Hilbert-Huang transformation and physical prior fused bridge pile group system earthquake damage rapid assessment method

Abstract

The invention discloses a bridge pile group system earthquake damage rapid assessment method fusing Hilbert-Huang transformation and physical prior, belongs to the technical field of civil engineering and building earthquake resistance, and aims at solving the problems that a traditional method is lagged, physically unreasonable and limited in local monitoring. The method comprises the following steps: S1, model training and embedding: before an earthquake, constructing a deep learning model, and embedding a pile-soil vibration control equation and boundary conditions in the model as loss function constraint terms; s2, establishing a damage mapping relation: before an earthquake, establishing a quantitative mapping relation between a frequency domain index representing the overall dynamic characteristics of the system and an engineering demand parameter representing the local plastic damage degree; and S3, post-earthquake real-time evaluation: determining the overall damage degree and damage position of the structure according to the extracted frequency domain index difference on the basis of the established quantitative mapping relation during an earthquake.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of civil engineering and building anti-seismic technology, and particularly relates to a pile foundation structure earthquake damage prediction and analysis technology, which is mainly applied to foundation engineering and aims to explore the damage exploration of pile foundation as a concealed engineering, and is suitable for pile foundation structures in earthquake-prone areas and special foundation engineering with high requirements for anti-seismic performance. BACKGROUND

[0002] With the acceleration of urbanization process and large-scale construction of transportation infrastructure, the safety and seismic performance of bridge engineering are increasingly valued. As an important part of bridge foundation, the damage assessment of pile group system under earthquake has always been a key and difficult problem in the engineering field. The traditional pile group earthquake damage assessment method mainly has the following technical bottlenecks:

[0003] 1. The problem of post-disaster assessment lag

[0004] The method commonly used in current engineering practice mainly relies on post-earthquake manual investigation or offline data processing. This kind of method needs professional technicians to carry equipment to the scene, and evaluates through visual inspection, local sampling or static / dynamic testing. This evaluation method often needs several hours or even several days to complete, and it is difficult to provide timely technical support for post-disaster emergency decision-making. Especially after a strong earthquake, transportation, communication and other infrastructure may be damaged, further delaying the evaluation process and seriously affecting the timeliness of emergency rescue.

[0005] 2. The problem of missing physical mechanism

[0006] In recent years, although deep learning technology has been widely applied in the field of earthquake engineering, traditional deep learning models based on time series prediction are mostly pure "data-driven" black box models. This kind of model lacks consideration of the dynamics mechanism of pile-soil system, and it is difficult to ensure that the prediction results meet the pile-soil dynamics equation and the upper inertia action law, and the prediction results may appear to violate the physical law. For example, in predicting the seismic response of pile group, the existing model may not accurately reflect the key physical phenomena such as pile top inertia mass effect and pile-soil interaction nonlinearity, resulting in large deviation of prediction results under extreme working conditions.

[0007] 3. The problem of limitation of monitoring means

[0008] Conventional monitoring methods primarily rely on local strain gauges or accelerometers. These point-based monitoring methods have significant shortcomings: firstly, their monitoring range is limited, making it difficult to capture the overall response characteristics of the pile group system; secondly, the complex noise at the pile-soil interface easily interferes with the reliability of the monitoring data. Furthermore, existing frequency domain analysis methods (such as Fourier transform) used for processing distributed fiber optic sensor data have significant limitations when processing non-stationary seismic signals. For example, they involve a large computational load, difficulty in accurately extracting damage-related characteristic frequency components, and a lack of rapid and effective full-domain frequency domain analysis methods.

[0009] 4. The problem of one-sidedness in evaluation indicators

[0010] Current assessments often employ time-domain indicators such as displacement and strain, or single-frequency-domain indicators. However, these indicators lack a clear quantitative relationship with the degree of plastic damage to the structure. Especially when local plastic hinges appear in pile group systems, existing indicators often fail to accurately reflect the overall damage state of the structure, posing challenges to maintenance decisions. Summary of the Invention

[0011] To address the problems of lagging, physical inconsistencies, and limitations in local monitoring of traditional methods, this invention provides a rapid seismic damage assessment method for bridge pile group systems that integrates Hilbert-Huang transform and physical priors.

[0012] The present invention discloses a rapid seismic damage assessment method for bridge pile group systems that integrates Hilbert-Huang transform and physical priors. The method includes the following steps:

[0013] Step S1, Model Training and Embedding: Before the earthquake, a physical information-enhanced deep learning model is constructed to predict the seismic displacement response of a pile group system in the absence of plastic damage. The pile-soil vibration control equation and boundary conditions are embedded in the deep learning model as loss function constraints. The model is trained using a pile group system inertial action dataset. The trained deep learning model is then embedded into the structure.

[0014] Step S2, Damage Mapping Relationship Establishment: Before the earthquake, incremental dynamic analysis is performed on the finite element model of the pile group system, and combined with the undamaged baseline response predicted by the deep learning model described in Step S1, a quantitative mapping relationship is established between the frequency domain index characterizing the overall dynamic characteristics of the system and the engineering requirement parameters characterizing the degree of local plastic damage.

[0015] Step S3: Real-time post-earthquake assessment: During the earthquake, the deep learning model embedded in step S1 is used to predict the system response in the undamaged state based on the input ground motion data; simultaneously, the actual response data of the structure is acquired through distributed fiber optic sensors; time-frequency analysis is performed on the predicted undamaged response and the acquired actual response respectively to extract the same frequency domain index; based on the quantitative mapping relationship established in step S2, the overall damage degree and damage location of the structure are determined according to the differences in the extracted frequency domain indexes.

[0016] Preferably, the training process of the physical information-enhanced deep learning model in step S1 includes:

[0017] S11. Construct a dataset of inertial effects of a pile group system as a training dataset. The training dataset includes the main dataset obtained by simulating the pile group system using a finite element model, and the physical prior augmentation dataset obtained by analyzing a single-degree-of-freedom system.

[0018] S12. Construct a deep learning model based on the Transformer architecture;

[0019] S13. Define the loss function of the deep learning model. Its expression is:

[0020]

[0021] in, The data loss term is used to calculate the mean square error between the predicted and actual displacement response data;

[0022] This is the loss term in the differential equation governing pile foundation vibration. These are the weighting coefficients for the loss term in the differential equation governing pile foundation vibration.

[0023] Let j be the loss term of the boundary condition for the pile foundation. , which is the weighting coefficient of the j-th boundary condition loss term of the pile foundation, where j=1,2,3,4 represent four constraint types: pile top sliding, pile top free, pile bottom elastic hinge, and pile bottom fixed, respectively.

[0024] S14. Train the deep learning model using the training dataset, and update the model parameters by optimizing the loss function.

[0025] Preferably, the loss term in the differential equation for pile foundation vibration control Obtain by the following formula:

[0026]

[0027] in, ρA represents the horizontal deflection of the pile body, h is the axial coordinate with the pile top as the origin and the downward direction of the pile body as the positive direction, h=0 represents the pile top, h=H represents the pile bottom, H is the pile height; t is the duration of the earthquake; EI is the lateral bending stiffness of the pile; ρA is the mass per unit length of the pile. The soil reaction force borne by the pile.

[0028] Preferably, the boundary condition loss term for the pile foundation is jointly constructed from one constraint type at the pile top and one constraint type at the pile bottom;

[0029] The constraint types at the pile top include two items: pile top sliding and pile top free for j=1,2.

[0030] The types of constraints at the pile bottom include two items: elastic hinge at the pile bottom for j=3,4 and fixed pile bottom.

[0031] When the pile top is a sliding constraint type (j=1), the loss term of the j=1th boundary condition of the pile foundation is:

[0032]

[0033] When the pile top is a free constraint type (j=2), the loss term of the j=2nd boundary condition of the pile foundation is:

[0034]

[0035] Where m is the inertial mass at the top of the pile group;

[0036] When the pile bottom is a j=3 pile bottom elastic hinge constraint type, the loss term of the j=3rd boundary condition of the pile foundation is:

[0037]

[0038] When the pile bottom is a fixed constraint type (j=4), the loss term of the j=4th boundary condition of the pile foundation is:

[0039]

[0040] in, For the rotational stiffness of the pile bottom, .

[0041] Preferably, the inertial action dataset of the pile group system in step S11 is constructed in the following way: calculating the equivalent period of the pile group system. and equivalent damping ratio , in (0.7 1.5 ) and (0.5 ,2 The parameters are randomly combined within the range, and the Newmark-β method is used to calculate the seismic displacement response of a single-degree-of-freedom system under each parameter combination, thus forming the physical prior augmentation dataset.

[0042] Main dataset: By applying amplitude-modulated small ground motion sets to the finite element model of the pile group system, a pairwise input-output dataset of the displacement response of the structural system under seismic excitation and non-plastic damage is obtained.

[0043] Preferably, the frequency domain index mentioned in steps S2 and S3 is the energy-weighted root frequency. The calculation formula is as follows:

[0044]

[0045] Where n is the number of intrinsic mode functions (IMFs) obtained after empirical mode decomposition. Let be the average instantaneous frequency of the i-th IMF obtained by Hilbert-Huang transform. Let be the signal energy of the i-th IMF, which is obtained by squaring and integrating the time history of the IMF component.

[0046] Preferably, the establishment of the quantitative mapping relationship in step S2 specifically involves:

[0047] S21. Perform incremental dynamic analysis (IDA) on the finite element model of the pile group system and record the curvature ductility coefficient μ of the plastic hinge of the pile body under different intensities of ground motion.

[0048] S22. For each IDA analysis step, extract the actual seismic displacement response calculated from the finite element model. The deep learning model trained in step S1 is used to predict the undamaged displacement response under the same seismic input. ;

[0049] S23, respectively for and Empirical mode decomposition and Hilbert-Huang transform are performed, and their corresponding energy-weighted root frequencies are calculated respectively. and ;

[0050] S24. Calculate the normalized energy-weighted root frequency. ;

[0051] S25. Establish the normalized energy-weighted root frequency. The regression relationship between the plastic hinge curvature ductility coefficient μ, i.e., establishing -μ quantitative mapping relationship.

[0052] Preferably, the specific method for determining the overall damage level of the structure in step S3 is as follows:

[0053] Calculate the ratio of the energy-weighted root frequency of the actual response to the energy-weighted root frequency of the undamaged predicted response, i.e., the normalized energy-weighted root frequency. ;

[0054] Based on the establishment in step S2 -μ quantitative mapping relationship to determine the corresponding damage level;

[0055] The damage severity levels include at least the following: 1≤μ<2 indicates mild damage, indicated by a green light; 2≤μ<5 indicates moderate damage, indicated by a yellow light; 5≤μ<8 indicates severe damage, indicated by a red light; and μ≥8 indicates complete destruction, indicated by both a red and a yellow light.

[0056] Preferably, the specific method for determining the damage location of the structure in step S3 is as follows:

[0057] Based on the arrangement of distributed optical fiber sensors, the actual displacement response at different depths along the pile is obtained;

[0058] Calculate the different depths along the pile body respectively Energy weighted root frequency of the actual response And the energy-weighted root frequency of the undamaged response predicted by the deep learning model at the corresponding depth. , where z is the depth coordinate;

[0059] Search from the bottom of the pile towards the top of the pile to locate the position. First time below The depth of the location indicates the location of the plastic damage zone.

[0060] Preferably, the deep learning model trained in step S1 is embedded into the damage degree integration processing and indication module at the structural end, and is available for emergency repair personnel to access via a universal interface, providing a basis for subsequent reinforcement and maintenance.

[0061] The beneficial effects of this invention are:

[0062] (1) Integrating physical constraints with deep learning to improve the physical rationality of the prediction model: In the time series prediction of pile group seismic response, the pile-soil vibration control equation (Euler-Bernoulli beam equation) and boundary conditions are embedded as loss function terms into the deep learning model of the Transformer architecture for the first time, which solves the problem that the prediction results of the traditional black box model violate physical laws and ensures that the prediction of pile group seismic response conforms to the dynamic principle (such as inertial mass effect and pile-soil interaction).

[0063] (2) Establish the mapping relationship between frequency domain index and plastic damage: The normalized energy weighted root frequency (EWRF) based on Hilbert-Huang transform is proposed. The regression relationship between it and the curvature ductility coefficient is established through incremental dynamic analysis. By the regression relationship between the normalized energy weighted root frequency and the curvature ductility coefficient of the pile body, the degree of plastic development can be quantitatively characterized, the interference of local strain monitoring noise can be overcome, and the quantitative assessment of the degree of plastic damage in the whole domain can be realized (such as the accurate classification of mild, moderate and severe damage).

[0064] (3) Achieve rapid damage assessment after disaster: By training a deep learning model before the earthquake and embedding it into the structure, and combining distributed fiber optic sensor data with Hilbert-Huang transform analysis, we can achieve rapid frequency domain assessment of the seismic response of pile groups and real-time indication of damage level, thus solving the problem of lag in traditional methods that rely on offline calculation and manual investigation, and providing immediate support for emergency decision-making. Attached Figure Description

[0065] Figure 1 This is a framework diagram of a rapid seismic damage assessment method for bridge pile group systems that integrates Hilbert-Huang transform and physical priors, as described in this invention.

[0066] Figure 2 This is a schematic diagram of a single-degree-of-freedom system involved in customizing the dataset for the top inertial effect of a pile group system;

[0067] Figure 3 This is a schematic diagram of the pile group and its superstructure;

[0068] Figure 4 This is a schematic diagram of the frequency domain state of the upper structure under non-plastic damage;

[0069] Figure 5 This is a schematic diagram of the frequency domain state of the superstructure under plastic damage;

[0070] Figure 6 This is a schematic diagram of the frequency domain state of the potential plastic region under non-plastic damage;

[0071] Figure 7 This is a schematic diagram of the frequency domain state of the plastic region under plastic damage;

[0072] Figure 8 This is a graph showing the quantitative relationship between the normalized energy-weighted root frequency and the degree of plastic damage. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0075] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0076] Specific Implementation Method 1: The following is combined with... Figures 1 to 8 This embodiment describes a rapid seismic damage assessment method for bridge pile group systems that integrates Hilbert-Huang transform and physical priors. The method includes the following steps:

[0077] Step S1, Model Training and Embedding: Before the earthquake, a physical information-enhanced deep learning model is constructed to predict the seismic displacement response of a pile group system in the absence of plastic damage. The pile-soil vibration control equation and boundary conditions are embedded in the deep learning model as loss function constraints. The model is trained using a pile group system inertial action dataset. The trained deep learning model is then embedded into the structure.

[0078] Step S2, Damage Mapping Relationship Establishment: Before the earthquake, incremental dynamic analysis is performed on the finite element model of the pile group system, and combined with the undamaged baseline response predicted by the deep learning model described in Step S1, a quantitative mapping relationship is established between the frequency domain index characterizing the overall dynamic characteristics of the system and the engineering requirement parameters characterizing the degree of local plastic damage.

[0079] Step S3: Real-time post-earthquake assessment: During the earthquake, the deep learning model embedded in step S1 is used to predict the system response in the undamaged state based on the input ground motion data; simultaneously, the actual response data of the structure is acquired through distributed fiber optic sensors; time-frequency analysis is performed on the predicted undamaged response and the acquired actual response respectively to extract the same frequency domain index; based on the quantitative mapping relationship established in step S2, the overall damage degree and damage location of the structure are determined according to the differences in the extracted frequency domain indexes.

[0080] The above summarizes the core concept of the invention, namely the three-step method of "physical information deep learning model + frequency domain index mapping + real-time difference evaluation".

[0081] The core objective of this patent is to propose a rapid seismic damage assessment method for pile group foundations that integrates Hilbert-Huang transform and physical priors, aiming to address several technical bottlenecks in the current field of earthquake engineering regarding response analysis and damage assessment of pile group foundations. Each step is described in detail.

[0082] Step S1 involves model building, customizing a dataset of inertial effects on a pile group system, introducing the non-plastic control equations for pile-soil system vibration and vibration boundary condition losses to avoid black-box predictions in deep learning and ensure dynamic rationality. The trained deep learning model is embedded into the damage level integration and indication module at the structural end, and made available to emergency repair personnel via a universal interface, providing a basis for subsequent reinforcement and maintenance.

[0083] The training process of the physical information-enhanced deep learning model includes:

[0084] S11. Construct a dataset of inertial effects of a pile group system as a training dataset. The training dataset includes the main dataset obtained by simulating the pile group system using a finite element model, and the physical prior augmentation dataset obtained by analyzing a single-degree-of-freedom system.

[0085] S12. Construct a deep learning model based on the Transformer architecture;

[0086] S13. For time-series prediction of pile group seismic response based on deep learning, vibration control differential equations and boundary conditions of piles with different constraints are added as one of the loss constraints of the deep learning. The Transformer deep learning model is used to analyze the relationship between the seismic time history and the displacement time history at various locations of the pile group. For different pile end fixing conditions, non-plastic constraints are applied to the loss function: Define the loss function of the deep learning model. Its expression is:

[0087]

[0088] in, The data loss term is used to calculate the mean square error between the predicted and actual displacement response data;

[0089] This is the loss term in the differential equation governing pile foundation vibration. These are the weighting coefficients for the loss term in the differential equation governing pile foundation vibration.

[0090] Let j be the loss term of the boundary condition for the pile foundation. , which is the weighting coefficient of the j-th boundary condition loss term of the pile foundation, where j=1,2,3,4 represent four constraint types: pile top sliding, pile top free, pile bottom elastic hinge, and pile bottom fixed, respectively.

[0091] Each weighting coefficient can be manually adjusted based on expert experience to ensure that the contributions of each loss are balanced during training, preventing the loss with a larger scale from dominating.

[0092] S14. Train the deep learning model using the training dataset, and update the model parameters by optimizing the loss function.

[0093] Loss term in the differential equation for pile foundation vibration control in step S13 Obtain by the following formula:

[0094]

[0095] in, ρA represents the horizontal deflection of the pile body, h is the axial coordinate with the pile top as the origin and the downward direction of the pile body as the positive direction, h=0 represents the pile top, h=H represents the pile bottom, H is the pile height; t is the duration of the earthquake; EI is the lateral bending stiffness of the pile; ρA is the mass per unit length of the pile. The soil reaction force borne by the pile is 0, based on in-situ experimental research or using API specifications, for the soil reaction force on the part of the pile foundation that is exposed above ground.

[0096] The boundary condition loss term for pile foundations is jointly constructed from one constraint type at the pile top and one constraint type at the pile bottom;

[0097] The constraint types at the pile top include two items: pile top sliding and pile top free for j=1,2.

[0098] The types of constraints at the pile bottom include two items: elastic hinge at the pile bottom for j=3,4 and fixed pile bottom.

[0099] Examples illustrating common combinations of boundary condition loss terms for pile foundations:

[0100] Condition 1: Pile top free + pile bottom fixed, then the boundary condition loss term for the pile foundation is: .

[0101] Condition 2: Pile top sliding + pile bottom fixed, then the boundary condition loss term of the pile foundation is: .

[0102] Condition 3: Free top of pile + elastic hinge at bottom of pile, then the boundary condition loss term of the pile foundation is: .

[0103] Condition 4: Pile top sliding + pile bottom elastic hinge, then the boundary condition loss term of the pile foundation is: .

[0104] When the pile top is a sliding constraint type (j=1), the loss term of the j=1th boundary condition of the pile foundation is:

[0105]

[0106] When the pile top is a free constraint type (j=2), the loss term of the j=2nd boundary condition of the pile foundation is:

[0107]

[0108] Where m is the inertial mass at the top of the pile group;

[0109] When the pile bottom is a j=3 pile bottom elastic hinge constraint type, the loss term of the j=3rd boundary condition of the pile foundation is:

[0110]

[0111] When the pile bottom is a fixed constraint type (j=4), the loss term of the j=4th boundary condition of the pile foundation is:

[0112]

[0113] in, For the rotational stiffness of the pile bottom, .

[0114] In step S11, the inertial action dataset of the pile group system is constructed in the following way: the equivalent period of the pile group system is calculated. and equivalent damping ratio , in (0.7 1.5 ) and (0.5 ,2 The parameters are randomly combined within the range, and the Newmark-β method is used to calculate the seismic displacement response of a single-degree-of-freedom system under each parameter combination, thus forming the physical prior augmentation dataset.

[0115] Main dataset: By applying amplitude-modulated small ground motion sets to the finite element model of the pile group system, a pairwise input-output dataset of the displacement response of the structural system under seismic excitation and non-plastic damage is obtained.

[0116] After training with the main data and physical prior augmentation data, the trained deep learning model is embedded into the damage degree integration processing and indication module before the earthquake.

[0117] This paper describes specific techniques for enhancing the protection of datasets. The method aims to address the scarcity of high-quality training data in civil engineering by artificially generating effective data through physical mechanisms to improve the model's generalization ability.

[0118] The frequency domain index mentioned in steps S2 and S3 is the energy-weighted root frequency. The calculation formula is as follows:

[0119]

[0120] Where n is the number of intrinsic mode functions (IMFs) obtained after empirical mode decomposition. Let be the average instantaneous frequency of the i-th IMF obtained by Hilbert-Huang transform. Let be the signal energy of the i-th IMF, which is obtained by squaring and integrating the time history of the IMF component.

[0121] The "normalized energy-weighted root frequency" is a bridging parameter proposed in this invention to connect the overall dynamic characteristics of a system with local damage.

[0122] Step S2 utilizes deep learning non-plastic benchmark analysis and finite element incremental dynamic analysis to establish the relationship between normalized frequency domain indices and plastic damage based on Hilbert-Huang transform, thereby achieving quantitative mapping.

[0123] The establishment of the quantitative mapping relationship specifically refers to:

[0124] S21. Perform incremental dynamic analysis (IDA) on the finite element model of the pile group system and record the curvature ductility coefficient μ of the plastic hinge of the pile body under different intensities of ground motion.

[0125] S22. For each IDA analysis step, simultaneously extract the actual seismic displacement response calculated from the finite element model. The deep learning model trained in step S1 is used to predict the undamaged displacement response under the same seismic input. ;

[0126] S23, respectively for and Empirical mode decomposition and Hilbert-Huang transform are performed, and their corresponding energy-weighted root frequencies are calculated respectively. and The calculation formula remains the same. .

[0127] S24. Calculate the normalized energy-weighted root frequency. The normalized energy-weighted root frequency is the ratio of the energy-weighted root frequency of the actual response to the energy-weighted root frequency of the undamaged predicted response.

[0128] S25. Establish the normalized energy-weighted root frequency. The regression relationship between the plastic hinge curvature ductility coefficient μ, i.e., establishing -μ quantitative mapping relationship. See also Figure 8 As shown.

[0129] Step S3 automatically determines the degree of damage and the location of the plastic zone by comparing the frequency domain characteristics of the deep learning non-plasticity prediction and the monitoring results of the distributed sensors, and displays them in real time along with the structure.

[0130] After the bedrock seismic inversion module records the ground ground motion time history during an earthquake, the bedrock ground motion is obtained using the one-dimensional equivalent linear wave method. Based on the deep learning model in the damage degree integration and indication module, the seismic response under conditions of no plastic damage across the entire structure is predicted. This seismic response is then decomposed into several intrinsic mode functions (EMFs) using empirical mode decomposition. The Hilbert-Huang transform is applied to each EMF to obtain its instantaneous frequency. For the entire structure, the proposed frequency eigenvalues ​​under non-plastic damage conditions, i.e., the energy-weighted root frequencies, are further calculated. Simultaneously, based on the actual displacement time history data retrieved from distributed fiber optic sensors, empirical mode decomposition and Hilbert-Huang transform are also performed to calculate the energy-weighted root frequencies under the actual conditions across the entire structure. Based on the normalized energy-weighted root frequency-plastic damage relationship, the damage location and degree are determined.

[0131] The specific method for determining the overall damage level of the structure is as follows:

[0132] Calculate the ratio of the energy-weighted root frequency of the actual response to the energy-weighted root frequency of the undamaged predicted response, i.e., the normalized energy-weighted root frequency. ;

[0133] Based on the establishment in step S2 -μ quantitative mapping relationship to determine the corresponding damage level;

[0134] The degree of damage includes at least the following:

[0135] When 1≤μ<2, the pile group system is considered to be slightly damaged, meaning it has no impact on the bridge's functionality. The integrated treatment and indicator module will light up a green light according to the degree of bridge damage, indicating that minor repairs are needed.

[0136] When 2≤μ<5, the pile group system is moderately damaged but can still maintain normal use. It needs to be repaired or reinforced. Traffic control should be implemented as appropriate, and the yellow light will illuminate along with the bridge indicator module.

[0137] When 5≤μ<8, the pile group system is severely damaged, which seriously affects the bridge's functionality and makes it impassable. The red light will illuminate along with the bridge indicator module, indicating that major repairs or reinforcement are required and traffic control should be implemented immediately.

[0138] When μ≥8, the pile group system is completely destroyed and passage is prohibited. The red and yellow lights of the bridge indicator module will light up simultaneously, and reconstruction or renovation is required.

[0139] The specific method for determining the location of damage to the structure is as follows:

[0140] Based on the arrangement of distributed optical fiber sensors, the actual displacement response at different depths along the pile is obtained;

[0141] Calculate the different depths along the pile body respectively Energy weighted root frequency of the actual response And the energy-weighted root frequency of the undamaged response predicted by the deep learning model at the corresponding depth. , where z is the depth coordinate;

[0142] Search from the bottom of the pile towards the top of the pile to locate the position. First time below The depth of the location indicates the location of the plastic damage zone.

[0143] Positioning is also based on Figure 8 The quantitative relationship shown is complete.

[0144] The method of the present invention will be illustrated below with a specific example.

[0145] A bridge pile group system consists of piles, a pile cap, and a superstructure block. The pile cap is 2.3m thick and has a plan dimension of 7.5m. The superstructure block weighs 450 tons. The piles are arranged in a 3×3 pattern, with a center-to-center spacing of 3m and a diameter of 800mm. The longitudinal reinforcement ratio is 1.5%, and the pile group system uses C30 concrete. The upper layer is medium sand with a relative density of 55%, and the lower bearing layer is dense sand with a relative density of 80%. The objective is to quickly analyze the structural damage of the pile group foundation under earthquake conditions. The methodology framework is as follows: Figure 1 :

[0146] (1) For time-series prediction of pile group seismic response based on deep learning, a customized inertial action dataset was developed. Single-degree-of-freedom analysis was performed under different equivalent periods (0.8s~1.8s) and equivalent damping ratios (0.03~0.1) (e.g., Figure 2 Physical prior augmentation data was obtained and added to the deep learning dataset as supplementary data. The main deep learning data was obtained by building a finite element model of the pile group system, yielding 160 pairs of input-output data for the seismic excitation-structure non-plastic damage displacement response.

[0147] (2) A deep learning model based on the Transformer model architecture under the encoder-decoder framework was established. The aforementioned vibration control differential equation and vibration boundary conditions were added as one of the loss constraints for deep learning. The mapping relationship between the earthquake time history and the global position displacement time history of the pile group was analyzed. In this embodiment, the pile bottom is a fixed support, that is, the rotational stiffness is set to a sufficiently large value, and the pile top is in free contact. The lateral bending stiffness of the pile in the vibration control equation is 427000 kN·m. 2 The mass per unit length was set to 1.2 tons / m, and the soil reaction force borne by the piles was based on the API standard. The soil reaction force at the pile foundation protruding from the ground was 0. After training with the main data and physical prior augmentation data, the trained deep learning model was embedded into the pre-earthquake damage level integration and indication module.

[0148] (3) Before the earthquake, incremental dynamic analysis of PGA from 0.1g to 1g was carried out on the finite element model of the pile group system. During the analysis, the development of plastic hinges in the pile body was recorded, and the corresponding curvature ductility coefficients were extracted. At the same time, empirical mode decomposition was performed on the actual seismic response of the pile superstructure obtained by the finite element model and the seismic response of the pile superstructure under non-plastic state predicted by deep learning, and several intrinsic mode functions were obtained. Subsequently, Hilbert-Huang transform was applied to each intrinsic mode function to extract the instantaneous frequency. Further, the proposed energy-weighted root frequency was calculated. Based on the energy-weighted root frequency normalized by finite element analysis and deep learning prediction, a regression relationship between it and the curvature ductility coefficient at the plastic hinge was established, thereby realizing the quantitative mapping between the frequency characteristic index of the superstructure and the degree of plastic damage. Figure 8 (Red line).

[0149] (4) During an earthquake, the bedrock seismic inversion module ( Figure 3 After recording the ground motion time history, the bedrock ground motion was obtained using the one-dimensional equivalent linear wave method. Based on the deep learning model in the damage degree integrated processing and indication module, the seismic response under conditions of no plastic damage across the entire structure was predicted. The original seismic response was decomposed into several intrinsic mode functions (EMFs) using empirical mode decomposition. The Hilbert-Huang transform was applied to each EMF to obtain its instantaneous frequency. For the entire structure, the proposed frequency eigenvalues ​​under non-plastic damage conditions, i.e., the energy-weighted root frequencies, were further calculated. Simultaneously, based on the actual displacement time history data retrieved from distributed fiber optic sensors, empirical mode decomposition and Hilbert-Huang transform were also performed to calculate the energy-weighted root frequencies under the actual conditions across the entire structure. Based on the normalized energy-weighted root frequency-plastic damage degree relationship, the damage location and degree were determined.

[0150] 1) Damage Determination Method: For the frequency domain state of the superstructure of the pile group system under non-plastic damage obtained by deep learning (e.g., Figure 4 ), and the frequency domain state of the superstructure of the pile group system obtained from the monitoring data of distributed optical fiber sensors ( Figure 5 The energy-weighted root frequencies were calculated to be 1.012 Hz and 0.337 Hz, respectively. Based on the normalized energy-weighted root frequency-plastic damage relationship, [the following was found...]. Figure 8 When the normalized energy-weighted root frequency is 0.3, the degree of plastic damage is complete failure, indicating the module ( Figure 3 When the red and yellow lights are on, passage is prohibited.

[0151] 2) Damage Location Method: For the frequency characteristic values ​​of the entire structure under actual state obtained by distributed fiber optic sensor monitoring, compare them with the frequency characteristic values ​​of the entire non-plastic damage state calculated by deep learning, and search for the initial node where the energy-weighted root frequency decreases from the pile bottom to the pile top, such as... Figure 6 Frequency domain state of the potential plastic region under non-plastic damage. Figure 7 The frequency domain state of the plastic region under plastic damage has energy-weighted root frequencies of 1.292 Hz and 1.006 Hz, respectively. At this point, the node represents the location of the plastic region, as shown below. Figure 3 .

[0152] The above results verify that this patented technology has a significant effect on rapid prediction and damage analysis of seismic response of pile group foundation systems and has broad engineering application prospects.

[0153] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A rapid seismic damage assessment method for bridge pile group systems integrating Hilbert-Huang transform and physical priors, characterized in that, The method includes the following steps: Step S1, Model Training and Embedding: Before the earthquake, a physical information-enhanced deep learning model is constructed to predict the seismic displacement response of a pile group system in the absence of plastic damage. The pile-soil vibration control equation and boundary conditions are embedded in the deep learning model as loss function constraints. The model is trained using a pile group system inertial action dataset. The trained deep learning model is then embedded into the structure. Step S2, Damage Mapping Relationship Establishment: Before the earthquake, incremental dynamic analysis is performed on the finite element model of the pile group system, and combined with the undamaged baseline response predicted by the deep learning model described in Step S1, a quantitative mapping relationship is established between the frequency domain index characterizing the overall dynamic characteristics of the system and the engineering requirement parameters characterizing the degree of local plastic damage. Step S3: Real-time post-earthquake assessment: During the earthquake, the deep learning model embedded in step S1 is used to predict the system response in the undamaged state based on the input ground motion data; simultaneously, the actual response data of the structure is acquired through distributed fiber optic sensors; time-frequency analysis is performed on the predicted undamaged response and the acquired actual response respectively to extract the same frequency domain index; based on the quantitative mapping relationship established in step S2, the overall damage degree and damage location of the structure are determined according to the differences in the extracted frequency domain indexes.

2. The rapid seismic damage assessment method for bridge pile group systems integrating Hilbert-Huang transform and physical priors as described in claim 1, characterized in that, The training process of the physical information augmentation deep learning model in step S1 includes: S11. Construct a dataset of inertial effects of a pile group system as a training dataset. The training dataset includes the main dataset obtained by simulating the pile group system using a finite element model, and the physical prior augmentation dataset obtained by analyzing a single-degree-of-freedom system. S12. Construct a deep learning model based on the Transformer architecture; S13. Define the loss function of the deep learning model. Its expression is: in, The data loss term is used to calculate the mean square error between the predicted and actual displacement response data; This is the loss term in the differential equation governing pile foundation vibration. These are the weighting coefficients for the loss term in the differential equation governing pile foundation vibration. For the j-th boundary condition loss term of the pile foundation, , which is the weighting coefficient of the j-th boundary condition loss term of the pile foundation, where j=1,2,3,4 represent four constraint types: pile top sliding, pile top free, pile bottom elastic hinge, and pile bottom fixed, respectively. S14. Train the deep learning model using the training dataset, and update the model parameters by optimizing the loss function.

3. The rapid seismic damage assessment method for bridge pile group systems integrating Hilbert-Huang transform and physical priors as described in claim 2, characterized in that, Loss term in the differential equation for vibration control of pile foundation Obtain by the following formula: in, ρA represents the horizontal deflection of the pile body, h is the axial coordinate with the pile top as the origin and the downward direction of the pile body as the positive direction, h=0 represents the pile top, h=H represents the pile bottom, H is the pile height; t is the duration of the earthquake; EI is the lateral bending stiffness of the pile; ρA is the mass per unit length of the pile. The soil reaction force borne by the pile.

4. The rapid seismic damage assessment method for bridge pile group systems integrating Hilbert-Huang transform and physical priors as described in claim 3, characterized in that, The boundary condition loss term for pile foundations is jointly constructed from one constraint type at the pile top and one constraint type at the pile bottom; The constraint types at the pile top include two items: pile top sliding and pile top free for j=1,2. The types of constraints at the pile bottom include two items: elastic hinge at the pile bottom for j=3,4 and fixed pile bottom. When the pile top is a sliding constraint type (j=1), the loss term of the j=1th boundary condition of the pile foundation is: When the pile top is a free constraint type (j=2), the loss term of the j=2nd boundary condition of the pile foundation is: Where m is the inertial mass at the top of the pile group; When the pile bottom is a j=3 pile bottom elastic hinge constraint type, the loss term of the j=3rd boundary condition of the pile foundation is: When the pile bottom is a fixed constraint type (j=4), the loss term of the j=4th boundary condition of the pile foundation is: in, For the rotational stiffness of the pile bottom, .

5. The rapid seismic damage assessment method for bridge pile group systems integrating Hilbert-Huang transform and physical priors as described in claim 2, characterized in that, In step S11, the inertial action dataset of the pile group system is constructed in the following way: the equivalent period of the pile group system is calculated. and equivalent damping ratio , in (0.7 1.5 ) and (0.5 ,2 The parameters are randomly combined within the range, and the Newmark-β method is used to calculate the seismic displacement response of a single-degree-of-freedom system under each parameter combination, thus forming the physical prior augmentation dataset. Main dataset: By applying amplitude-modulated small ground motion sets to the finite element model of the pile group system, a pairwise input-output dataset of the displacement response of the structural system under seismic excitation and non-plastic damage is obtained.

6. The rapid seismic damage assessment method for bridge pile group systems integrating Hilbert-Huang transform and physical priors as described in claim 1, characterized in that, The frequency domain index mentioned in steps S2 and S3 is the energy-weighted root frequency. The calculation formula is as follows: Where n is the number of intrinsic mode functions (IMFs) obtained after empirical mode decomposition. Let be the average instantaneous frequency of the i-th IMF obtained by Hilbert-Huang transform. Let be the signal energy of the i-th IMF, which is obtained by squaring and integrating the time history of the IMF component.

7. The rapid seismic damage assessment method for bridge pile group systems integrating Hilbert-Huang transform and physical priors as described in claim 1, characterized in that, The establishment of the quantitative mapping relationship in step S2 specifically refers to: S21. Perform incremental dynamic analysis (IDA) on the finite element model of the pile group system and record the curvature ductility coefficient μ of the plastic hinge of the pile body under different intensities of ground motion. S22. For each IDA analysis step, extract the actual seismic displacement response calculated from the finite element model. The deep learning model trained in step S1 is used to predict the undamaged displacement response under the same seismic input. ; S23, respectively for and Empirical mode decomposition and Hilbert-Huang transform are performed, and their corresponding energy-weighted root frequencies are calculated respectively. and ; S24. Calculate the normalized energy-weighted root frequency. ; S25. Establish the normalized energy-weighted root frequency. The regression relationship between the plastic hinge curvature ductility coefficient μ, i.e., establishing -μ quantitative mapping relationship.

8. A rapid seismic damage assessment method for bridge pile group systems integrating Hilbert-Huang transform and physical priors as described in claim 1 or 7, characterized in that, The specific method for determining the overall damage level of the structure in step S3 is as follows: Calculate the ratio of the energy-weighted root frequency of the actual response to the energy-weighted root frequency of the predicted response without damage, i.e., the normalized energy-weighted root frequency. ; Based on the establishment in step S2 -μ quantitative mapping relationship to determine the corresponding damage level; The damage severity levels include at least the following: 1≤μ<2 indicates mild damage, indicated by a green light; 2≤μ<5 indicates moderate damage, indicated by a yellow light; 5≤μ<8 indicates severe damage, indicated by a red light; and μ≥8 indicates complete destruction, indicated by both a red and a yellow light.

9. A rapid seismic damage assessment method for bridge pile group systems integrating Hilbert-Huang transform and physical priors as described in claim 1 or 7, characterized in that, The specific method for determining the location of damage to the structure in step S3 is as follows: Based on the arrangement of distributed optical fiber sensors, the actual displacement response at different depths along the pile is obtained; Calculate the different depths along the pile body respectively Energy weighted root frequency of the actual response And the energy-weighted root frequency of the undamaged response predicted by the deep learning model at the corresponding depth. , where z is the depth coordinate; Search from the bottom of the pile towards the top of the pile to locate the position. First time below The depth of the location indicates the location of the plastic damage zone.

10. The rapid seismic damage assessment method for bridge pile group systems integrating Hilbert-Huang transform and physical priors as described in claim 1, characterized in that, The deep learning model trained in step S1 is embedded into the damage level integration processing and indication module at the structural end, and is available for emergency repair personnel to access via a general interface, providing a basis for subsequent reinforcement and maintenance.

Images