For H-shaped steel members, select the section height-to-width ratio h / b and the flange width-to-thickness ratio b f / t f 、Web height-to-thickness ratio h w / t w , axial compression ratio n, slenderness ratio L / h, and five other parameters, and considering the different service ages T in the offshore atmospheric environment, a finite element analysis model of existing steel frame columns is established.

[0052] In order to obtain the relationship between the weight loss rate η of the steel and the corrosion depth d, it is assumed that: 1) the density of the steel does not change before and after corrosion, that is, ρ0 = ρ T ; 2) Since the length of the component is much larger than the width and height of the cross section, it can be considered that corrosion only occurs in the width and height directions of the cross section of the component, that is, L0 = L T , then the weight loss rate η can be expressed as:

[0053]

[0054] Where: m0, ρ0, A0, L0 are the initial mass, density, cross-sectional area and length of the component respectively; m T , T , A T , L T are the corresponding physical characteristics of the components when they are in service for T years; A corr Indicates the cross-sectional corrosion area. For H-shaped steel members, the cross-sectional specifications are h×b×t w ×t f , then the initial cross-sectional area is A0 = 2bt f +t w (h-2t f ), the cross-sectional corrosion area at the service age of T years is A corr =2bd+d(h-2t f ), d = r corr ·T, and then the weight loss rate η can be converted to:

[0055]

[0056] Combining this formula with the mechanical property degradation formula (1) of corroded steel, the mechanical property index of corroded steel at the service age of T years can be obtained.

[0057] The skeleton curves of the physical test model and numerical analysis model of existing corroded steel frame columns in offshore atmospheric environment are extracted, and the displacement angle data corresponding to each damage state are obtained. The test and simulation data are integrated, and the outliers in the data are eliminated according to the Pierce principle. According to formulas (6)-(8), the fragility curve parameters (logarithmic mean θ and logarithmic standard deviation β) of each damage state of existing steel frame columns under different service ages in offshore atmospheric environment are obtained, and the evolution law of fragility parameters θ and β under each damage state with the increase of service age T is revealed; through regression analysis, the correlation between fragility parameters (logarithmic mean θ and logarithmic standard deviation β) and service age T is established, so as to construct a time-varying fragility model of existing steel frame columns in offshore atmospheric environment.

[0058] Furthermore, in step six,

[0059] Based on the established refined finite element models of existing steel frame columns under different design parameters and service ages and the determined damage levels, the residual seismic capacity of earthquake-damaged steel frame columns was evaluated. First, each model was preloaded to make it pre-damaged to five different damage levels (basically intact, slightly damaged, moderately damaged, severely damaged and collapsed, numbered DS0 to DS4), and then the damaged model was subjected to main loading to examine its residual seismic performance. Referring to the loading system recommended by AISC, loading was performed at displacement angles θ of 0.375%, 0.5%, 0.75%, 1%, 1.5%, 2%, 3%, 4%..., and each displacement level was cycled twice. In the preloading stage, after 5 cycles at the target displacement of the set damage level, unloading was performed in the form of cyclic loading to simulate a real vibration cycle in an earthquake; then, main loading was implemented.

[0060] Extract the residual stiffness K of each damaged steel frame column within the parameter range res (initial stiffness of the hysteresis curve of the main loading section), and compared with the initial stiffness K of the intact steel frame column; the residual bearing capacity M res (i.e. the maximum load on the hysteresis curve of the main loading section), and the maximum bearing capacity of the intact steel frame column M max For comparison; Remaining energy consumption capacity E res (accumulated energy consumption of the main loading section), and the accumulated energy consumption of the intact steel frame column E cum By comparison, the influence of various design parameters and service age on the residual seismic performance of earthquake-damaged steel frame columns is studied.

[0061] Furthermore, in step seven,

[0062] The residual stiffness K of the existing corroded steel frame columns under different design parameters, service age and damage levels is obtained. res / K, residual bearing capacity M res / M max , remaining energy consumption Eres / E cum The BP neural network model based on particle swarm optimization (PSO-BPNN) is used to build a nonlinear mapping relationship between the design parameters, service age, damage degree and residual seismic capacity of existing steel frame columns:

[0063]

[0064] Where: K res / K(DS i )、M res / M max (DS i ) and E res / E cum (DS i ) are the various levels of damage states DS of the existing steel frame columns i The residual stiffness, residual bearing capacity and residual energy dissipation under K res / K(DS i )、M res / M max (DS i ) and E res / E cum (DS i ) corresponding to the nonlinear function; T is the service age, h / b is the aspect ratio of the cross section, b f / t f is the flange width-to-thickness ratio, h w / t w is the web height-to-thickness ratio, n is the axial compression ratio, and L / h is the slenderness ratio.

[0065] In the obtained data set, 70% of the data are randomly selected as the training set, and 50% of the remaining data are selected as the validation set and 50% as the test set, that is, the training set, validation set and test set account for 70%, 15% and 15% of the total samples respectively.

[0066] The service age T, the section aspect ratio h / b, and the flange width-to-thickness ratio b f / t f 、Web height-to-thickness ratio h w / t w , axial compression ratio n, slenderness ratio L / h, damage level DS as input vectors, residual stiffness K res / K, residual bearing capacity M res / M max 、Remaining energy consumption E res / E cum3 performance indicators are used as output vectors. According to the empirical formula and combined with trial and error analysis, the number of hidden layer neuron nodes l=8 is determined. Finally, the PSO-BPNN model parameters are as follows: 1) the input layer contains 7 neurons, the output layer contains 1 neuron, and the hidden layer contains 8 neurons; 2) the maximum number of iterations is 1000 steps; the learning rate is set to 0.01; the error tolerance is 0.001; and the momentum constant is set to 0.5. Thus, the optimal evaluation model of the residual seismic capacity of earthquake-damaged existing multi-age steel frame columns is established.

[0067] Further, in step eight,

[0068] Based on the proposed time-varying fragility model of existing multi-age steel frame columns, the probability of occurrence of each level of damage under a given displacement angle θ can be obtained: matrix [DS i |θ](T), the damage probability matrix:

[0069]

[0070] Where: P[DS i |θ](T) is the existing steel frame column reaching or exceeding a certain damage state DS i The probability of; T is the service age.

[0071] The residual stiffness K obtained based on PSO-BPNN res / K, residual bearing capacity M res / M max and the remaining energy consumption E res / E cum Combined with the vulnerability failure probability matrix, the expected value of the residual seismic capacity ratio of existing multi-age steel frame columns under a given displacement angle θ in the offshore atmospheric environment can be obtained in a probabilistic sense, and this expected value can be used to predict the corresponding seismic capacity safety reserve.

[0072]

[0073]

[0074] Where: E(K res )、E(M res )、E(E res ) are the expected values ​​of residual stiffness, residual bearing capacity and residual energy dissipation respectively; P matrix [DS i |θ](T) is the probability of occurrence of each level of damage state; K res / K(DS i )、M res / M max (DS i ) and E res / Ecum (DS i ) are the destruction states DS at each level i The residual stiffness, residual bearing capacity and residual energy dissipation under the load condition; T is the service age.

[0075] Beneficial effects of the present invention:

[0076] (1) The seismic damage mechanism and seismic performance degradation law of existing multi-age steel frame columns under different design parameters and service ages are revealed.

[0077] (2) The established three-dimensional refined finite element model of steel frame columns taking into account corrosion damage and ductile metal damage criteria can truly reflect the stress and deformation performance of corroded steel frame columns under low-cycle reciprocating loads.

[0078] (3) Considering the influence of environmental factors and service age on the performance indicators of existing steel components, the classification criteria and theoretical characterization of the damage status of existing steel frame columns are proposed, and the classification standards for the damage level of existing steel structures in the existing specifications are improved.

[0079] (4) Considering the uncertainties of design parameters and durability damage, a time-varying vulnerability model of existing multi-age steel frame columns is established to achieve continuous assessment and prediction of the seismic vulnerability of steel frame columns throughout their life cycle.

[0080] (5) Based on the PSO-BPNN neural network model, a nonlinear mapping relationship between the design parameters, service age, degree of seismic damage and residual seismic capacity of steel frame columns is established, and an optimal assessment model for the residual seismic capacity of existing multi-age steel frame columns is established to overcome the problems of high assessment cost, low efficiency and difficulty in quantifying the actual seismic capacity of the structure in traditional seismic capacity assessment methods.

[0081] (6) Combining the results of vulnerability and residual seismic capacity analysis, the residual seismic capacity of existing multi-age steel frame columns under specific displacement angles in the offshore atmospheric environment can be directly given in a probabilistic sense, thus achieving a scientific and efficient evaluation of their residual seismic capacity.

[0082] (7) It eliminates the need for professional personnel to go to the earthquake site for investigation, thus solving the shortcomings of traditional methods such as low efficiency, poor accuracy of quantitative characterization, and reliance on expert experience. It improves the intelligence and reliability of "revealing the evolution law of seismic performance of existing steel structures before an earthquake - assessing the residual seismic capacity of damaged steel structures after an earthquake", providing theoretical support for the effective implementation of seismic performance assessment, earthquake damage prediction, and post-earthquake safety appraisal of existing steel structures in earthquake-prone areas, and providing a technical basis for engineering and technical personnel engaged in earthquake damage early warning, assessment, and reinforcement of existing engineering structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1It is a machine learning-based assessment process for the time-varying vulnerability and residual seismic capacity of existing steel frame columns.

[0084] Figure 2 It is a climate environment simulation system.

[0085] Figure 3 It is a low-cycle reciprocating loading test device.

[0086] Figure 4 It is a low-cycle reciprocating loading system.

[0087] Figure 5 This is the analysis of the low-cycle reciprocating loading test results; Figure 5 (a) is the test failure mode, Figure 5 (b) is the hysteresis curve. Figure 5 (c) is the skeleton curve, Figure 5 (d) is the stiffness degradation comparison curve, Figure 5 (e) is the energy consumption comparison curve.

[0088] Figure 6 It is the corrosion form of steel surface.

[0089] Figure 7 is the finite element model of the corroded steel frame column; Figure 7 (a) is the finite element model of a typical corroded steel frame column. Figure 7 (b) is the first-order buckling mode.

[0090] Figure 8 Comparison of finite element analysis and test results Figure 8 (a) is the comparison of destruction morphology. Figure 8 (b) is the comparison of hysteresis curves.

[0091] Fig. 9 Establish a process for the fragility of existing steel frame columns.

[0092] Fig.10 Quantitative standard for damage and failure state of existing steel frame columns.

[0093] Fig.11 Comparison of the cross sections of H-shaped steel before and after corrosion.

[0094] Fig.12 is the fragility curve of existing steel frame columns at different service ages in offshore atmospheric environment; Fig.12 (a) is the fragility curve under DS1 damage state, Fig.12 (b) is the fragility curve under DS2 damage state, Fig.12 (c) is the fragility curve under DS3 damage state, Fig.12 (d) is the fragility curve under DS4 damage state.

[0095] Fig.13 Loading system for residual seismic capacity analysis of components.

[0096] Fig.14 is the remaining capacity K res / K,M res / M max and E res / E changes with the service age T; Fig.14 (a) is K res / K changes with the service age T, Fig.14 (b) is M res / M max The relationship between the service age T and the service age T is: Fig.14 (c) is E res / The relationship between E and service age T.

[0097] Fig.15 This is the operation flow of the PSO-BPNN neural network model.

[0098] Fig.16 Comparison between the prediction results of the PSO-BPNN neural network model and the measured results; Fig.16 (a) is K res / K prediction and measured results comparison, Fig.16 (b) is M res / M max Comparison of predicted and measured results. Fig.16 (c) is E res / EComparison of predicted and measured results.

[0099] Fig.17 Schematic diagram of the relevant dimensions of the specimen. DETAILED DESCRIPTION

[0100] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0101] The overall flow chart of the process of the present invention is as follows Figure 1 As shown:

[0102] Assessment method of time-varying vulnerability and post-earthquake residual seismic capacity of existing corroded steel frame columns and steel frame nodes based on machine learning:

[0103] The method comprises the following steps:

[0104] Step 1: Conduct low-cycle reciprocating loading tests on corroded steel frame columns in offshore atmospheric environment to reveal the seismic damage mechanism and seismic performance degradation law of existing corroded steel frame columns in offshore atmospheric environment.

[0105] Step 2: Based on the physical test model, a three-dimensional refined finite element model of the steel frame column considering corrosion damage and ductile metal damage criteria is established, and the accuracy and effectiveness of the constructed numerical model are verified by comparing the test results.

[0106] Step 3: Determine the component vulnerability function and sort out the process of constructing the time-varying vulnerability model of existing corroded steel frame columns.

[0107] Step 4: Select the engineering requirement parameter EDP index that can characterize the seismic performance of existing corroded steel frame columns; classify the damage levels of existing corroded steel frame columns and determine the corresponding quantitative limits.

[0108] Step 5: Combine the physical test data and numerical simulation analysis data to construct a parameter data set of the time-varying vulnerability function of existing corroded steel frame columns, and establish the time-varying vulnerability curve of existing multi-age steel frame columns in the offshore atmospheric environment.

[0109] Step 6: Based on the established numerical model of existing corroded steel frame columns and the determined damage level, reveal the evolution law of the residual seismic capacity of existing corroded steel frame columns under different damage levels, different service ages, and different design parameters;

[0110] Step 7: Based on the BP neural network model optimized by particle swarm optimization, a nonlinear mapping relationship is established between the design parameters, service age, damage degree and residual stiffness, residual bearing capacity and residual energy consumption of existing corroded steel frame columns, and an optimal assessment model for the residual seismic capacity of earthquake-damaged existing multi-age steel frame columns is established.

[0111] Step 8: Combined with the results of vulnerability analysis, the residual seismic capacity of existing multi-age steel frame columns under specific horizontal displacements in the offshore atmospheric environment is given in a probabilistic sense to achieve a rapid and reasonable assessment of their residual seismic capacity.

[0112] In step one,

[0113] With reference to current specifications and engineering examples, 12 steel frame column specimens were designed. All specimens were made of hot-rolled H-shaped steel with a cross-sectional specification of HW 250×250×9×14mm and a material of Q235B. For detailed dimensions, see Fig.17 (Unit: mm).

[0114] The ZHT / W2300 climate environment simulation system was used to conduct accelerated corrosion tests on all specimens (steel standard specimens, steel frame column specimens, and steel frame node specimens) in the offshore atmospheric environment. Figure 2 At the same time, the system is equipped with advanced intelligent digital control facilities to achieve full monitoring of the test. According to the neutral salt spray test conditions specified in "Artificial Atmosphere Corrosion Test-Salt Spray Test" (GB / T10125-2012), the system environmental parameter settings are shown in Table 1.

[0115] Table 1 Environmental parameters

[0116]

[0117] Low cycle reciprocating test loading device such as Figure 3 As shown. The horizontal reciprocating load at the column end is provided by a 50-ton MTS electro-hydraulic servo actuator; the vertical load at the column top is applied by a 100-ton synchronous hydraulic jack. The specimen is fixed to the rigid floor of the test room by a compression beam and anchor bolts. In addition, lateral supports are added on both sides of the specimen to prevent the specimen from buckling out of the plane. The horizontal load adopts a displacement-controlled loading method, referring to the loading system recommended by AISC, and the displacement angles θ are 0.375%, 0.5%, 0.75%, 1%, 1.5%, 2%, 3%, 4%..., respectively. The loading system is shown in Figure 4 .

[0118] Typical test results are as follows Figure 5 Through the analysis of hysteresis curve, skeleton curve, stiffness degradation, ductility and energy dissipation, the influence of corrosion degree on the seismic performance of existing steel frame columns is revealed: with the increase of corrosion degree, the bearing capacity, deformation capacity and energy dissipation capacity of existing steel frame columns gradually decrease.

[0119] At the same time, according to "Steel and Steel Products Mechanical Properties Test Sampling Location and Test Preparation" (GB / T 2975-2018), three kinds of standard steel specimens with thickness of 6.5mm, 9mm and 14mm were cut from the same batch of plates as the steel frame components for accelerated corrosion and tensile failure tests. The accelerated corrosion time was used as the research parameter, and the weight loss rate η was used for quantification. The least squares method was used for regression analysis of the test results to obtain the functional relationship between the mechanical properties of corroded steel and its degree of corrosion (weight loss rate) in the offshore atmospheric environment.

[0120]

[0121] Where: f y0 、f u0 , δ0, and E0 are the yield strength, ultimate strength, elongation and elastic modulus of the steel before corrosion, respectively; and f y 、f u , δ, and E are the corresponding mechanical properties of steel after corrosion; η is the weight loss rate (less than 14%).

[0122] In step 2,

[0123] Based on ABAQUS general software, a three-dimensional finite element model considering corrosion damage and ductile metal damage criteria was established to perform numerical simulation on the test steel frame column. The corrosion forms of steel are mainly divided into uniform corrosion and pitting corrosion, such as Figure 6For this reason, the corrosion damage is divided into uniform corrosion and pitting corrosion to comprehensively consider the different mechanisms of their influence on steel performance:

[0124] (1) Uniform corrosion is reflected in the reduction of plate thickness, that is, the reduction of component cross-sectional dimensions;

[0125] (2) Pitting is reflected in the reduction of mechanical properties of materials, that is, the degradation law of steel yield strength, ultimate strength, elastic modulus, and elongation as the weight loss rate increases is introduced. Among them, the maximum residual thickness is used to determine the mechanical property index.

[0126] In addition, in order to accurately reflect the bearing capacity and stiffness degradation characteristics of steel components and structures caused by the accumulation of plastic strain under large deformation, the ductile metal damage criterion (Ductile Damage) based on stress triaxiality is added to the material properties, where the damage initiation criterion can be defined as:

[0127]

[0128] In the formula, is the damage initiation strain; η ave is the isotropic stress triaxiality; ε pl is the equivalent plastic strain; σ m and σ e are hydrostatic stress and Mises equivalent stress respectively; ε1, ε2 and ε3 are three principal strains. Based on existing literature and repeated tests, the fracture damage initiation strain can be taken as 0.1-0.3. The steel damage evolution path adopts a linear model, and the fracture displacement is 1.5-2.5mm.

[0129] All parts of the model use C3D8R solid elements, and the minimum element size is 3mm. The boundary conditions and loading conditions of the model are consistent with the test. At the same time, in order to accurately simulate the local buckling phenomenon at the bottom of the steel frame column, the initial defect based on the first-order buckling mode is introduced. Figure 7 Finite element model of corroded steel frame columns established.

[0130] In terms of test failure morphology, hysteresis curve, etc., the finite element analysis results are in good agreement with the test results. Figure 8 As shown, the accuracy of the constructed finite element model of existing corroded steel frame columns is verified, indicating that it can better characterize the mechanical performance of existing corroded steel frame columns under low-cycle reciprocating action.

[0131] In step three,

[0132] The component vulnerability function represents the conditional probability of a component reaching or exceeding a specific damage state (DS) under a given engineering demand parameter (EDP). In the study, it is generally assumed that the component vulnerability follows a log-normal distribution, and the corresponding probability density function and cumulative distribution function are:

[0133]

[0134] Where: P(DS≥ds i |EDP=x) is the damage level (destruction state) reached or exceeded by the component ds i The probability under the condition of Φ[·] is the standard normal cumulative distribution function; x is the index parameter of the engineering demand parameter EDP; θ i and β i is the median and logarithmic standard deviation of the probability distribution of parameter x.

[0135] Among them, the median value θ i and the logarithmic standard deviation β i It can be calculated as follows:

[0136]

[0137] Where: n is the number of samples of statistical data; x j is the jth engineering requirement parameter value of the component; β r,i and β u,i are the logarithmic standard deviations of the vulnerability function considering intrinsic uncertainty and subjective uncertainty, respectively.

[0138] The establishment of the fragility curve of existing steel frame columns mainly includes the following contents: (1) selecting the EDP index that can characterize the seismic performance of components; (2) classifying the damage and destruction levels of components; (3) establishing the fragility function; (4) eliminating outliers according to the Peirce criterion; (5) determining the median value θ under each damage level. i and the logarithmic standard deviation β i , and then construct the fragility curve; (6) Use the Lilliefors method to test the goodness of fit of the fragility curve. Using the Lilliefors test method, the formula is as follows:

[0139] D=max|P(x j )-S n (x j )| (9)

[0140] S n (x j )=(j-0.5) / n (10)

[0141] Dcrit =0.895 / (n 0.5 -0.01+0.85n -0.5 ) (11)

[0142] In the formula, P and S n are the fitted vulnerability function value and the sample cumulative distribution function value respectively; D is the maximum absolute difference between the two functions. crit is the upper limit of D at the 5% significance level, which can be determined according to formula (6). crit , the goodness of fit test is passed; otherwise, the test fails. The process of establishing the fragility curve of existing steel frame columns is as follows: Fig. 9 shown.

[0143] In step four,

[0144] At present, there are many descriptions of the damage states of steel structures and components in performance-based seismic design and damage assessment research. However, the impact of environmental corrosion on the performance indicators of existing steel columns has not been considered. Combining the existing research and the damage characteristics of steel frame columns in this experiment, the damage states of steel frame columns are divided into five levels: basically intact, slightly damaged, moderately damaged, severely damaged and collapsed, numbered DS1 to DS5 respectively. The description of the damage phenomenon under each damage state and the photos of typical damage characteristics are detailed in Table 2, and the corresponding repair methods are also given.

[0145] Existing studies have shown that displacement angle, as an indicator of the seismic performance of steel frame components, can better characterize the relationship between the degree of component damage and engineering demand parameters; and based on the moment-rotation skeleton curve of steel frame column components and connection units, a theoretical characterization of each damage state is given. Based on this, the present invention defines the equivalent yield point B of the skeleton curve as reaching a slight damage state DS2; point D at 2 / 3 of the BC segment on the skeleton curve is defined as reaching a moderate damage state DS3; the peak point C on the skeleton curve is defined as reaching a severe damage state DS4; and point E on the skeleton curve where the bending bearing capacity drops by 25% is defined as reaching a collapse state DS5. Fig.10 shown.

[0146] Table 2 Definition of damage and failure states of existing steel frame columns

[0147]

[0148]

[0149] In step five,

[0150] Since the test data are limited and the component fragility curve established based on non-homogeneous test data will result in distorted evaluation results, a vulnerability function parameter data set is constructed based on the verified finite element analysis method of corroded steel frame columns, and then the vulnerability curve of existing steel frame columns in the offshore atmospheric environment is established.

[0151] For H-shaped steel members, the width-to-thickness ratio of the plate and its assembly, axial compression ratio and corrosion damage are the main factors affecting its seismic performance. f / t f 、Web height-to-thickness ratio h w / t w , axial compression ratio n, slenderness ratio L / h, and five other parameters. Meanwhile, the different service ages T in the offshore atmospheric environment were taken into consideration. A finite element analysis model of 800 existing steel frame columns was established, covering most of the situations that may occur in actual projects. The design parameters are shown in Table 3.

[0152] Table 3 Design parameters and values

[0153] Design parameters Value h / b 1、1.3、1.5、1.8、2 <![CDATA[b f / t f ]]> 8.61、11、13、15、17、19 <![CDATA[h w / t w ]]> 10、15、24.67、35、50、65、80 n 0、0.1、0.15、0.2、0.25、0.3、0.4、0.5 L / h 3.5、4.8、6、7.2、8、10 T(years) 0、30、40、50、60、70、80

[0154] The offshore atmospheric environment refers to the coastal atmospheric area affected by the marine chloride ion environment. The rusting time of steel structures is mainly determined by the failure time of their anti-corrosion coatings. However, the factors and mechanisms that affect the failure of anti-corrosion coatings are relatively complex, and it is difficult to accurately assess the initial rusting time of steel materials through coating failure. Therefore, based on the research results of the research team, relevant specifications and regulations for steel structure anti-corrosion, and existing relevant research results, the initial rusting time of in-service steel structure materials in the offshore atmospheric environment is defined as 30 years; referring to the C3 corrosion environment defined in ISO 12944-2 (low-salinity coastal areas, steel corrosion rate K is 25-50μm / a) and related data defined by the International Organization for Standardization, the average corrosion rate of steel is defined as 0.04mm / a.

[0155] In order to obtain the relationship between the weight loss rate η of the steel and the corrosion depth d, it is assumed that: 1) the density of the steel does not change before and after corrosion, that is, ρ0 = ρ T ; 2) Since the length of the component is much larger than the width and height of the cross section, it can be considered that corrosion only occurs in the width and height directions of the cross section of the component, that is, L0 = L T . Then the weight loss rate η can be expressed as:

[0156]

[0157] Where: m0, ρ0, A0, L0 are the initial mass, density, cross-sectional area and length of the component respectively; m T , T , A T , L Tare the corresponding physical characteristics of the components when they are in service for T years; A corr Indicates the cross-sectional corrosion area. For H-shaped steel members (see Fig.11 ), the cross-sectional specifications are h×b×t w ×t f , then the initial cross-sectional area is A0 = 2bt f +t w (h-2t f ), the cross-sectional corrosion area at the service age of T years is A corr =2bd+d(h-2t f ), d = r corr ·T, and then the weight loss rate η can be converted to:

[0158]

[0159] Combining this formula with the mechanical property degradation formula (1) of corroded steel, the mechanical property index of corroded steel at the service age of T years can be obtained.

[0160] Existing experimental studies have also shown that under the influence of environmental erosion, the mechanical and seismic performance of steel structural components and structures degrade significantly with the increase of service age. Therefore, when statistically analyzing the displacement angles of various steel frame column models under different damage limit states, the influence of service age on steel columns is considered, while the influence of factors such as section aspect ratio, plate width-to-thickness ratio, axial compression ratio and slenderness ratio are ignored.

[0161] The skeleton curves of the physical test model and numerical analysis model of 800 existing corroded steel frame columns in offshore atmospheric environment were extracted, and the displacement angle data corresponding to each damage state were obtained. The test and simulation data were integrated, and the outliers in the data were eliminated according to the Pierce principle. According to formulas (6)-(8), the fragility curve parameters (logarithmic mean and logarithmic standard deviation) of each damage state of existing steel frame columns under different service ages in offshore atmospheric environment were obtained, as shown in Table 4.

[0162] Table 4 Parameters of vulnerability function of existing steel frame columns at different service ages

[0163]

[0164] Substituting the vulnerability parameters in Table 4 into the vulnerability function formula (5) can obtain the vulnerability curves of existing steel frame columns under different service ages in the offshore atmospheric environment, as shown in Fig.12 It can be seen that with the increase of service age, the exceedance probability of each damage state of the steel frame column at the same displacement angle continues to increase.

[0165] In order to continuously evaluate and predict the seismic vulnerability of steel frame columns throughout their life cycle, a time-varying vulnerability model for existing steel frame columns is proposed. Based on the vulnerability function parameter data in Table 4, the correlation between the vulnerability parameters and the service age can be established through regression analysis, thereby constructing a time-varying vulnerability model for existing steel frame columns in offshore atmospheric environment, as shown in Equation (14) Table 5.

[0166]

[0167] Table 5 Regression coefficients of vulnerability function parameters for different damage states

[0168]

[0169] In step six,

[0170] Based on the established numerical models of 800 existing steel frame columns and the determined damage levels (Table 4), the residual seismic capacity of earthquake-damaged steel frame columns was evaluated. First, each model was preloaded to make it pre-damaged to five different damage levels (DS0~DS4), and then the damaged model was subjected to main loading to study its residual seismic performance. The loading system is as follows Fig.13 As shown in the figure, referring to the loading system recommended by AISC, loading is performed at displacement angles θ of 0.375%, 0.5%, 0.75%, 1%, 1.5%, 2%, 3%, 4%..., and each displacement level is cycled twice. In the preloading stage, after 5 cycles at the target displacement of the set damage level, unloading is performed in the form of cyclic loading, and then the main loading is implemented.

[0171] Extract the residual stiffness K of each damaged steel frame column within the parameter range res (initial stiffness of the hysteresis curve of the main loading section), and compared with the initial stiffness K of the intact steel frame column; the residual bearing capacity M res (i.e. the maximum load on the hysteresis curve of the main loading section), and the maximum bearing capacity of the intact steel frame column M max For comparison; Remaining energy consumption capacity E res (accumulated energy consumption of the main loading section), and the accumulated energy consumption of the intact steel frame column E cum By comparison, the influence of various design parameters and service age on the residual seismic performance of earthquake-damaged steel frame columns is studied. Fig.14 Given the remaining capacity K res / K,M res / M max and E res / E cum The relationship with the service age T.

[0172] In step seven,

[0173] The BP (Back propagation) neural network model is a multi-layer feedforward neural network trained according to the error back propagation algorithm. It has a simple grid structure and good generalization ability. It can avoid redundant calculations and approximate any nonlinear function under arbitrary precision requirements, accurately reflecting and storing a large number of input-output mode mapping relationships. The BP neural network model consists of three layers: input layer, hidden layer, and output layer. Each neuron receives input signals from the upper layer, calculates the weighted sum of its inputs, and transmits outputs to the next layer through the activation function:

[0174] Hidden Layer:

[0175] Output layer:

[0176] Activation function (such as Sigmoid function): In the formula, x i is the input vector; w ij and b j is the weight parameter corresponding to the input vector and is the bias vector; j is the hidden layer output vector; w kj and b k is the weight parameter and bias vector corresponding to the output vector; k is the output vector of the output layer; f() is the activation function.

[0177] It should be noted that when the BP neural network model has inappropriate initial weights and bias values ​​during the training process, it will cause the model to fall into the local optimal problem and the fitting problem will occur, that is, it performs well in the training phase, but the prediction accuracy is poor in the test phase. Studies have shown that integrating multiple intelligent learning models can improve the prediction accuracy and robustness of machine learning. Therefore, considering the accuracy and efficiency of machine learning, the BP neural network model optimized by particle swarm optimization (PSO-BPNN) is used to evaluate the residual capacity of corroded steel frame columns.

[0178] In the particle swarm algorithm, each particle adjusts the initial weight and deviation of the optimization solution according to the fitness function. and initial velocity As the input weight vector and bias vector in the BPNN model. Then the inverse of the root mean square error (RMSE) between the prediction vector and the target vector is used as the fitness function to judge the quality of the solution. For each iteration, the particle selects the local best position P in the history i and the global optimal position P g The particle's velocity v i and position x i The update equation is as follows:

[0179]

[0180] In the formula, t is the number of iterations; ω is the correlation coefficient; c1 and c2 are acceleration constants; ε1 and ε2 are combined random values ​​between [0,1]. The particle velocity and position updates are repeated many times until the optimization goal is achieved. The operation process of the BP neural network model based on particle swarm optimization (PSO-BPNN) is as follows: Fig.15 As shown in the figure. 70% of the data in the established database are randomly selected as the training set, 50% of the remaining data are selected as the validation set, and the rest of the data are selected as the test set, that is, the proportion of the training set, validation set and test set in the total samples is 70%, 15% and 15% respectively. The component design parameters (h / b, b f / t f ,h w / t w ,n and L / h), service age T, damage degree DS as input vectors, and residual performance index K res / K,M res / M max or E res / E cum As the output vector. According to formula (20), the number of hidden layer neuron nodes l=8 is determined by trial and error analysis. Finally, the PSO-BPNN model parameters are as follows: 1) the input layer contains 7 neurons, the output layer contains 1 neuron, and the hidden layer contains 8 neurons; 2) the maximum number of iterations is 1000 steps; the learning rate is set to 0.01; the error tolerance is 0.001; and the momentum constant is set to 0.5.

[0181]

[0182] In the formula, l is the number of hidden layer neuron nodes; m and n are the number of input layer and output layer neuron nodes respectively; a is a constant between 1 and 10. The nonlinear mapping relationship of the PSO-BPNN model is as follows:

[0183]

[0184] In the formula, K res / K(DS i )、M res / M max (DS i ) and E res / E cum (DS i ) are the various levels of damage states DS of the existing steel frame columns i The residual stiffness, residual bearing capacity and residual energy dissipation under K are respectively res / K(DS i )、M res / Mmax (DS i ) and E res / E cum (DS i ) corresponding to the nonlinear function; T is the service age, h / b is the aspect ratio of the cross section, b f / t f is the flange width-to-thickness ratio, h w / t w is the web height-to-thickness ratio, n is the axial compression ratio, and L / h is the slenderness ratio.

[0185] Fig.16 The comparison between the prediction results of the PSO-BPNN neural network model and the measured results shows that the residual stiffness K res / K, remaining load capacity M res / M max 、Remaining energy consumption E res / E cum The correlation coefficients between the predicted values ​​and the true values ​​are both above 0.98, and basically fall within the 95% prediction interval, indicating that the constructed PSO-BPNN neural network model has good prediction ability.

[0186] In step eight,

[0187] Based on the proposed time-varying fragility model of existing steel frame columns (Eq. 14), the probability of occurrence of each level of damage (failure probability matrix) P at a given displacement angle θ can be obtained: matrix [DS i |θ](T), that is:

[0188]

[0189] Where: P[DS i |θ](T) is the existing steel frame column reaching or exceeding a certain damage state DS i The probability of; T is the service age.

[0190] Then the residual stiffness K predicted by the PSO-BPNN neural network model is res / K, remaining load capacity M res / M max and the remaining energy consumption E res / E cum Combined with formula (21), the expected value of the residual seismic capacity ratio of the existing steel frame column at a given displacement angle θ can be obtained in a probabilistic sense (formula (22) to (24)). This expected value can be used to predict the corresponding seismic capacity safety reserve. For example, a steel frame column with a service age of 50 years (design parameters n = 0.3, b f / t f =8.61,h w / tw =24.67, h / b=1, L / h=6), under the elastic-plastic inter-story displacement angle limit of 0.02 specified in the seismic code, the residual stiffness K res / K is 0.72, and the remaining load capacity M res / M max is 0.83, and the remaining energy consumption E res / E cum of 0.70.

[0191]

[0192] Where: E(K res )、E(M res )、E(E res ) are the expected values ​​of residual stiffness, residual bearing capacity and residual energy dissipation respectively; P matrix [DS i |θ](T) is the probability of occurrence of each level of damage state; K res / K(DS i )、M res / M max (DS i ) and E res / E cum (DS i ) are the destruction states DS at each level i The residual stiffness, residual bearing capacity and residual energy dissipation under the load condition; T is the service age.

Claims

1. A method for evaluating the time-varying vulnerability and residual seismic capacity of existing steel columns based on machine learning, characterized in that: The steps include: Step 1: Construct a physical test model and carry out low-cycle reciprocating loading tests on corroded steel frame columns in offshore atmospheric environment to reveal the seismic damage mechanism and seismic performance degradation law of existing corroded steel frame columns in offshore atmospheric environment; Step 2: Based on the physical test model in step 1, a three-dimensional refined finite element model of the steel frame column considering corrosion damage and ductile metal damage criteria is established, and the accuracy and effectiveness of the established numerical model are verified by comparing the test results; Step 3: Determine the component vulnerability function and sort out the process of constructing the time-varying vulnerability model of the existing corroded steel frame columns; Step 4: Select the engineering requirement parameter EDP index that can characterize the seismic performance of existing corroded steel frame columns; classify the damage level of existing corroded steel frame columns and determine the corresponding quantitative limit values; Step 5: Combine the physical test data and numerical simulation analysis data to construct a parameter data set of time-varying vulnerability function of existing corroded steel frame columns, and establish a time-varying vulnerability curve of existing multi-age steel frame columns under offshore atmospheric environment; Step 6: Based on the three-dimensional refined finite element model of the steel frame column established in step 2 and the damage level determined in step 4, the evolution law of the residual seismic capacity of the existing corroded steel frame columns under different damage levels, different service ages, and different design parameters is revealed, and the residual seismic capacity analysis results are obtained; Step 7: Based on the residual seismic capacity analysis results of step 6, a BP neural network model optimized by particle swarm optimization is used to build a nonlinear mapping relationship between the design parameters, service age, damage degree and residual stiffness, residual bearing capacity and residual energy dissipation of existing corroded steel frame columns, and establish an optimal evaluation model for the residual seismic capacity of earthquake-damaged existing multi-age steel frame columns; Step 8: Combining the vulnerability curve in step 5 with the optimal assessment model of residual seismic capacity in step 7, the residual seismic capacity of existing multi-age steel frame columns under specific horizontal displacement in offshore atmospheric environment is given in a probabilistic sense, so as to achieve a rapid and reasonable assessment of its residual seismic capacity; In step seven, The residual stiffness ratio K of the existing corroded steel frame columns under different design parameters, service age and damage level is obtained. res / K, residual bearing capacity ratio M res / M max , remaining energy consumption ratio E res / E cum The BP neural network model based on particle swarm optimization is used to build a nonlinear mapping relationship between the design parameters, service age, damage degree and residual seismic capacity of existing steel frame columns: Where: K res / K(DS i ), M res / M max (DS i ) and E res / E cum (DS i ) are the various levels of damage states DS of the existing steel frame columns i The residual stiffness, residual bearing capacity and residual energy dissipation under K res / K(DS i ), M res / M max (DS i ) and E res / E cum (DS i ) corresponding to the nonlinear function; T is the service age, h / b is the aspect ratio of the cross section, b f / t f is the flange width-to-thickness ratio, h w / t w is the height-to-thickness ratio of the web, n is the axial compression ratio, and L / h is the slenderness ratio; In the obtained data set, a part of the data is randomly selected as the training set, and a part of the remaining data is selected as the validation set, and the other part is selected as the test set; The service age T, the section aspect ratio h / b, and the flange width-to-thickness ratio b f / t f 、Web height-to-thickness ratio h w / t w , axial compression ratio n, slenderness ratio L / h, damage level DS as input vectors, residual stiffness K res / K, residual bearing capacity M res / M max 、Remaining energy consumption E res / E cum Three performance indicators are used as output vectors; thus, an optimal assessment model for the residual seismic capacity of existing multi-age steel frame columns damaged by earthquakes is established.

2. The method for evaluating time-varying vulnerability and residual seismic capacity of existing steel columns based on machine learning according to claim 1 is characterized in that: In step 1, steel frame column specimens are designed with reference to existing specifications and engineering examples. The specimens are all made of hot-rolled H-shaped steel. According to the "Artificial Atmosphere Corrosion Test-Salt Spray Test", the ZHT / W2300 climate environment simulation system is used to conduct salt spray accelerated corrosion on all specimens. Then, low-cycle reciprocating loading tests are conducted on the corroded specimens. Through the tensile performance test of corroded steel, the least square method was used to perform regression analysis on the test results, and the functional relationship between the mechanical performance index and the weight loss rate of corroded steel in the offshore atmospheric environment was obtained. Where: f y0 、f u0 , δ0 and E0 are the yield strength, ultimate strength, elongation and elastic modulus of the steel before corrosion, respectively; and f y 、f u , δ and E are the corresponding mechanical properties of steel after corrosion; η is the weight loss rate, η is less than 14%; Through the experimental failure morphology, hysteresis curve, skeleton curve, stiffness degradation, ductility and energy dissipation analysis of corroded steel frame columns, the influence of corrosion degree on the seismic performance of existing steel frame columns is revealed.

3. The method for evaluating time-varying vulnerability and residual seismic capacity of existing steel columns based on machine learning according to claim 1, characterized in that: In step 2, a three-dimensional refined finite element model considering corrosion damage and ductile metal damage criteria was established, and a numerical simulation of the test steel frame column was performed. The corrosion damage was divided into two parts: uniform corrosion and pitting corrosion: (1) Uniform corrosion is reflected in the reduction of plate thickness, that is, the reduction of component cross-sectional dimensions; (2) Pitting is manifested in the reduction of mechanical properties of materials, that is, the functional relationship between the yield strength, ultimate strength, elastic modulus, and elongation of steel and the weight loss rate is introduced; Where: f y0 、f u0 , δ0 and E0 are the yield strength, ultimate strength, elongation and elastic modulus of the steel before corrosion, respectively; and f y 、f u , δ and E are the corresponding mechanical properties of steel after corrosion; η is the weight loss rate; Among them, the maximum residual thickness is used to determine the mechanical performance index; Add a stress triaxiality-based ductile metal damage criterion in the material properties. The damage criterion is defined as follows: Where: is the damage initiation strain; ε pl is the equivalent plastic strain; η ave is stress triaxiality; σ m is the hydrostatic stress; e is the Mises yield stress; ε1, ε2 and ε3 are the three principal strains; All components of the model adopt C3D8R solid elements, and the grid is divided by structured meshing technology; binding constraints are used to simulate the welding effect between components; the boundary conditions and loading conditions of the model are consistent with the experiment; at the same time, in order to accurately simulate the local buckling phenomenon at the bottom of the steel frame column, an initial defect based on the first-order buckling mode is introduced.

4. The method for evaluating time-varying vulnerability and residual seismic capacity of existing steel columns based on machine learning according to claim 1, characterized in that: The step three is specifically as follows: The component vulnerability function represents the conditional probability of a component reaching or exceeding a specific damage state under given engineering demand parameters. In this study, it is assumed that the component vulnerability follows a log-normal distribution, and the corresponding probability density function and cumulative distribution function are: Where: P(DS≥ds i |EDP=x) is the damage level ds reached or exceeded by the component i The probability under the condition of Φ[·] is the standard normal cumulative distribution function; x is the index parameter of the engineering demand parameter EDP; θ i and β i is the median and logarithmic standard deviation of the probability distribution of parameter x; Among them, the median value θ i and the logarithmic standard deviation β i It can be calculated as follows: Where: N is the number of samples of statistical data; x j is the jth engineering requirement parameter value of the component; β r,i and β u,i are the logarithmic standard deviations of the vulnerability function considering intrinsic uncertainty and subjective uncertainty, respectively; The establishment of the fragility curve of existing steel frame columns includes the following contents: (1) Select the EDP index that can characterize the seismic performance of the component; (2) Classification of component damage and failure levels; (3) Establish vulnerability function; (4) Eliminate outliers based on Peirce's criterion; (5) Determine the median value θ at each damage level i and the logarithmic standard deviation β i , and then construct the vulnerability curve; (6) The Lilliefors method is used to test the goodness of fit of the fragility curve. The Lilliefors test method is used, and the formula is as follows: D=maxP(x j )-S n (x j )| (9) S n (x j )=(j-0.5) / n (10) D crit =0.895 / (n 0.5 -0.01+0.85n -0.5 ) (11) In the formula, P and S n are the fitted vulnerability function value and the sample cumulative distribution function value respectively; D is the maximum absolute difference between the two functions, D crit is the upper limit of D at the 5% significance level, determined according to formula (6), when D ≤ D crit , the goodness-of-fit test passes; otherwise, the test fails.

5. The method for evaluating time-varying vulnerability and residual seismic capacity of existing steel columns based on machine learning according to claim 1, characterized in that: In step four, The damage status of steel frame columns is divided into five levels: basically intact, slightly damaged, moderately damaged, severely damaged and collapsed; The displacement angle is selected as the seismic performance index of steel frame members, and the theoretical characterization of each damage state is given based on the moment-rotation skeleton curve of steel frame column members.

6. The method for evaluating time-varying vulnerability and residual seismic capacity of existing steel columns based on machine learning according to claim 1, characterized in that: In step five, The finite element analysis method of corroded steel frame columns based on the verification is used to construct a vulnerability function parameter data set, and the vulnerability curve of existing steel frame columns in offshore atmospheric environment is established; For H-shaped steel members, select the section height-to-width ratio h / b and the flange width-to-thickness ratio b f / t f 、Web height-to-thickness ratio h w / t w , axial compression ratio n, slenderness ratio L / h, and taking into account the different service ages T in offshore atmospheric environments, a finite element analysis model of existing steel frame columns is established; In order to obtain the relationship between the weight loss rate η of the steel and the corrosion depth d, it is assumed that: 1) the density of the steel does not change before and after corrosion, that is, ρ0 = ρ T ; 2) Since the length of the component is much larger than the width and height of the cross section, it is believed that corrosion only occurs in the width and height directions of the cross section of the component, that is, L0 = L T , then the weight loss rate η is expressed as: Where: m0, ρ0, A0, L0 are the initial mass, density, cross-sectional area and length of the component respectively; m T , T , A T , L T are the corresponding physical characteristics of the components when they are in service for T years; A corr Indicates the cross-sectional corrosion area; for H-shaped steel members, the cross-sectional specifications are h×b×t w ×t f , then the initial cross-sectional area is A0 = 2bt f +t w (h-2t f ), the cross-sectional corrosion area at the service age of T years is A corr =2bd+d(h-2t f ), d = r corr ·T, and then the weight loss rate η is converted to: Combining this formula with the mechanical property degradation formula (1) of corroded steel, the mechanical property index of corroded steel at the service age of T years can be obtained; The skeleton curves of the physical test model and numerical analysis model of existing corroded steel frame columns in offshore atmospheric environment are extracted, and the displacement angle data corresponding to each damage state are obtained; the test and simulation data are integrated, and the outliers in the data are eliminated according to the Pierce principle. According to formulas (6)-(8), the fragility curve parameters of each damage state of existing steel frame columns under different service ages in offshore atmospheric environment are obtained, and the fragility parameters θ under each damage state are revealed. n The evolution law of and β with the increase of service age T; through regression analysis, the correlation between vulnerability parameters and service age T is established, so as to construct a time-varying vulnerability model of existing steel frame columns in offshore atmospheric environment.

7. The method for evaluating time-varying vulnerability and residual seismic capacity of existing steel columns based on machine learning according to claim 1, characterized in that: In step six, Based on the established refined finite element models of existing steel frame columns under different design parameters and service ages and the determined damage levels, the residual seismic capacity of earthquake-damaged steel frame columns is evaluated; First, each model was preloaded to make it pre-damaged to five different damage levels, and then the damaged model was subjected to main loading to examine its residual seismic performance. Referring to the loading system recommended by AISC, the displacement angles θ were loaded at 0.375%, 0.5%, 0.75%, 1%, 1.5%, 2%, 3%, 4%..., and each displacement level was cycled twice. In the preloading stage, after 5 cycles at the target displacement reaching the set damage level, it was unloaded in the form of cyclic loading to simulate a real vibration cycle in an earthquake; then, the main loading was implemented; Extract the residual stiffness K of each damaged steel frame column within the parameter range res and compared with the initial stiffness K of the intact steel frame column; the residual bearing capacity M res The maximum bearing capacity of the intact steel frame column M max For comparison; Remaining energy consumption capacity E res The accumulated energy consumption of intact steel frame columns is E cum By comparison, the influence of various design parameters and service age on the residual seismic performance of earthquake-damaged steel frame columns is studied.

8. The method for evaluating time-varying vulnerability and residual seismic capacity of existing steel columns based on machine learning according to claim 1, characterized in that: In step eight, Based on the proposed time-varying fragility model of existing multi-age steel frame columns, the probability of occurrence of each level of damage under a given displacement angle θ can be obtained: matrix [DS i |θ](T), the damage probability matrix: Where: P[DS i |θ](T) is the existing steel frame column reaching or exceeding a certain damage state DS i The probability of; T is the service age; The residual stiffness K obtained based on PSO-BPNN res / K, residual bearing capacity ratio M res / M max and the remaining energy consumption ratio E res / E cum Combined with the vulnerability failure probability matrix, the expected value of the residual seismic capacity ratio of existing multi-age steel frame columns under a given displacement angle θ in the offshore atmospheric environment is obtained in a probabilistic sense. This expected value can be used to predict the corresponding seismic capacity safety reserve. Where: E(K res )、E(M res )、E(E res ) are the expected values ​​of residual stiffness, residual bearing capacity and residual energy dissipation respectively; P matrix [DS i |θ](T) is the probability of occurrence of each level of damage state; K res / K(DS i ), M res / M max (DS i ) and E res / E cum (DS i ) are the destruction states DS at each level i The residual stiffness ratio, residual bearing capacity ratio and residual energy consumption ratio under ; T is the service age.

Images