Inertial-motion load coupled pile foundation seismic safety intelligent evaluation system

Abstract

The application belongs to the technical field of pile foundation seismic resistance evaluation, and relates to an inertia-motion load coupled pile foundation seismic resistance safety intelligent evaluation system, which comprises: an equivalent feature generation module, which extracts frequency response and peak acceleration to generate time-frequency features; an initial state construction module, which constructs a grid and generates external load and initial deformation vectors; an input vector generation module, which splices displacement and features to generate a state vector; a constitutive parameter reasoning module, which inputs the network to output soil resistance and stiffness; a global matrix assembly module, which integrates internal forces and assembles a global stiffness matrix; a state update solution module, which updates deformation according to the unbalanced force and stiffness to obtain a correction; a convergence judgment and output module, which calculates a norm, iterates when the norm exceeds a limit, and generates an envelope curve when convergence is achieved. The application solves the problem of insufficient accuracy of existing evaluation methods in evaluating nonlinear load coupling responses under complex seismic motion characteristics.

Description

Technical Field

[0001] This invention belongs to the technical field of pile foundation seismic assessment and relates to an intelligent assessment system for pile foundation seismic safety based on inertial-motion load coupling. Background Technology

[0002] In the field of seismic engineering, assessing the safety performance of pile foundation structures under seismic loading is a crucial issue. The seismic response of pile foundations is jointly determined by the inertial load of the superstructure and the kinetic load of soil deformation; the coupling between these two factors increases the difficulty of mechanical analysis. Simultaneously, the constitutive relations of soil mechanics exhibit nonlinear characteristics, displaying phenomena such as strain softening and stiffness degradation. Its response mode is closely related to the frequency components, energy magnitude, and duration of the earthquake motion. Establishing an assessment model that comprehensively reflects these complex factors is a fundamental requirement in this field.

[0003] Currently, the industry commonly employs numerical evaluation methods based on nonlinear static analysis. This method typically constructs a numerical model of the pile-soil system, treating the pile as an elastic beam and the surrounding soil as a series of independent nonlinear springs distributed along the pile, thus establishing a Winkler foundation model. The model uses pre-defined static empirical curves to characterize the nonlinear behavior of the soil, thereby defining the correspondence between soil resistance and pile lateral displacement at various depths. During the evaluation process, estimated peak inertial loads and peak kinetic displacement loads are applied to the model, and iterative calculations are used to obtain the deformation and internal force state of the pile.

[0004] The aforementioned traditional methods have certain technical limitations when dealing with complex seismic scenarios. First, existing models mostly use static empirical parameters, which, once set, are difficult to dynamically reflect the dependence of soil constitutive relations on characteristics such as the frequency distribution of specific seismic energy bands, and are prone to evaluation bias when facing atypical seismic events. Second, the independent spring model based on the Winkler foundation assumption lacks consideration of spatial synergistic effects and does not fully reflect the nonlocal coupling effect of the overall pile deformation on the local soil stress field, making it difficult to reflect the changes in the stress state of distant soil caused by large local displacements in conventional models. In addition, conventional load estimation and application methods are relatively simplified and cannot fully demonstrate the coupled evolution characteristics of inertial and kinetic effects in the nonlinear response process. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides an intelligent assessment system for the seismic safety of pile foundations based on inertial-motion load coupling.

[0006] An intelligent assessment system for seismic safety of pile foundations coupled with inertial-motion loads, comprising: The equivalent feature generation module is used to acquire the initial seismic wave signal, extract the frequency response parameters, and combine them with the peak ground acceleration to generate the time-frequency feature vector of the ground motion; The initial state construction module is used to construct a discrete grid based on the pile foundation design parameters, apply the peak inertial load to the top node to form an external load vector, calculate and generate a pure motion displacement profile, and assign zero displacement values ​​to the grid nodes to generate the initial deformation vector of the entire pile. The input vector generation module is used to extract the nodal displacement components of the discrete grid and combine them into a total pile deformation vector. The pile-soil relative displacement components are calculated by subtracting the nodal displacement components from the components at the depth corresponding to the pure motion displacement profile. The pile-soil relative displacement components, the total pile deformation vector, and the ground motion time-frequency characteristic vector are concatenated to generate a nodal state vector. The constitutive parameter inference module is used to input the node state vectors at each depth into the constitutive relation prediction neural network and output the node soil resistance and node soil tangential stiffness at the corresponding depth. The global matrix assembly module is used to integrate and accumulate the soil resistance of all nodes to obtain the comprehensive internal force vector, and assemble the global tangent stiffness matrix based on the tangent stiffness of the soil at each node. The state update solution module is used to calculate the difference between the external load vector and the comprehensive internal force vector to generate the unbalanced force vector. It solves the displacement correction amount based on the global tangent stiffness matrix and the unbalanced force vector, and adds the displacement correction amount to the current full pile deformation vector to generate the updated full pile deformation vector. The convergence determination and output module is used to extract the components corresponding to the lateral translational degrees of freedom in the unbalanced force vector and calculate their norms. If the norm is greater than the preset allowable error threshold, the updated full pile deformation vector is used as the input for the next iteration step, and the step of extracting nodal displacement components is returned. If the norm is less than or equal to the allowable error threshold, the iteration stops, and the pile internal force envelope is calculated based on the current full pile deformation vector.

[0007] A further aspect of the present invention involves acquiring the initial seismic wave signal, including: Obtain the environmental white noise variance threshold and the continuous signal collected at the bedrock surface; Configure an adaptive cutoff frequency for the Wiener filter that matches the variance threshold of the ambient white noise. A Wiener filter was used to filter out baseline drift and high-frequency noise in the continuous signal based on an adaptive cutoff frequency, and the filtered time series was used as the initial seismic wave signal.

[0008] A further aspect of the present invention involves extracting frequency response parameters and combining them with peak ground acceleration to generate a time-frequency feature vector of ground motion, including: The initial seismic wave signal is decomposed based on wavelet packet transform; Extract the coefficient values ​​of each preset frequency band after decomposition, sum them together with the time step, and obtain the frequency response parameters; The maximum absolute value of the initial seismic wave signal is extracted as the peak ground acceleration; Peak ground acceleration is concatenated with frequency response parameters to generate a time-frequency feature vector of ground motion.

[0009] A further aspect of the present invention generates an external load vector, comprising the following steps: Read the pile length, cross-sectional diameter and elastic modulus from the pile foundation design parameters, and discretize the pile body into multiple foundation beam elements to generate a discrete mesh; The peak inertial load of the superstructure was determined using the limit equilibrium method, and the pure motion displacement profile was calculated. The peak inertial load is applied to the translational degrees of freedom of the top node of the discrete mesh to generate the external load vector.

[0010] A further aspect of the present invention generates a node state vector, comprising the following steps: Extract the lateral displacement components of the discrete grid nodes in the current iteration step as the node displacement components; The nodal displacement components of all depth nodes are spliced ​​together in depth order to obtain the total pile deformation vector; The component corresponding to the depth in the pure motion displacement profile is extracted and subtracted from the nodal displacement component to obtain the pile-soil relative displacement component. The relative displacement components of the pile and soil, the deformation vector of the entire pile, and the time-frequency characteristic vector of the ground motion are concatenated and standardized to generate the node state vector.

[0011] A further embodiment of the present invention outputs the corresponding nodal soil resistance and nodal soil tangential stiffness, including the following steps: The node state vectors are imported into the constitutive relation prediction neural network for feature encoding; The spatial coupling characteristics of the deformation vector of the whole pile are extracted using the spatiotemporal attention mechanism module; The displacement space coupling features are transformed into weighted feature vectors by combining the weight matrix; By analyzing the weighted eigenvectors through a nonlinear activation layer, the corresponding nodal soil resistance and nodal soil tangential stiffness are output.

[0012] A further aspect of this invention utilizes a spatiotemporal attention mechanism module to extract the displacement spatial coupling features of the whole pile deformation vector, including the following steps: The deformation vector of the entire pile is reshaped into a spatial sequence feature matrix, and then mapped into a query matrix, a key matrix, and a value matrix, respectively. Perform a dot product operation between the query matrix and the key matrix to obtain the initial similarity score matrix; The attention weights are obtained by normalizing the initial similarity score matrix. The attention weights are weighted and summed with the value matrix to generate displacement space coupling features.

[0013] A further aspect of the present invention involves forming a global tangent stiffness matrix, comprising the following steps: The soil resistance at the nodes is integrated along the discrete grid to obtain the comprehensive internal force vector; Retrieve the structural stiffness matrix; The tangential stiffness of the soil at each node is assigned to the position corresponding to the lateral displacement degree of freedom in the diagonal matrix according to the pile foundation connectivity topology, and then added to the structural stiffness matrix to assemble a global tangential stiffness matrix.

[0014] A further aspect of the present invention involves generating an updated full-pile deformation vector, comprising the following steps: Calculate the difference between the external load vector and the combined internal force vector to generate the unbalanced force vector; A system of equations is constructed based on the global tangent stiffness matrix and the unbalanced force vector, and the displacement correction is obtained by solving the equations. The displacement correction is added to the current full pile deformation vector to generate the updated full pile deformation vector.

[0015] A further aspect of the present invention involves calculating and generating the envelope of internal forces within the pile, comprising the following steps: Extract the components corresponding to the lateral translational degrees of freedom from the unbalanced force vector to form a translational subvector, and calculate the norm of the translational subvector; If the norm is greater than the preset tolerance threshold, the updated full pile deformation vector will be used as the input for the next iteration step, and the process will return to the step of extracting nodal displacement components. If the norm is less than or equal to the allowable error threshold, extract the current full pile deformation vector, and calculate the bending moment and shear force distribution based on the finite difference method to generate the pile internal force envelope.

[0016] In summary, the present invention has the following beneficial technical effects: 1. This invention generates a seismic motion time-frequency feature vector by extracting the frequency response parameters and peak ground acceleration of the initial seismic wave signal, and uses this vector as the input feature of a constitutive relation prediction neural network. Because this feature vector combines peak ground acceleration information with energy frequency domain distribution information extracted based on wavelet packet transform, the neural network can identify the specific effects of different frequency components on soil stiffness degradation and strength reduction during inference. This mechanism changes the traditional calculation mode of soil constitutive response, which relies solely on displacement amplitude, allowing soil constitutive parameters to dynamically adapt to the spectral characteristics of the input seismic motion signal, thereby improving the accuracy of the system's prediction of the nonlinear behavior of pile-soil systems in the face of seismic events with different frequency characteristics.

[0017] 2. This invention constructs a full pile deformation vector by extracting the lateral displacement components of the nodes and utilizes a spatiotemporal attention mechanism to extract the spatial coupling features of this vector, thereby achieving quantitative processing of the spatial nonlocal coupling effect of pile-soil interaction. In each iteration predicting the soil resistance at a specific depth, the input features not only include the local relative displacement between the pile and soil but also encompass global deformation information representing the overall bending and shear modes of the pile. The spatiotemporal attention mechanism incorporates the influence of the deformation of distant nodes on the local soil stress state of the current node by calculating the dependency weights between node deformations at different depths. This approach ensures that the calculated soil resistance distribution reflects the mechanical characteristics of a continuous medium, providing a data foundation for predicting objective and reasonable pile bending moment and shear force distributions.

[0018] 3. This invention employs a constitutive relation prediction neural network to establish an adaptive constitutive parameter inference mechanism, and embeds this neural network within a state update iterative solution framework. Within each iteration of the evaluation system, the neural network, based on real-time local deformation, global spatial deformation state, and seismic motion characteristics, infers and outputs nodal soil resistance and nodal soil tangential stiffness compatible with the current stress state. This scheme introduces data-driven, nonlinear soil mechanics constitutive parameters into the real-time updating and solving process of the macroscopic structural mechanics equilibrium equations, ensuring that the final displacement convergence solution satisfies both the static equilibrium equations of the physical system and the fundamental laws of nonlinear material evolution, thereby enhancing the applicability of the evaluation system under different geological conditions and seismic loads. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention.

[0020] Figure 1 This is a schematic diagram of the framework in the embodiments of this application.

[0021] Figure 2 This is a flowchart illustrating an embodiment of this application.

[0022] Figure 3 This is an energy distribution curve of a preset frequency band after wavelet packet transform decomposition in this embodiment of the application.

[0023] Figure 4 This is a graph showing the distribution of pure motion displacement profile along depth in the embodiments of this application.

[0024] Figure 5 This is a curve showing the distribution of the tangential stiffness of the soil at the node as a function of depth in the embodiments of this application.

[0025] Figure 6This is a schematic diagram of the bending moment envelope of the internal forces in the pile body in the embodiments of this application. Detailed Implementation

[0026] The following is in conjunction with the appendix Figures 1-6 A preferred description of the present invention is provided below.

[0027] See attached document Figures 1-2 This invention proposes an intelligent assessment system for the seismic safety of pile foundations based on inertial-motion load coupling, comprising the following modules: The equivalent feature generation module is used to acquire the initial seismic wave signal, extract the frequency response parameters, and combine them with the peak ground acceleration to generate the time-frequency feature vector of the ground motion; The initial state construction module is used to construct a discrete grid based on the pile foundation design parameters, apply the peak inertial load to the top node to form an external load vector, calculate and generate a pure motion displacement profile, and assign zero displacement values ​​to the grid nodes to generate the initial deformation vector of the entire pile. The input vector generation module is used to extract the nodal displacement components of the discrete grid and combine them into a total pile deformation vector. The pile-soil relative displacement components are calculated by subtracting the nodal displacement components from the components at the depth corresponding to the pure motion displacement profile. The pile-soil relative displacement components, the total pile deformation vector, and the ground motion time-frequency characteristic vector are concatenated to generate a nodal state vector. The constitutive parameter inference module is used to input the node state vectors at each depth into the constitutive relation prediction neural network and output the node soil resistance and node soil tangential stiffness at the corresponding depth. The global matrix assembly module is used to integrate and accumulate the soil resistance of all nodes to obtain the comprehensive internal force vector, and assemble the global tangent stiffness matrix based on the tangent stiffness of the soil at each node. The state update solution module is used to calculate the difference between the external load vector and the comprehensive internal force vector to generate the unbalanced force vector. It solves the displacement correction amount based on the global tangent stiffness matrix and the unbalanced force vector, and adds the displacement correction amount to the current full pile deformation vector to generate the updated full pile deformation vector. The convergence determination and output module is used to extract the components corresponding to the lateral translational degrees of freedom in the unbalanced force vector and calculate their norms. If the norm is greater than the preset allowable error threshold, the updated full pile deformation vector is used as the input for the next iteration step, and the step of extracting nodal displacement components is returned. If the norm is less than or equal to the allowable error threshold, the iteration stops, and the pile internal force envelope is calculated based on the current full pile deformation vector.

[0028] In one embodiment of the present invention, the equivalent feature generation module is configured to perform the following steps: The initial seismic wave signal is decomposed based on wavelet packet transform; Extract the coefficient values ​​of each preset frequency band after decomposition, sum them together with the time step, and obtain the frequency response parameters; The maximum absolute value of the initial seismic wave signal is extracted as the peak ground acceleration; Peak ground acceleration is concatenated with frequency response parameters to generate a time-frequency feature vector of ground motion.

[0029] The system acquires the raw seismic motion input sequence and performs multidimensional signal analysis to generate time-frequency characteristic vectors of the seismic motion. This process is performed by a data preprocessing module integrated within the evaluation system. This module first acquires continuous initial seismic wave signals collected at the bedrock surface of the target analysis site from a specified storage medium or via a network interface. The signal is stored in a digital format, such as a multi-line data file containing timestamps and acceleration amplitudes. Before transformation, the system first acquires the environmental white noise variance threshold returned by the input digital channel monitoring equipment and configures an adaptive cutoff frequency for the Wiener filter matched to the environmental white noise variance threshold accordingly.

[0030] A Wiener filter is used to filter out baseline drift and high-frequency noise from the initial seismic signal according to an adaptive cutoff frequency. The filtered and smoothed time series is then updated to the initial seismic signal used in subsequent analyses. After acquiring the smoothed signal, the system calls internal signal processing libraries, such as the PyWavelets library in the Python environment or the Wavelet Toolbox in the MATLAB environment, to perform wavelet packet transform decomposition on the initial seismic signal.

[0031] When performing this transformation, the system will preset the mother wavelet function, such as the db4 wavelet, and the number of decomposition levels. The transform operation expands the original signal in the time-frequency domain, generating... The system generates coefficient sequences for each of several mutually orthogonal sub-bands. Then, for each sub-band's coefficient sequence, the system calculates its energy, which serves as a frequency response parameter characterizing the intensity of seismic energy distribution within that band. Specifically, the energy is calculated by summing the squares of all coefficients within the sub-band along with the time step; the formula is as follows: In the formula, Indicates the index of the sub-band. This indicates the index of the coefficients within that sub-band. This represents the total number of coefficients within this frequency band. The constant sampling period of the signal. By sampling all Perform this calculation on each sub-band, and the system obtains a set of frequency response parameters.

[0032] At the same time, the processing module also iterates through the initial seismic wave signal, extracts the maximum absolute value, and records it as the peak ground acceleration. .

[0033] Finally, the system uses the peak ground acceleration as an element and concatenates it with a set of frequency response parameters obtained from the previous calculations to form a one-dimensional vector. This vector is the time-frequency characteristic vector of the seismic motion corresponding to the current seismic resistance condition. This vector fully encapsulates the peak intensity information of the earthquake and the energy distribution characteristics in the frequency domain, and outputs it as structured data to subsequent steps.

[0034] Initial seismic wave signal refers to a signal with a constant sampling period. Collected one-dimensional discrete array ,in This represents the sampling point number. Wavelet packet transform is a signal analysis technique that iteratively decomposes a signal into different frequency subspaces using high-pass and low-pass filter banks, providing a more refined frequency division for the high-frequency components compared to standard wavelet transform.

[0035] The preset frequency band is determined by the number of decomposition levels of the wavelet packet transform. and signal sampling frequency Jointly decided, the first The frequency range covered by each frequency band is: Set the number of decomposition layers Typically, based on engineering experience, the sampling frequency is set between 3 and 5. Assuming the application scenario is typical seismic analysis in civil engineering, the sampling frequency of the seismic records... Set to 100 Hz. The Nyquist frequency range of 0 to 50 Hz can be divided into 16 sub-bands, each 3.125 Hz wide. This resolution is sufficient to identify the main frequency components of most earthquakes. The frequency response parameter is the energy value calculated for each preset frequency band. The set of, where The range of values ​​is arrive Peak ground acceleration Defined as the maximum absolute value of acceleration throughout the entire time series signal, i.e. Seismic motion time-frequency eigenvectors It is a dimension of The column vector is defined as follows: .

[0036] In a specific example, the system acquires an initial seismic wave signal with a duration of 20.48 s, and its sampling frequency is... The frequency is 100 Hz, and it contains 2048 sampling points. The data preprocessing module determines the peak ground acceleration by traversing these 2048 data points. for The system configuration uses the aforementioned db4 mother wavelet and sets the number of decomposition levels. The initial seismic signal is subjected to wavelet packet transform (WPT). This operation decomposes the signal into 16 sub-bands, each with a bandwidth of 3.125 Hz. Subsequently, the system applies an energy calculation formula to the coefficients of each band. The resulting set of frequency response parameters, composed of the energy values ​​of the 16 bands, is [15.2, 45.8, 22.1, 8.4, 5.1, 2.3, 1.8, 1.1, 0.9, 0.7, 0.6, 0.5, 0.4, 0.3, 0.2, 0.1], where the units are all [missing information]. For the frequency response parameters (energy values) distribution of each preset frequency band, please refer to [link / reference]. Figure 3 As shown. The final system will determine the peak ground acceleration. By concatenating this with a set of frequency response parameters, a 17-dimensional time-frequency feature vector of ground motion is generated. Its vector values ​​are [4.2, 15.2, 45.8, 22.1, 8.4, 5.1, 2.3, 1.8, 1.1, 0.9, 0.7, 0.6, 0.5, 0.4, 0.3, 0.2, 0.1]. T The vector, after being formatted, is passed as output to subsequent computation processes.

[0037] In one embodiment of the present invention, the initial state construction module is configured to perform the following steps: Read the pile length, cross-sectional diameter and elastic modulus from the pile foundation design parameters, discretize the pile body into multiple foundation beam elements to generate a discrete mesh; use the limit equilibrium method to determine the peak inertial load of the superstructure and calculate the pure motion displacement profile; apply the peak inertial load to the translational degrees of freedom of the top node of the discrete mesh to generate the external load vector.

[0038] A nonlinear static physical solver for pile foundations is constructed, and the predicted site load boundaries are analyzed based on this solver to generate external load vectors and initial deformation vectors for the entire pile. The physical model construction and solution module within the system constructs the initial numerical model discrete mesh based on the nonlinear Winkler foundation beam theory in a programmed manner, using site pile foundation design parameters retrieved from the project database.

[0039] Specifically, this module reads parameters such as the total length, cross-sectional diameter, material elastic modulus, and moment of inertia of the pile, and discretizes the continuous pile body along the depth direction into a series of beam elements connected by nodes. The pile top is designated as node 0, and the elements are numbered sequentially downwards along the depth. At each node location, the system arranges nonlinear soil springs along the side of the pile body to simulate the interaction between the pile and the soil. The mechanical behavior of these springs follows a preset... - Constitutive relations. The module estimates two main seismic loads using the limit equilibrium method: On the one hand, the independent peak inertial load of the superstructure acting on the pile top is estimated, which represents the maximum horizontal shear force generated by the mass of the superstructure during the earthquake.

[0040] On the other hand, the free-field pure motion displacement profile caused by an earthquake can be estimated through site response analysis or simplified normative methods. This profile defines the peak displacement of the soil relative to the bedrock at different depths.

[0041] The module applies independent peak inertial loads as concentrated forces to the pile top nodes of the discrete grid of the numerical model, forming a separate external load vector that defines the force boundary conditions of the system.

[0042] The calculated pure motion displacement profile is saved as a site displacement reference vector and used to calculate the actual pile-soil relative deformation in subsequent iterations.

[0043] To initiate subsequent iterative calculations, the module assigns zero displacement parameters to all nodes of the discrete mesh of the numerical model, assuming that the pile is in an undeformed vertical state at the moment the calculation begins, thereby establishing an initial deformation vector of the entire pile with all components being zero. This vector serves as the starting point for subsequent iterative solutions.

[0044] The pile foundation design parameters include pile length. Pile diameter Elastic modulus of pile material and moment of inertia of cross section Structured datasets, which are typically derived from engineering design drawings or configuration files.

[0045] The nonlinear Winkel foundation beam theory simplifies the pile body into a one-dimensional Euler-Bernoulli beam and the surrounding soil into a series of depth-independent nonlinear springs in a soil mechanics model. Its governing differential equations for sections without external lateral load distribution are: The numerical model discrete mesh decomposes the continuous pile body into components using the finite difference method or the finite element method. A set of nodes, with node spacing This can be preset, for example, to 0.5 m. The limit equilibrium method is a classic method in geotechnical engineering for assessing slope or foundation stability; here it is used to conservatively estimate the maximum surface displacement that an earthquake might cause when a complete time-history analysis is lacking. Independent peak inertial load. It is usually estimated as ,in It is the effective mass of the superstructure. It is based on the site type and basic structural cycle. The spectral acceleration was obtained from the design reaction spectrum.

[0046] Pure motion displacement profile It is a description of the peak lateral displacement of soil in a free site as a function of depth. A changing function or discrete vector. External load vector. It is a generalized force vector that represents all external active forces applied to the pile-soil system; in this framework, it only includes the forces at the pile top. Its corresponding elements at the remaining nodes are all zero. Initial deformation vector of the entire pile. It is a dimension The zero vector, where correspond Each node has two degrees of freedom: lateral displacement and rotation. The form is as follows: .

[0047] In a specific example, the system needs to analyze a bored pile with a length of 30m, a circular cross-section with a diameter of 1.0m, and cast-in-place concrete using C30 concrete. The system reads the pile foundation design parameters from the database, including the pile length. elastic modulus And calculate the moment of inertia of the cross section. The system sets the discrete node spacing to 1.0m, thereby generating a discrete mesh for the numerical model containing 31 nodes.

[0048] Next, the module performs load estimation. The effective mass of the superstructure is 500 tons, and the corresponding spectral acceleration found from the design response spectrum is 0.5 g, approximately... The independent peak inertial load is calculated to be 2450 kN. Simultaneously, the peak surface displacement is determined to be 0.15 m using the limit equilibrium method. Assuming that the soil displacement decreases linearly along a depth of 20 m, a pure motion displacement profile is generated. The distribution pattern of the pure motion displacement profile along the depth is as follows: Figure 4 As shown. Based on this, the system constructs the external load vector. This vector is a 62-dimensional column vector, with a value of 2450 kN only at the first element corresponding to the lateral translational degree of freedom at node 0, and the remaining 61 elements being 0. Simultaneously, the system generates a pure motion displacement profile. Save it into memory. Finally, the system initializes a full-pile initial deformation vector with a dimension of 62. And set all its elements to 0. This zero vector is related to the external load vector. Together with the initial conditions of the physics solver, they are passed to subsequent iterative analysis steps.

[0049] In one embodiment of the present invention, the input vector generation module is configured to perform the following steps: Extract the lateral displacement components of the discrete grid nodes in the current iteration step as node displacement components; concatenate the node displacement components of all depth nodes in depth order to obtain the whole pile deformation vector; extract the component corresponding to the depth in the pure motion displacement profile and subtract it from the node displacement components to obtain the pile-soil relative displacement components; concatenate the pile-soil relative displacement components, the whole pile deformation vector and the ground motion time-frequency feature vector and perform standardization processing to generate the node state vector.

[0050] The dynamic nodal spatial response of the numerical model is collected and combined with the time-frequency feature vector of ground motion to generate nodal state vectors for subsequent neural network inference. At the initial moment of the iteration, i.e., the... In the next iteration, the system's physical model construction and solution module extracts the nodal displacement components at each relative depth position of the numerical model in the current iteration state from its internal data structure. For the ... The th iteration Each node has a signal representing its lateral displacement component. .exist The initial iteration step, these displacement values ​​are derived from the initial deformation vector of the entire pile. Therefore, all values ​​are zero.

[0051] The module will be all The nodal displacement components of each node and their corresponding nodal rotation signals By combining the two equations, a complete pile deformation vector representing the global bending and shear stress modes of the pile is formed. .exist At that time, this That is, the initial deformation vector of the entire pile. Next, in order to provide comprehensive and context-sensitive input information to subsequent neural networks, the system performs a data stream concatenation operation.

[0052] For each depth node on the pile The system extracts the free-field motion displacement components corresponding to the depth from the pure motion displacement profile. The difference between the current node displacement component and the site displacement is calculated to obtain the pile-soil relative displacement component that characterizes the actual stress deformation. .

[0053] The system will measure the pile-soil relative displacement components at this node. Deformation vector of the entire pile in the initial state and the generated time-frequency feature vector of ground motion. The sequences are sequentially concatenated and superimposed. To eliminate the interference of numerical differences between different physical quantities on the deep learning feedforward computation, the system performs Z-score normalization on the concatenated sequences to generate high-dimensional sequences corresponding to the depth nodes. The input vector is the node state vector. This input vector, as an information packet, contains not only the local deformation of the node, but also the global deformation mode of the pile and the energy frequency fingerprint of the external seismic motion, providing sufficient physical information for the intelligent reconstruction of constitutive relations in subsequent steps. This process is repeated for each node on the pile to generate a set of node state vectors, ready to be input into the neural network in the next step.

[0054] Among them, nodal displacement components It is in the In the next iteration, the pile body stored in the solver memory is the first... Lateral displacement values ​​at discrete nodes, in meters (m). Total pile deformation vector. It is in the In the next iteration, the displacements of all nodes are... and corner The vector formed by arranging the nodes in order, i.e. Its dimensions are This represents the complete deformation shape of the pile body during this iteration step. Node state vector This is the first The iteration of the ... A one-dimensional vector constructed from nodes is formed by concatenating the aforementioned three data components; its mathematical expression is: The semicolon represents the vertical concatenation of vectors. That is The dimension of this vector is... The design of this input vector enables the neural network not only to sense local displacement, but also to capture nonlocal coupling effects through the global deformation vector, and to perceive the characteristics of the excitation source through the seismic feature vector.

[0055] In a specific example, the system begins the first... This is the second iteration. The physical model contains 31 nodes, i.e. The system first processes the initial deformation vector of the entire pile. This is a vector where all 62 elements are 0. For any node in the discrete grid of the pile body, such as node number 5 5m below the pile top, the system first extracts its nodal displacement components. Since this is the initial iteration, this value is derived from the initial deformation vector of the entire pile. The corresponding position is obtained, and its value is 0m. Simultaneously, the total pile deformation vector of the current iteration step... The known 62-dimensional zero vector is then used. The system then retrieves the seismic motion time-frequency characteristic vector. , it is The dimension is 17. At this point, the system extracts the free-field motion displacement components corresponding to a depth of 5m. Based on the profile distribution, its value is 0.11m, thus the pile-soil relative displacement component of node 5 is calculated. Subsequently, the system generates the corresponding node state vector for node number 5. By using scalars ,vector sum vector By concatenating and superimposing the vectors, a vector with a dimension of 80 is finally obtained.

[0056] The vector The specific value is After being standardized, the data is temporarily stored in memory, waiting to be fed into the neural network in the next step.

[0057] In one embodiment of the present invention, the constitutive parameter inference module is configured to perform the following steps: The node state vectors are imported into the constitutive relation prediction neural network for feature encoding; The spatial coupling characteristics of the deformation vector of the whole pile are extracted using the spatiotemporal attention mechanism module; The displacement space coupling features are transformed into weighted feature vectors by combining the weight matrix; By analyzing the weighted eigenvectors through a nonlinear activation layer, the corresponding nodal soil resistance and nodal soil tangential stiffness are output.

[0058] Specifically, as the iteration process begins, the system activates a pre-configured constitutive relation prediction neural network to perform deep analysis on the node state vectors generated in the previous step, and generates nonlinear soil evolution constitutive parameters accordingly. The execution entity is the pre-trained offline constitutive relation prediction neural network. For any depth node on the pile... The system generates node state vectors that encapsulate local deformation, global modes, and seismic motion characteristics. The data is imported as input into the deep learning feedforward computation channel of the constitutive relation prediction neural network.

[0059] Within this channel, the various components of the node state vector are processed in a specific orientation. Specifically, this includes the overall pile deformation vector. A portion of the data is fed into the built-in spatiotemporal attention mechanism module. This module, combined with the pre-trained network weight matrix, calculates self-attention weights to map in real time the spatial coupling characteristics of displacement caused by the overall deformation of the pile, i.e., the influence of the displacement of one part of the pile on the resistance of another part of the soil. The other parts of the network process the relative displacement components between the pile and the soil. and the time-frequency eigenvector of ground motion After multiple nonlinear transformations and feature fusion, the network finally predicts and outputs two key constitutive parameters: The first output is the number of iterations in the next iteration. Each depth node should provide an intrinsic coupled soil resistance value, which is output as a signal, representing the nodal soil resistance. .

[0060] The second output is the differential tangent stiffness value of the soil resistance at the current displacement point. This value is output as another signal, namely the tangent stiffness of the soil at the node. This process occurs in the current iteration step. In the process, all piles will be processed in parallel or serially. Each node executes, ultimately generating a set of nodal soil resistance and nodal soil tangential stiffness covering the entire pile depth, and assembling these signals into a global mechanical matrix.

[0061] The computation process of the spatiotemporal attention mechanism within this neural network can be abstractly expressed as follows: Before performing this calculation, the dimension is The total pile deformation vector The first thing to be reshaped into dimensions is... The spatial sequence feature matrix, where each row corresponds to the displacement and rotation features of a depth node.

[0062] Subsequently, the spatial sequence feature matrix is ​​linearly mapped through different pre-trained network weight matrices to generate query matrices. Key matrix Sum matrix , is the scaling factor for the feature dimension. This is achieved through the transpose dot product. Calculated The similarity score matrix, whose off-diagonal elements quantify the nonlocal coupling effect of deformation of remote nodes on the current node. The entire constitutive relation prediction neural network operation process can be represented as a function. Its input is the node state vector. The output is ,in The set of pre-trained network weight matrices representing the entire network.

[0063] Among them, the constitutive relation prediction neural network is a deep neural network structure containing an input layer, multiple hidden layers and an output layer. Its pre-configuration state means that the weight matrix of the network has been determined through offline training.

[0064] Specifically, the network architecture includes an input mapping encoding layer for feature dimensionality reduction, a multi-head spatiotemporal attention layer for extracting nonlocal effects, and a fully connected decoder layer for outputting constitutive parameters. The training dataset typically consists of thousands of paired input-output examples, with inputs such as seismic motion, site conditions, and pile foundation parameters; and outputs such as the entire pile body obtained from three-dimensional finite element nonlinear analysis. - response.

[0065] In one embodiment of the present invention, to ensure the reliability and reproducibility of the training data, i.e., the real labels, the three-dimensional finite element nonlinear analysis specifically employs general-purpose finite element software for batch parametric modeling and generation. In the finite element model, the pile is simulated using three-dimensional solid elements with a linear elastic constitutive model; the surrounding soil is also simulated using three-dimensional solid elements, and its constitutive relationship adopts the Mohr-Coulomb elastoplastic model or the modified Cambridge model, which can reflect the nonlinear yield characteristics of the soil; the pile-soil contact interface adopts a master-slave contact algorithm, the tangential direction adopts the Coulomb friction model, and the normal direction adopts hard contact to allow pile-soil separation; the lateral and bottom boundaries of the model adopt viscoelastic artificial boundary conditions to absorb the reflected energy of seismic waves. By inputting seismic waves with different spectra and peak values ​​to the bottom of the model, the time history data of soil node reactions and relative displacements at each depth are extracted, thereby generating the full-process pile body py response labels for neural network training.

[0066] The training objective of the network is to minimize its predicted node resistance. and differential tangent stiffness The multi-task joint loss between the finite element simulation data and the data is defined by the following loss function: in To balance the magnitudes of the two prediction targets and the adaptive weighting coefficients, due to tangent stiffness... The value is usually higher than the resistance. Larger orders of magnitude The introduction of this feature prevents stiffness error from dominating the gradient update direction during network backpropagation. The training process employs the Adam adaptive optimization algorithm to iteratively update the network weights.

[0067] The spatiotemporal attention mechanism is a neural network module whose spatial dimension is reflected in its ability to process the full pile deformation vector as a sequence input. It calculates the dependency weights between any two nodes; its time dimension is implicit in the macroscopic iterative loop of the entire method, that is, the network in the ______ ... The output of step 1 will affect the output of step 2. The input state of each step is used to simulate the characteristics of physical processes evolving over time. Displacement-space coupling is a complex phenomenon in soil mechanics, referring to how displacement at a certain point in the pile can alter the stress state of the soil at a distance through the continuity of the pile and soil, thus affecting its resistance. This effect is traditionally considered to exist at various depths... - In Winkel models where springs are independent, the soil resistance at the nodes is neglected. The physical unit is force per unit length, for example, kN / m. (Nodal soil tangential stiffness) The physical unit of force is the square of a unit length, for example... .

[0068] In a specific example, the system processes node 5 in the 0th iteration ( The node state vector generated at time ) This 80-dimensional vector The data is input into a constitutive relation prediction neural network. Inside the network, the full pile deformation vector containing 62 zero values ​​is fed into a spatiotemporal attention module. Since the initial displacement of all nodes is zero, the attention weights tend towards a baseline distribution, reflecting the inherent correlation between different parts of the pile in a static state. Simultaneously, other parts of the network process non-zero relative displacement inputs and 17-dimensional seismic motion feature inputs. Assuming the network undergoes multiple layers of matrix multiplication and activation function calculations, such as ReLU, the computational logic in the simplified output layer is as follows: as well as ,in It is the activation value vector of the second to last layer. and These are the weights and biases of the output layer. In this initial state, due to the initial pile-soil relative displacement caused by free-field motion, the network predicts that the soil has been activated and generates resistance. The predicted nodal soil resistance is assumed to be... The initial tangential stiffness is -80 kN / m. However, the soil possesses a certain stiffness in its initial state, therefore the network predicts a non-zero initial tangential stiffness. Assuming the network, based on input information, particularly the seismic eigenvector, predicts that the initial stiffness of the soil at a depth of 5 m is affected under this condition, the nodal tangential stiffness is calculated. for These two values, and This constitutes the output of the constitutive prediction step for node 5. The system repeats this process for all the remaining 30 nodes, ultimately generating two 31-dimensional vectors: one for the node soil resistance and the other for the node soil tangential stiffness. These two vectors serve as outputs, providing all the necessary information for the next step of global matrix assembly.

[0069] In one embodiment of the present invention, the global matrix assembly module is configured to perform the following steps: The soil resistance at the nodes is integrated along the discrete grid to obtain the comprehensive internal force vector; Retrieve the structural stiffness matrix; The tangential stiffness of the soil at each node is assigned to the position corresponding to the lateral displacement degree of freedom in the diagonal matrix according to the pile foundation connectivity topology, and then added to the structural stiffness matrix to assemble a global tangential stiffness matrix.

[0070] The system's physical solver core module converts the discrete constitutive information at each node output by the neural network into a global matrix component required for macroscopic mechanical analysis. First, the module receives the constitutive relation predictions from the system memory for the neural network at the current iteration step. For the full depth of the pile The soil resistance of all nodes returned by the discrete node and the tangential stiffness of the soil at the node The module obtains both dimensions as... The vectors represent the soil resistance density and tangential stiffness density distributed along the pile, respectively.

[0071] Next, based on the principle of virtual work, the module generates the comprehensive internal force vector for the current iteration step by integrating and accumulating the soil resistance at the nodes along the pile length. In the discretized numerical model, this integration operation is converted into a weighted summation, calculated using the following formula: in It is a dimension of Boolean mapping matrix, which only corresponds to the first... The position element for each node's lateral translational degree of freedom is 1, and all others are 0. This operation is equivalent to distributing the soil resistance as concentrated forces to the corresponding translational degrees of freedom, without generating direct constraint moments on the rotational degrees of freedom; This is the length of the discrete element. This calculation generates the combined internal force vector for the current iteration step. Each component of this vector represents the soil deformation vector of the current full pile. The total resisting force or moment generated by the corresponding degree of freedom of the pile body.

[0072] Simultaneously, the module utilizes the nodal soil tangent stiffness to assemble a global tangent stiffness matrix compatible with the current displacement state. This matrix consists of two parts: the structural stiffness matrix of the pile itself. and the stiffness matrix contributed by the soil The module calculates the nodal soil tangential stiffness for each node. Based on the interconnected topology of the pile foundations, the soil stiffness matrix is ​​mapped and assembled. Then, it is compared with the pre-calculated structural stiffness matrix. Add them together to form the global tangent stiffness matrix. Its assembly process can be represented as follows: The soil stiffness matrix It is a dimension of A sparse matrix, whose non-zero elements are located only on the main diagonal at positions corresponding to the lateral displacement degrees of freedom of each node. Assuming that the row and column indices of the matrix both start from 1, and that the lateral displacement and rotation degrees of freedom of the nodes alternate, then for a depth index of... ( For nodes of type ), the filling rule for their corresponding non-zero diagonal elements is as follows: matrix All other elements in the set are zero.

[0073] Finally, the calculated integrated internal force vector and global tangent stiffness matrix This output is used as the result of this step for calculating unbalanced forces and correcting displacements.

[0074] Combined internal force vector It is a column vector with the same dimension as the total number of degrees of freedom of the system, for example This reflects the projection of the resultant force of the resistance generated by all nonlinear soil springs onto each global degree of freedom under the current deformation state. Global tangent stiffness matrix. It is a dimension of The square matrix is ​​the Jacobian matrix of the total internal forces of the system with respect to the total displacement, which is used in the Newton-Raphson iterative method to solve for the displacement correction.

[0075] The application of the principle of virtual work ensures the equivalence of energy contributions of forces at discrete nodes when converted into global internal force vectors. The interconnected topology of the pile foundation defines the connection method between pile-beam elements, which determines the stiffness matrix of the pile structure. The distribution pattern of non-zero elements is usually a banded matrix.

[0076] In a specific example, during the 0th iteration, the system receives the nodal soil resistance generated for all 31 nodes of the full pile. and the tangential stiffness of the soil at the node Due to the influence of the foundation movement profile in the initial state, some nodes generate non-zero initial nodal soil resistance, such as those mentioned above. For the tangential stiffness of the soil at the pile joint, its value varies linearly from the pile top to the pile bottom, for example, at the pile top joint. Node No. 5 And the 30th node at the bottom of the pile The distribution of tangential stiffness of the soil at the node with depth is as follows: Figure 5 As shown. The discrete node spacing is known. The system first calculates the combined internal force vector. According to the integral formula, the system will convert non-zero values ​​at each depth. The final product is generated by integrating and accumulating through the mapping matrix. This is a non-zero internal force vector containing the initial soil resistance at each node. Next, the system constructs a global tangent stiffness matrix. It first retrieves a pre-generated pile structure stiffness matrix with a size of 62×62. Then, it constructs the soil stiffness matrix based on the nodal soil tangent stiffness vector. This matrix is ​​a 62×62 diagonal matrix, with its diagonal elements filled with elements corresponding to the lateral displacement degrees of freedom of the nodes. For example, the diagonal element corresponding to the displacement of node 0 is... The diagonal element corresponding to the displacement of node 5 is: The diagonal element corresponding to the displacement of node 30 is: All other elements, including the diagonal elements corresponding to the degrees of freedom of rotation, are zero.

[0077] Finally, the system adds the two matrices together. This yields a symmetric banded matrix. This non-zero combined internal force vector... and the assembled global tangent stiffness matrix Both are outputs of this step.

[0078] In one embodiment of the present invention, the state update solution module is used to perform the following operations: Calculate the difference between the external load vector and the combined internal force vector to generate the unbalanced force vector; A system of equations is constructed based on the global tangent stiffness matrix and the unbalanced force vector, and the displacement correction is obtained by solving the equations. The displacement correction is added to the current full pile deformation vector to generate the updated full pile deformation vector.

[0079] The system utilizes the generated global mechanics matrix component to calculate the force imbalance differences in the macroscopic system and solves for displacement corrections to update the global state. This step is the core of the Newton-Raphson iterative solution and is executed by the physical solver core module.

[0080] The module first obtains the predefined external load vector. and obtain the current iteration step. Comprehensive internal force vector The module calculates the vector difference between the two vectors and sets the result as the unbalanced force vector for the current iteration step. The calculation formula is as follows: This vector The physical meaning of is the imbalance between external and internal forces acting on the system under the current deformation state. Its existence indicates that the system has not yet reached static equilibrium.

[0081] Next, the modules establish this unbalanced force vector. With the current step global tangent stiffness matrix A large set of nonlinear evolution equations is constructed. The equations are in the following form: The solver calls upon a built-in, efficient linear equation solver, such as one based on LU decomposition or the conjugate gradient method, to solve the equation and calculate the displacement correction used to compensate for the current unbalanced force deviation. This vector Each component represents the amount of adjustment required for the corresponding degree of freedom in order to bring the system closer to equilibrium.

[0082] Finally, the module will calculate the displacement correction amount. The total pile deformation vector superimposed on the physical system at the current iteration step This generates an updated full pile deformation vector for the next physical prediction. Its update rules are as follows: in This is the step size factor. Through this operation, the deformed state of the system is updated, completing the iteration from step [number missing]. to the iteration step The state evolution. This updated full pile deformation vector. This will serve as the basis for the next convergence determination, and if convergence is not determined, it will be used as the input for the next iteration.

[0083] Unbalanced force vector The unit is consistent with the unit of force or torque, for example... or The norm of the equation directly reflects how close the current solution is to the true equilibrium solution. The nonlinear evolution equations are the standard form of the Newton-Raphson method in multi-degree-of-freedom nonlinear systems. Displacement correction. The unit is consistent with the unit of displacement or rotation, such as m or rad. Step factor The range of values ​​is within A scalar value between these two values ​​is used to control the size of the update step to improve the stability and convergence of the iteration process. In most cases, especially within the neighborhood of the solution, this value can be directly taken. Full-step updates are performed; however, when the nonlinearity is strong or the initial guess is poor, an optimal step size factor less than 1 may be selected through methods such as linear search to avoid iterative divergence. The updated full pile deformation vector. It is the core variable in the iterative process, carrying complete path information of the system's evolution from the initial state to the final equilibrium state.

[0084] In a specific example, in the 0th iteration, the system retrieves the external load vector. Since this vector only contains the 2450kN concentrated inertial force at the pile top, all its elements in the 62 degrees of freedom are 0 except for the first element, which is 2450. The system obtained the comprehensive internal force vector. This is a non-zero vector reflecting the initial pile-soil relative deformation resistance. According to the formula... The calculated unbalanced force vector By integrating the combined effects of external inertial load input and internal initial soil motion resistance, the stress imbalance state of the macroscopic system at this point is accurately characterized. Then, the system is coupled... and global tangent stiffness matrix Construct a system of linear equations The displacement correction can be obtained by matrix inversion or iterative solution. .

[0085] The vector obtained by solving The first few components are This indicates that the pile top node needs to generate a lateral displacement of 0.015m and a rotation angle of -0.0002rad. The system uses full-step updates, i.e. This correction is then added to the initial displacement. Due to the total pile deformation vector... It is a zero vector, so the updated total pile deformation vector For example, the updated pile top displacement is 0.015 m, and the rotation angle is -0.0002 rad. This updated full pile deformation vector contains 62 non-zero components. This result is determined as the final result of this iteration and convergence is checked. If it does not converge, it will be used in the next iteration. The basis for the force on the pile body under the condition.

[0086] In one embodiment of the present invention, the convergence determination and output module is used to perform the following steps: Extract the components corresponding to the lateral translational degrees of freedom from the unbalanced force vector to form a translational subvector, and calculate the norm of the translational subvector; If the norm is greater than the preset tolerance threshold, the updated full pile deformation vector will be used as the input for the next iteration step, and the process will return to the step of extracting nodal displacement components. If the norm is less than or equal to the allowable error threshold, extract the current full pile deformation vector, and calculate the bending moment and shear force distribution based on the finite difference method to generate the pile internal force envelope.

[0087] The system determines the convergence of the iterative process and triggers corresponding logical branches based on the determination result until the internal force envelope of the pile body that satisfies physical and mechanical equilibrium is generated. In each iteration step... At the end, the solution module first extracts the unbalanced force vector calculated in the previous step. To avoid the loss of physical meaning caused by directly adding forces and torques with different dimensions, the system starts from the unbalanced force vector. In this process, force components corresponding to the lateral translational degrees of freedom of each node are specifically extracted, forming a dimension of Translational unbalanced force vector To quantify the magnitude of this subvector into a single metric, the system calculates its Euclidean norm (L2 norm), which is calculated using the following formula: in It is the vector of translational unbalanced force. The This norm eliminates the dimensional conflict caused by the mixing of force and bending moment, and its value... Having the dimension of force, such as , which serves as a rigorous convergence metric for the current iteration step.

[0088] The system reads the preset tolerance threshold from its internal configuration or designated storage medium. This threshold represents the maximum acceptable unbalanced force in engineering, and its magnitude determines the accuracy of the calculation results. The system will calculate the unbalance norm. With tolerance threshold Compare them. If an imbalance norm is determined... Greater than the allowable error threshold This means that the system has not yet reached mechanical equilibrium.

[0089] In this situation, the system will trigger a loop logic, updating the generated full pile deformation vector. Deferred to the beginning of the next iteration, and automatically triggered to reuse the new deformed state. The process involves querying the neural network, reconstructing the constitutive relation, assembling the matrix, and solving for the new displacement correction.

[0090] Conversely, if determining the unbalanced norm Less than or equal to the tolerance error threshold This indicates that the current unbalanced force in the system is within acceptable limits, and it can be considered to have reached static equilibrium. At this point, the system will truncate the iterative loop mechanism and stop the calculation. The system locks the current full pile deformation vector. or This serves as the reference displacement solution vector under this load condition.

[0091] Based on this final displacement solution, the system further extracts and calculates various response indicators of the pile body. For example, it calculates the bending moment and shear force distribution along the pile body by differentiating the displacement solutions, and finally outputs these calculation results to form the internal force envelope of the pile body that characterizes the most dangerous stress state of the pile foundation under a specific seismic load.

[0092] Among them, the imbalance norm It is a non-negative scalar, and its unit is consistent with the unit of force, such as kN. Tolerance threshold. This is a small positive number set according to the engineering accuracy requirements. Its setting is usually a percentage relative to the total effect of the external load, for example, 0.1% of the total external force. A typical value is between 1.0 kN and 10.0 kN, depending on the specific requirements of the project. The reference displacement solution vector is the converged total pile deformation vector. It includes the final lateral displacement and rotation of all discrete nodes of the pile body.

[0093] The internal force envelope of a pile is a set of curves, typically including bending moment-depth and shear force-depth curves, illustrating the distribution of internal stresses within the pile under specific loads. It serves as a direct basis for structural engineers to verify the safety of pile foundations and design reinforcement. To avoid truncation errors caused by higher-order differences, bending moment... The calculation directly utilizes the rotational degrees of freedom in the displacement solution vector. Perform a first-order difference solution, for example shear force Then, based on the bending moment distribution, further differentiation is performed to obtain the result.

[0094] In a specific example, in the first After the iteration, the system calculates the unbalanced force vector. The system extracts the translational components and calculates the unbalance norm as follows: The system reads the preset tolerance threshold from the configuration file. The system determines that since 5.6 kN is greater than 1.0 kN, the convergence is not achieved. Therefore, the system performs a cyclic deferred operation, updating the current full pile deformation vector. As the first The input of the nth iteration is returned, and execution continues. Assume that after several iterations, at the nth iteration... After the iteration, the unbalanced force vector calculated by the system Its L2 norm value The system makes another determination: since 0.8 kN ≤ 1.0 kN, the determination result is convergence.

[0095] At this point, the system truncates the iteration loop. It then calculates the final total pile deformation vector. The baseline displacement solution vector is locked. For example, this vector records a series of precise deformation data, such as the final displacement at the pile top being 0.112 m and the displacement at a depth of 15 m in the pile body being 0.058 m. Finally, the system's post-processing module calls this baseline displacement solution vector and uses the internal force calculation formula of the beam element to calculate the bending moment and shear force at each section of the pile body. For example, the calculated bending moment at the pile top is 2800 kN·m, the maximum positive bending moment occurs at 8 m underground at 1500 kN·m, and the maximum negative bending moment occurs at 18 m underground at -2100 kN·m. Finally, the internal force envelope of the pile body is generated based on the above calculation results. (See the generated internal force (bending moment) envelope for details.) Figure 6 As shown, the final analysis results are output to the interactive interface or report file.

[0096] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0097] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A smart assessment system for the seismic safety of pile foundations coupled with inertial-motion loads, characterized in that, include: The equivalent feature generation module is used to acquire the initial seismic wave signal, extract the frequency response parameters, and combine them with the peak ground acceleration to generate the time-frequency feature vector of the ground motion; The initial state construction module is used to construct a discrete grid based on the pile foundation design parameters, apply the peak inertial load to the top node to form an external load vector, calculate and generate a pure motion displacement profile, and assign zero displacement values ​​to the grid nodes to generate the initial deformation vector of the entire pile. The input vector generation module is used to extract the nodal displacement components of the discrete grid and combine them into a total pile deformation vector. The pile-soil relative displacement components are calculated by subtracting the nodal displacement components from the components at the depth corresponding to the pure motion displacement profile. The pile-soil relative displacement components, the total pile deformation vector, and the ground motion time-frequency characteristic vector are concatenated to generate a nodal state vector. The constitutive parameter inference module is used to input the node state vectors at each depth into the constitutive relation prediction neural network and output the node soil resistance and node soil tangential stiffness at the corresponding depth. The global matrix assembly module is used to integrate and accumulate the soil resistance of all nodes to obtain the comprehensive internal force vector, and assemble the global tangent stiffness matrix based on the tangent stiffness of the soil at each node. The state update solution module is used to calculate the difference between the external load vector and the comprehensive internal force vector to generate the unbalanced force vector. It solves the displacement correction amount based on the global tangent stiffness matrix and the unbalanced force vector, and adds the displacement correction amount to the current full pile deformation vector to generate the updated full pile deformation vector. The convergence determination and output module is used to extract the components corresponding to the lateral translational degrees of freedom in the unbalanced force vector and calculate their norms. If the norm is greater than the preset allowable error threshold, the updated full pile deformation vector is used as the input for the next iteration step, and the step of extracting nodal displacement components is returned. If the norm is less than or equal to the allowable error threshold, the iteration stops, and the pile internal force envelope is calculated based on the current full pile deformation vector.

2. The intelligent assessment system for seismic safety of pile foundations based on inertial-motion load coupling according to claim 1, characterized in that, Acquiring the initial seismic wave signal includes: Obtain the environmental white noise variance threshold and the continuous signal collected at the bedrock surface; Configure an adaptive cutoff frequency for the Wiener filter that matches the variance threshold of the ambient white noise. A Wiener filter was used to filter out baseline drift and high-frequency noise in the continuous signal based on an adaptive cutoff frequency, and the filtered time series was used as the initial seismic wave signal.

3. The intelligent assessment system for seismic safety of pile foundations based on inertial-motion load coupling according to claim 1, characterized in that, Frequency response parameters are extracted and combined with peak ground acceleration to generate a time-frequency feature vector of ground motion, including: The initial seismic wave signal is decomposed based on wavelet packet transform; Extract the coefficient values ​​of each preset frequency band after decomposition, sum them together with the time step, and obtain the frequency response parameters; The maximum absolute value of the initial seismic wave signal is extracted as the peak ground acceleration; Peak ground acceleration is concatenated with frequency response parameters to generate a time-frequency feature vector of ground motion.

4. The intelligent assessment system for seismic safety of pile foundations based on inertial-motion load coupling according to claim 1, characterized in that, Generating the external load vector includes the following steps: Read the pile length, cross-sectional diameter and elastic modulus from the pile foundation design parameters, and discretize the pile body into multiple foundation beam elements to generate a discrete mesh; The peak inertial load of the superstructure was determined using the limit equilibrium method, and the pure motion displacement profile was calculated. The peak inertial load is applied to the translational degrees of freedom of the top node of the discrete mesh to generate the external load vector.

5. The intelligent assessment system for seismic safety of pile foundations based on inertial-motion load coupling according to claim 1, characterized in that, Generating node state vectors includes the following steps: Extract the lateral displacement components of the discrete grid nodes in the current iteration step as the node displacement components; The nodal displacement components of all depth nodes are spliced ​​together in depth order to obtain the total pile deformation vector; The component corresponding to the depth in the pure motion displacement profile is extracted and subtracted from the nodal displacement component to obtain the pile-soil relative displacement component. The relative displacement components of the pile and soil, the deformation vector of the entire pile, and the time-frequency characteristic vector of the ground motion are concatenated and standardized to generate the node state vector.

6. The intelligent assessment system for seismic safety of pile foundations based on inertial-motion load coupling according to claim 1, characterized in that, Output the corresponding nodal soil resistance and nodal soil tangential stiffness, including the following steps: The node state vectors are imported into the constitutive relation prediction neural network for feature encoding; The spatial coupling characteristics of the deformation vector of the whole pile are extracted using the spatiotemporal attention mechanism module; The displacement space coupling features are transformed into weighted feature vectors by combining the weight matrix; By analyzing the weighted eigenvectors through a nonlinear activation layer, the corresponding nodal soil resistance and nodal soil tangential stiffness are output.

7. The intelligent assessment system for seismic safety of pile foundations based on inertial-motion load coupling according to claim 6, characterized in that, The displacement spatial coupling features of the whole pile deformation vector are extracted using a spatiotemporal attention mechanism module, including the following steps: The deformation vector of the entire pile is reshaped into a spatial sequence feature matrix, and then mapped into a query matrix, a key matrix, and a value matrix, respectively. Perform a dot product operation between the query matrix and the key matrix to obtain the initial similarity score matrix; The attention weights are obtained by normalizing the initial similarity score matrix. The attention weights are weighted and summed with the value matrix to generate displacement space coupling features.

8. The intelligent assessment system for seismic safety of pile foundations based on inertial-motion load coupling according to claim 1, characterized in that, The process of generating the global tangent stiffness matrix includes the following steps: The soil resistance at the nodes is integrated along the discrete grid to obtain the comprehensive internal force vector; Retrieve the structural stiffness matrix; The tangential stiffness of the soil at each node is assigned to the position corresponding to the lateral displacement degree of freedom in the diagonal matrix according to the pile foundation connectivity topology, and then added to the structural stiffness matrix to assemble a global tangential stiffness matrix.

9. The intelligent assessment system for seismic safety of pile foundations based on inertial-motion load coupling according to claim 1, characterized in that, Generating the updated full pile deformation vector includes the following steps: Calculate the difference between the external load vector and the combined internal force vector to generate the unbalanced force vector; A system of equations is constructed based on the global tangent stiffness matrix and the unbalanced force vector, and the displacement correction is obtained by solving the equations. The displacement correction is added to the current full pile deformation vector to generate the updated full pile deformation vector.

10. The intelligent assessment system for seismic safety of pile foundations based on inertial-motion load coupling according to claim 1, characterized in that, Calculating and generating the envelope of internal forces in the pile includes the following steps: Extract the components corresponding to the lateral translational degrees of freedom from the unbalanced force vector to form a translational subvector, and calculate the norm of the translational subvector; If the norm is greater than the preset tolerance threshold, the updated full pile deformation vector will be used as the input for the next iteration step, and the process will return to the step of extracting nodal displacement components. If the norm is less than or equal to the allowable error threshold, extract the current full pile deformation vector, and calculate the bending moment and shear force distribution based on the finite difference method to generate the pile internal force envelope.

Images