Artificial intelligence-based structural seismic performance evaluation method

Abstract

This invention belongs to the field of data processing and civil engineering technology, specifically disclosing an artificial intelligence-based method for assessing the seismic performance of structures. The method includes: acquiring multi-source heterogeneous response data through a sensor array and processing it to construct an operational state feature space; constructing a damage evolution model with nonlinear path-dependent characteristics and extracting irreversible damage evolution features; establishing a simulated neural synapse self-healing model, transforming physical damage features into changes in the connection strength of neural synapses; using an analog amplifier algorithm to search for the lowest energy point in the global parameter domain, identifying the adaptive optimization path and vitality core region in the post-earthquake stress redistribution process of the structure; comprehensively evaluating the remaining bearing capacity and recovery potential of the building structure and outputting an assessment report. This invention, by simulating the plasticity of the biological nervous system, achieves accurate capture of the structural damage evolution process and scientific prediction of post-earthquake recovery capabilities, improving the resilience and safety of structures in the face of earthquake disasters.

Description

Technical Field

[0001] This invention belongs to the field of data processing and civil engineering technology, and specifically relates to an artificial intelligence-based method for evaluating the seismic performance of structures. Background Technology

[0002] With the continuous advancement of disaster prevention and mitigation technologies, structural seismic performance assessment plays a crucial role in ensuring the safety of critical infrastructure and enhancing urban resilience. Artificial intelligence (AI) technology, with its powerful nonlinear fitting capabilities, has been widely applied to the prediction of building structure response and risk assessment under extreme seismic loads. Among these applications, the performance degradation analysis of complex structures under seismic sequences is a key research focus. To ensure the reliability of the assessment results, the assessment system needs to analyze the dynamic evolution of the structure under complex stress environments in real time. This places high demands on the algorithm's logical reasoning accuracy, data processing depth, and modeling ability for the evolution of physical damage.

[0003] Traditional AI assessment models typically employ static numerical attenuation logic when dealing with structural damage, making it difficult to accurately characterize the nonlinear path dependence and irreversible damage accumulation process caused by repeated aftershocks. Existing technologies lack effective modeling of a structure's post-earthquake adaptive optimization potential and cannot predict in real-time the dynamics of structural performance recovery and self-healing effects after stress redistribution. Furthermore, conventional assessment models are limited by fixed linear feature correlations, failing to capture the cross-effects between structural damage evolution and biological plasticity characteristics. This leads to biases in identifying the force paths of structures under extreme conditions, hindering in-depth deconstruction and ensemble optimization of structural vitality. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence-based method for evaluating the seismic performance of structures, which can effectively solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by this invention is: a method for evaluating the seismic performance of structures based on artificial intelligence, comprising the following specific steps:

[0006] Step 1: Acquire multi-source heterogeneous response data by deploying sensor arrays at key nodes of the building structure. The multi-source heterogeneous response data includes acceleration, displacement, tilt angle and strain information. Perform clock synchronization and noise removal processing on the multi-source heterogeneous response data to construct a high-dimensional structural operation state feature space.

[0007] Step 2: Construct a damage evolution model with nonlinear path dependence characteristics, map the high-dimensional structural operating state feature space to the damage evolution path, and extract irreversible damage evolution characteristics of the structure at different stress stages by simulating the energy dissipation and cumulative damage process under continuous seismic excitation.

[0008] Step 3: Establish a simulated neural synapse self-healing model based on the hybrid modeling logic of biological neural systems. The extracted irreversible damage evolution features are used as neuronal input excitations. By simulating the dynamic adjustment mechanism of synaptic weights in biological neural systems, the degree of physical damage to the structure is transformed into changes in the connection strength of neural synapses.

[0009] Step 4: Use the analog amplifier algorithm to search for the lowest energy point of the system potential energy surface in the global parameter domain, and combine it with the plasticity evolution rules in the simulated neural synapse self-healing model to identify the adaptive optimization path and core area of ​​structural vitality during the post-earthquake stress redistribution process.

[0010] Step 5: Based on the identified adaptive optimization path and structural vitality core area, comprehensively evaluate the remaining bearing capacity and dynamic recovery potential of the building structure, and output a seismic performance evaluation report that includes structural force path reconstruction scheme and ensemble optimization reinforcement recommendations.

[0011] Preferably, the process of acquiring multi-source heterogeneous response data in step 1 specifically involves arranging a predetermined number of sensor modules in the foundation, intermediate, and top layers of the building structure. The sensor modules are configured to convert the sensed physical vibration signals into discrete digital sequences using a high-precision analog-to-digital converter. During clock synchronization, the system uses a preset time reference signal to align the sampling frequencies of each sensor module, ensuring spatial coupling of features acquired at the same timestamp. For noise removal, this scheme employs adaptive filtering logic based on empirical mode decomposition to separate high-frequency environmental noise from low-frequency trend terms in the original signal, retaining effective frequency components reflecting the structure's own dynamic characteristics, thereby forming a stable structural operating state feature space.

[0012] Preferably, the damage evolution model with nonlinear path-dependent characteristics constructed in step 2 is based on the introduction of a memory mechanism for historical load effects. When dealing with damage accumulation caused by repeated aftershocks, the damage evolution model no longer uses simple linear stiffness reduction logic, but instead establishes an evolution function based on hysteretic energy dissipation. This function calculates the cumulative plastic energy dissipation after the structure enters the nonlinear stage by tracking the stress and deformation relationship of the structure throughout the earthquake process. The damage increment at each time step depends on the current stress state and the damage state of the previous stage, accurately characterizing the irreversibility and path dependence of the damage. In the specific feature extraction process, the model identifies the dominant damage mode of the structure during the stress process by calculating the changes in the eigenvalues ​​of the structural stiffness matrix.

[0013] Preferably, the establishment of the simulated neural synaptic self-healing model in step 3 deeply integrates the material damage theory in structural engineering with the synaptic plasticity theory in biological neuroscience. In this process, local damage points in the structure are abstracted as neuronal nodes in a neural network, and the mechanical transmission logic between components is transformed into synaptic connections between neurons. When irreversible damage occurs to the structure, the strength of the corresponding neural synaptic connections decays according to a preset inhibition function; conversely, when the structure exhibits a tendency for stress redistribution or has potentially redundant stress paths, the system triggers a simulated long-range enhancement effect, increasing the weight of synaptic connections on the corresponding paths. This hybrid modeling approach enables the evaluation model to simulate the self-organizing response process of the structure after being subjected to an impact.

[0014] Preferably, step 4 utilizes an analog amplifier algorithm to search for the lowest energy point. The logic behind this is to transform the complex nonlinear evaluation problem into a process of finding the globally optimal solution for the system's potential energy surface. During algorithm execution, the system constructs an equivalent energy function, which includes a first term reflecting the structure's elastic potential energy, a second term reflecting plastic dissipation energy, and a third term reflecting the synaptic adjustment and compensation energy. The analog amplifier algorithm iterates rapidly in the multidimensional solution space using the feedback suppression and gain adjustment principles of an analog operational amplifier. By monitoring the changing trend of the energy gradient, the algorithm can automatically avoid interference from local minima until it locates the globally lowest energy point representing the structure reaching steady-state equilibrium. At this point, the system parameters corresponding to this point represent the true stress state of the structure after the earthquake.

[0015] Preferably, the process of identifying the adaptive optimization path in step 4 involves in-depth analysis of the neural plasticity model. After the search for the lowest energy point is completed, the system analyzes the redistribution of synaptic connection weights. Those connection branches whose weights are enhanced or remain stable after damage are defined as the structurally most robust stress paths. This process not only assesses the degree of structural damage but, more importantly, discovers previously overlooked load-bearing potential within the structure by simulating the functional reorganization capabilities of biological systems in the face of damage. This simulation of a self-healing mechanism enables artificial intelligence to predict the dynamic evolution of a structure under subsequent aftershocks after the initial damage, achieving real-time prediction of the post-earthquake recovery effect.

[0016] Preferably, the process of comprehensively evaluating the remaining bearing capacity of the building structure in step 5 is based on the force path and energy steady-state characteristics identified above. The system calculates the ratio of the ultimate load of the structure in its current state to the initial design load to obtain the remaining bearing capacity coefficient. Simultaneously, by combining the recovery potential index output by the simulated neural synapse self-healing model, the system quantitatively classifies the difficulty of structural repair and the benefits of reinforcement. The recovery potential index reflects the possibility that the overall seismic performance of the structure can be improved through simple local reinforcement. This evaluation method shifts from a simple damage assessment to a deep deconstruction of the structural toughness and recoverability, providing a scientific basis for subsequent disaster prevention and mitigation decisions.

[0017] Preferably, the ensemble optimization strategy output in step 5 specifically includes component-level reinforcement schemes for weak structural links and system-level adjustment suggestions for the overall stress distribution. The ensemble optimization strategy determines the optimal reinforcement location and strength by comparing the differences between the initial stress path and the post-earthquake adaptive stress path. At the component level, the system indicates specific units that require increased material ductility or improved section stiffness; at the system level, the system suggests guiding energy flow to the path of maximum vitality by introducing damping devices or adjusting the structural layout, thereby achieving a synergistic improvement in the overall structural performance. This seismic performance assessment report presents all technical indicators in pure Chinese, avoiding the use of any uncertain evaluation terms to ensure the professionalism and guiding significance of the assessment results.

[0018] Preferably, the AI-based structural seismic performance assessment method further includes a pre-trained model training and parameter calibration stage in practical applications. In this stage, a pre-stored structural seismic damage database is used to calibrate the baseline parameters of the simulated neural synaptic self-healing model. By introducing a large amount of historical earthquake data, the AI ​​can learn the general laws governing damage evolution for different structural types under extreme loads. During calibration, the feedback gain coefficient in the analog amplifier algorithm is adjusted using backpropagation logic by comparing the damage morphology predicted by the model with the actual observed damage characteristics. When the prediction error decreases to within a preset threshold range, the model is considered to have the ability to accurately assess specific structural types.

[0019] Preferably, in the feature space construction of step 1, cross-correlation terms of structural dynamic characteristics are introduced to further enhance the expressive power of the data. Specifically, the system calculates the correlation matrix between acceleration signals at different nodes to characterize the overall coordinated vibration state of the structure. When local damage occurs in the structure, the phase difference and energy transfer function between nodes will shift significantly, and these shifts are added to the feature vector as key features. In this way, the evaluation model can capture microscopic damage initiation signals from macroscopic vibration morphological features, improving the sensitivity of the evaluation method in the early stages of structural damage.

[0020] Preferably, in the damage evolution model of step 2, a piecewise continuous analytical logic is employed to address the nonlinear degradation process of the material. When the strain amplitude is below a predetermined critical value, the material exhibits linear elastic behavior; once the strain amplitude exceeds the predetermined critical value, the model automatically switches to a plastic evolution mode based on the energy criterion. In this mode, the stiffness attenuation coefficient is defined as a function of the accumulated dissipated energy, and its variation follows a nonlinear power-law distribution. This refined approach ensures that the evaluation method can realistically reproduce the stiffness hardening or softening phenomenon of the structure under seismic wave action, providing an accurate physical basis for subsequent synaptic weight mapping.

[0021] Preferably, the simulated neural synapse self-healing algorithm also introduces a dynamic constraint mechanism when searching for the energy minimum point to simulate boundary constraints in a real physical environment. During the iterative search process, the range of change of all physical variables is strictly limited within a predetermined mechanically reasonable range. For example, the equivalent stiffness of a component cannot be adjusted to a negative value, and the displacement of a node cannot exceed the geometric deformation limit of the structure. These constraints are integrated into the control logic of the analog amplifier in the form of a penalty function. When the search path touches the boundary, the system automatically generates a reverse adjustment signal to guide the search process back to a feasible physical space, ensuring the authenticity and reliability of the evaluation results.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] 1. This invention solves the technical challenge of traditional artificial intelligence in assessing structural seismic performance by introducing a simulated neural synapse self-healing algorithm and a hybrid modeling logic based on biological neural networks, making it difficult to characterize nonlinear path dependence and irreversible damage accumulation. Compared with conventional static numerical attenuation models, this invention can deeply integrate the plasticity characteristics of biological neural systems, dynamically mapping physical damage to changes in the connectivity of neural synapses, thus achieving precise capture of the structural damage evolution process.

[0024] 2. By employing an analog amplifier algorithm to search for the global minimum energy point, the system can quickly pinpoint the equilibrium state of post-earthquake stress redistribution within a complex multi-dimensional parameter space. This not only assesses the degree of structural damage but also deeply identifies the structure's adaptive optimization potential and vital regions within the stress paths. This method breaks through the limitations of traditional assessments that focus solely on damage determination, enabling scientific prediction of structural self-healing effects and dynamic recovery capabilities.

[0025] 3. The ensemble optimization strategy output by this invention provides decision support with a system-level perspective for the repair and reinforcement of post-earthquake structures, improves the resilience and safety of structures in the face of extreme earthquake disasters, and provides reliable artificial intelligence support for the long-term service of major infrastructure. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0027] Figure 2 This is a schematic diagram of the core principle framework of the simulated neural synapse self-healing model based on bioneural hybrid modeling logic in this invention;

[0028] Figure 3 This is a flowchart illustrating the logical process of using an analog amplifier algorithm to search for the lowest energy point in the global parameter domain and identify an adaptive optimization path in this invention.

[0029] Figure 4 This is a data flow diagram illustrating how the present invention acquires multi-source heterogeneous response data through a sensor array and constructs a high-dimensional structural operating state feature space.

[0030] Figure 5 This is a logical flowchart of the process in this invention for comprehensively evaluating the remaining bearing capacity and recovery potential of a building structure based on force path reconstruction and energy steady-state characteristics. Detailed Implementation

[0031] Example 1: Please refer to the appendix Figure 1 To be continued Figure 5 To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0032] In the specific implementation process of the artificial intelligence-based structural seismic performance evaluation method provided by this invention, step 1 is first executed, which involves acquiring multi-source heterogeneous response data through sensor arrays deployed at key nodes of the building structure. In specific engineering practice of buildings, the deployment scheme of the sensor array follows the principles of full spatial coverage and densification of key nodes. The key nodes cover the foundation layer, intermediate layer, and top layer of the building structure. Specifically, in the foundation layer, sensor modules are installed at the column bases of load-bearing columns and the connection points of foundation beams to capture the initial vibration input from the foundation and the base shear force response of the structure; in the intermediate layer, sensor modules are arranged at the beam-column joints of each layer and the edge members of shear walls to monitor the inter-story drift angle and the strain distribution of the main components; in the top layer, sensor modules are mainly deployed at the geometric center of the structure and the vertices of the four corners to acquire the vertex displacement and torsional vibration signals of the structure.

[0033] The multi-source heterogeneous response data includes acceleration, displacement, tilt angle, and strain information. Acceleration information is acquired using a triaxial piezoelectric accelerometer, which can monitor the dynamic response of the structure in two orthogonal horizontal directions and the vertical direction in real time. Displacement information is acquired using a high-precision laser displacement meter or a draw-wire displacement sensor, with a sampling accuracy of 0.01 mm. Tilt angle information is acquired using a biaxial tilt sensor to assess the residual tilt of the structure after an earthquake. Strain information is acquired using resistance strain gauges or fiber optic grating sensors placed on the surface of the reinforcing steel or inside the concrete to characterize the stress level inside the component.

[0034] After acquiring the raw signal, this embodiment performs clock synchronization processing on the multi-source heterogeneous response data. The system uses a preset time reference signal to align the sampling frequencies of each sensor module. Specifically, the system adopts a master-slave synchronization architecture based on a high-precision temperature-controlled crystal oscillator. The master control unit periodically sends synchronization trigger pulses to each distributed acquisition node. After receiving the pulse, the acquisition node resets its internal counter and marks the acquired digital sequence according to the timestamp assigned by the master control unit. By calculating the physical delay of pulse transmission and performing software-level compensation, the data from sensors at different spatial locations at the same sampling instant have strict temporal consistency, ensuring the spatial coupling of features in each dimension during subsequent high-dimensional feature space construction.

[0035] To address environmental noise and electromagnetic interference introduced during the data acquisition process, this solution employs noise removal. An adaptive filtering logic based on empirical mode decomposition (EMD) is used. This logic decomposes the original non-stationary response signal into several intrinsic mode function (EMF) components with different characteristic scales. During the decomposition process, the system automatically identifies higher-order components representing high-frequency environmental noise and residual components representing low-frequency baseline drift. By setting an energy threshold criterion, the effective components containing the structure's own dynamic characteristics are reconstructed, while random disturbance components unrelated to the structural response are removed. This adaptive filtering process does not require pre-setting the filtering frequency band and can extract the most realistic structural operating state feature space based on the signal's own distribution characteristics.

[0036] After preprocessing, this embodiment proceeds to step 2, constructing a damage evolution model with nonlinear path-dependent characteristics. The core of this model lies in mapping the high-dimensional structural operating state feature space constructed in the preceding steps to the damage evolution path. Unlike traditional linear damage accumulation theory, this scheme introduces a memory mechanism for historical load effects. During the simulation of continuous seismic excitation or multiple aftershocks, the model establishes an evolution function based on hysteretic energy dissipation. Specifically, the system tracks the force and deformation response trajectory of the structure within each sampling time step and calculates the area enclosed by the hysteresis loop, which represents the plasticity consumed by the structure during the stress process.

[0037] The damage evolution model, when dealing with cumulative damage, defines the current damage state as the sum of the damage amount at the previous time step and the damage increment at the current time step. The damage increment depends not only on the current stress level but also on the cumulative energy dissipation experienced by the structure. By establishing a nonlinear mapping relationship between damage variables and cumulative energy dissipation, the model accurately characterizes the strength degradation and stiffness softening phenomena of structural materials during repeated loading. In the specific feature extraction stage, the model extracts the eigenvalue evolution sequence reflecting the overall stiffness degradation of the structure by performing eigenvalue decomposition on the instantaneous stiffness matrix. When an eigenvalue shows a step decrease, the system automatically identifies the corresponding irreversible damage evolution feature and marks it as a key turning point on the damage path.

[0038] To address the nonlinear degradation process of materials, this embodiment employs piecewise continuous analytical logic. When the strain amplitude fed back by the sensor is lower than the predetermined material proportional limit critical value, the damage model operates in the linear elastic submodule. At this time, the stiffness of the structure remains constant, and no permanent damage accumulation occurs. Once the strain amplitude exceeds the predetermined critical value, the system automatically switches to a plastic evolution mode based on the energy criterion. In this mode, the stiffness attenuation coefficient is defined as a nonlinear power-law function of the accumulated dissipated energy. As the dissipated energy increases, the stiffness attenuation coefficient evolves according to a preset exponential law until the material's failure limit is reached. This approach can realistically simulate the physical nature of a structure entering a nonlinear stage after experiencing a major earthquake.

[0039] Step 3 is then executed to establish a simulated neural synaptic self-healing model based on a hybrid bio-neural modeling logic. In this stage, the invention draws an analogy between macroscopic damage manifestations in structural engineering and microscopic synaptic mechanisms in biological neuroscience. Local damage points in a building structure are abstracted as neuronal nodes in a neural network, while the mechanical transmission logic between components, such as moment transmission between beams and columns or stress flow between shear walls, is transformed into the synaptic connection strength between neurons. When the irreversible damage evolution features extracted in step 2 show significant damage to a component, the corresponding synaptic connection strength will decay according to a preset inhibition function. This inhibition function simulates the degradation of synaptic function in a biological nervous system after damage.

[0040] The core innovation of this invention lies in simulating the self-healing or reorganization mechanisms of biological systems. When structural damage causes the failure of existing stress paths, the system monitors the trend of stress redistribution. If potential redundant stress paths exist within the structure, and these paths exhibit an increasing load-bearing capacity during stress redistribution, the system triggers a simulated long-range reinforcement effect. This long-range reinforcement effect increases the weight of synaptic connections on the corresponding paths, assigning higher mechanical weight coefficients to healthy or still-capable components in the model. Through this hybrid modeling approach, the evaluation model is no longer merely a record of damage, but a dynamic evolutionary system capable of simulating a structure's self-organizing response and search for survival paths after being struck.

[0041] Next, step 4 is executed, using an analog amplifier algorithm to search for the lowest energy point on the system's potential energy surface within the global parameter domain. This step transforms the complex nonlinear damage assessment problem into a process of finding the system's steady-state equilibrium solution. An equivalent energy function is constructed for the system, comprising a first term reflecting the structure's elastic potential energy, a second term reflecting plastic dissipation energy, and a third term reflecting the compensating energy from synaptic adjustments. The elastic potential energy term is positively correlated with the degree of structural deformation; the plastic dissipation energy term reflects the irreversible damage that has already occurred to the structure; and the compensating energy term reflects the change in the system's internal energy resulting from the weight adjustments in the aforementioned synaptic self-healing model.

[0042] The analog amplifier algorithm iterates within a multidimensional solution space comprised of thousands of structural parameters, utilizing the feedback suppression and gain adjustment principles of analog operational amplifiers. In each iteration, the algorithm calculates the gradient direction of the energy function and updates the system state variables along the direction of the fastest gradient descent. To avoid interference from local minima, a dynamic gain adjustment mechanism is introduced. When a slowdown in the energy descent rate is detected and the convergence criterion has not been met, the feedback gain coefficient is automatically increased, giving the search path greater momentum to overcome small fluctuations on the potential energy surface. This continues until the global energy minimum point, representing the structure reaching steady-state equilibrium after the earthquake, is located.

[0043] In the search for the minimum energy point, this embodiment introduces a dynamic constraint mechanism. The range of variation of all physical variables is strictly limited within a predetermined mechanically reasonable range. For example, the equivalent stiffness of the components is set to always be greater than or equal to 0, and the displacement of the nodes is limited within the geometric nonlinear deformation limit of the structure. These constraints are integrated into the control logic of the analog amplifier in the form of a penalty function. When the search path touches these physical boundaries, the penalty term generates a strong reverse adjustment signal, guiding the search process back to a feasible physical space, ensuring that the final searched minimum energy point has a true physical meaning.

[0044] After determining the point of lowest energy, the system proceeds to identify adaptive optimization paths and core areas of structural vitality. By analyzing the final distribution of connection weights in the simulated neural synapse model, the system identifies connection branches whose weights are enhanced or remain stable at a high level after experiencing earthquake impact. These branches are defined as the structurally most robust stress paths. This process transcends the traditional perspective of focusing solely on where things are damaged; more importantly, it discovers areas that can still bear load and how to utilize the remaining structure for balance. By uncovering hidden correlations in the neural plasticity model, artificial intelligence can identify redundant parts within the structure that were not originally primary load-bearing components but can play a crucial supporting role after an earthquake.

[0045] Finally, step 5 is executed to comprehensively evaluate the remaining bearing capacity and dynamic recovery potential of the building structure based on the identified adaptive optimization path and the core area of ​​structural vitality. The remaining bearing capacity is assessed by calculating the ratio of the ultimate lateral force of the structure in the current damaged state to the ultimate lateral force in the initial design state. After obtaining the remaining bearing capacity coefficient, the system further combines it with the recovery potential index for quantitative analysis. The recovery potential index reflects the ease with which the structure can recover to a safe service level through self-healing mechanisms (i.e., stress redistribution and local reinforcement).

[0046] Finally, the system outputs a seismic performance assessment report. This report includes a structural stress path reconstruction scheme and ensemble optimization reinforcement recommendations. The stress path reconstruction scheme details how post-earthquake stress transfers from severely damaged areas to the vital core areas, and specifies the new stress sequences requiring key maintenance in pure Chinese. The ensemble optimization strategy provides specific recommendations at both the component and system levels. At the component level, the report indicates which beams, columns, or walls need to have their ductility or stiffness improved by wrapping them with carbon fiber, increasing their cross-sectional dimensions, or replacing them with high-strength bolts. At the system level, the report recommends introducing damping devices, such as buckling-restrained braces or liquid viscous dampers, at specific adaptive optimization path nodes to guide more seismic energy to paths with high energy dissipation capacity, achieving a synergistic improvement in the overall structural performance.

[0047] In practical applications of this method, a model training and parameter calibration phase is also included. During this phase, a pre-stored large-scale structural earthquake damage database is used to calibrate the baseline parameters of the simulated neural synaptic self-healing model. By introducing a large amount of historical earthquake data, including examples of different structural systems, different fortification intensities, and different failure modes, artificial intelligence can learn the general laws of damage evolution. During calibration, the feedback gain coefficient in the analog amplifier algorithm is adjusted using backpropagation logic by comparing the model's predicted minimum energy point shape with the observed component failure characteristics in actual earthquake damage. When the error between the predicted damage distribution and the actual observed values ​​decreases, the model is considered to have accurate assessment capabilities.

[0048] Furthermore, in the feature space construction of step 1, to enhance the depth of data representation, the system calculates the correlation matrix between acceleration signals at different nodes. When the structure is in a healthy state, the vibration phases of each node exhibit strong regularity and correlation. Once local damage occurs, the phase difference between signals shifts, and the energy transfer function shows significant attenuation or peak shift in specific frequency bands. These shifts and changes are added as key features to the high-dimensional feature vector, significantly improving the sensitivity of the evaluation method to capturing early, minor damage to the structure.

[0049] Example 2: In Example 2, the present invention conducted a specific seismic performance evaluation exercise for a 30-story frame-shear wall structure located in a high-intensity seismic fortification zone.

[0050] First, step 1 was executed, deploying sensor substations on the 2nd basement level, 1st floor, 15th floor, 22nd floor, and the top of the 30th floor of the building. Each substation integrated a 3-axis accelerometer and a multi-channel strain data acquisition gateway. The acquisition system was set to a sampling frequency of 200 Hz. After a simulated strong seismic input, the system quickly acquired multi-source heterogeneous response data containing 512 channels. During the clock synchronization phase, the sampling time deviation of all acquisition points was controlled within 1 microsecond using the timing signal provided by the BeiDou Navigation Satellite System. For the complex signals generated by the whiplash effect on the top floor, the noise removal logic successfully separated the first three structural principal vibration modes with a frequency range of 0.2 Hz to 5 Hz using empirical mode decomposition, filtering out high-frequency mechanical vibration interference above 15 Hz generated by the rooftop air conditioning unit, thus constructing a clean structural operating state characteristic space.

[0051] Next, step 2 was performed, where the damage evolution model conducted an in-depth analysis of the critical shear walls from the 4th to the 8th floors of the building. Due to the significant cyclic deformation in this area during the earthquake, the model tracked the moment-curvature hysteresis relationship at the bottom section of the wall. Calculations revealed that the cumulative plastic energy dissipation of the left-side shear wall on the 6th floor had exceeded 40% of its design bearing capacity. Based on this energy dissipation data, the damage evolution function adjusted the stiffness attenuation coefficient of this area from 1.0 to 0.62, indicating that it had entered the nonlinear degradation stage. At this point, the extracted irreversible damage evolution features accurately located the concrete cracking and steel yielding trends of the shear wall edge members.

[0052] In step 3, the system constructed a corresponding simulated neural synapse self-healing model. The damaged shear wall on the 6th floor was mapped to a cluster of neurons with diminishing weights in a neural network. Simultaneously, the system detected an increase in stress levels in the frame columns adjacent to this wall and in the cross-floor bracing. The simulated neural synapse model automatically increased the synaptic connection weights between these adjacent components through a long-range enhancement effect. This weight transfer simulates the self-healing process of a building structure after damage to its main lateral force-resisting components, where mechanical loads are transferred through the floor slabs to the surrounding frame system, transforming the physical damage to the structure into a dynamic distribution of weights within the system.

[0053] Step 4 is then executed, where the analog amplifier algorithm searches for the lowest energy point within a space containing tens of thousands of component state parameters. The energy function constructed by the system uses the severe damage term of the 6th floor as the main plastic dissipation penalty term. Through 500 rapid iterations, the algorithm avoids local energy minima caused by local component instability, ultimately locking onto a steady-state point indicating that the structure as a whole is in gravitational equilibrium and can withstand aftershocks. At this point, the system identifies the central frame system located between the 10th and 12th floors as the core of vitality. Although these areas were not severely damaged in the initial earthquake, they bear more than 20% of the additional shear force in the post-earthquake stress redistribution, serving as the critical support path to prevent structural collapse.

[0054] Finally, step 5 was executed, and the system's comprehensive evaluation yielded a residual bearing capacity coefficient of 0.78 for the building. Based on the identification results of the vital core area, the generated seismic performance assessment report proposed an ensemble optimization strategy. Specifically, the recommendations were: high-pressure grouting repair and the addition of steel plate hoops to the damaged shear wall on the 6th floor to restore its energy dissipation capacity; and carbon fiber wrapping reinforcement for the core frame columns from the 10th to the 12th floors to improve their ductility under high loads. Furthermore, the report recommended adding two sets of active mass dampers to the 15th-floor refuge floor to actively suppress vibrations from subsequent aftershocks and protect the identified adaptive optimization path. This assessment scheme not only quantified the damage but also provided a scientific path for subsequent precise reinforcement.

[0055] Example 3: In Example 3, the present invention is applied to the seismic assessment of a gymnasium with a large-span spatial truss structure.

[0056] In step 1, the sensor array was primarily deployed at the truss support nodes, mid-span connectors, and spherical nodes. Because the spatial structure is highly sensitive to temperature changes and uneven settlement, the acquisition of tilt angle and temperature information was particularly emphasized in the multi-source heterogeneous data. For noise removal, the system employed a combined filtering logic based on wavelet packet decomposition and empirical mode decomposition, successfully eliminating the wind vibration noise caused by the large span while retaining the low-frequency large deformation characteristics reflecting the overall instability trend of the truss.

[0057] In step 2, for the compression buckling characteristics of truss members, the damage evolution model incorporates a nonlinear path tracing technique based on the arc-length method. When simulated seismic loads cause some compression members to approach the critical buckling load, the model no longer simply determines member failure. Instead, it extracts the irreversible damage characteristics of the structure during dynamic instability by calculating its post-buckling residual strength and path dependence features. This refined simulation ensures that the evaluation results reflect the complex geometric nonlinear behavior of large-span structures.

[0058] Step 3 involves simulating a neural synaptic self-healing model, treating each truss ball node as a neuron. In large-span structures, force transmission exhibits high multipath characteristics. When a critical tie rod is damaged, the system quickly identifies alternative force transmission loops by simulating the synaptic plasticity of the nervous system. For node connections with high redundancy, the system assigns higher self-healing scores in weight allocation, reflecting the robustness of the spatial structure after local damage.

[0059] In step 4, the analog amplifier algorithm specifically considers the geometric consistency constraints of the truss structure during the global search. When searching for the energy minimum point, the algorithm ensures that the coordinate updates of all nodes conform to the truss member length constraints. Through this physical boundary constraint, the adaptive optimization path identified by the system is not only mechanically steady-state but also geometrically feasible. The identified vital core region is located near the circumferential prestressed cables of the truss; the tension adjustment of these cables after structural damage plays a crucial role in maintaining overall stability.

[0060] Following step 5, the assessment report indicates that the residual bearing capacity coefficient of this large-span structure is 0.85, demonstrating high dynamic recovery potential. The optimized reinforcement recommendations focus on enhancing the rotational capacity of the support nodes and strengthening the sections of critical buckling members. Simultaneously, it is suggested that the stress path be reconstructed by adjusting the prestressed cable tension values ​​to guide the load to undamaged truss elements, achieving comprehensive seismic performance recovery at minimal cost.

[0061] Example 4: In Example 4, the present invention demonstrates detailed implementation details of the model training and parameter calibration phases.

[0062] In this phase, a database containing typical building earthquake damage cases from around the world over the past 50 years was first constructed. The data in the database is stored in the form of multidimensional tensors, including the geometric parameters of the structure, material properties, seismic waveform data, as well as actual damage images after the earthquake, reinforcement schemes, and final service status.

[0063] During the calibration process, the AI ​​system selected a concrete frame structure collapse instance as training input. Step 1 simulated and recreated the sensor data stream at that time. Step 2, the constructed damage evolution model initially predicted the failure location to be at the column head on the ground floor. The actual earthquake damage showed that the damage mainly occurred at the beam-column joints on the second floor. At this point, the system triggered the backpropagation logic. The system calculated the difference between the predicted damage distribution matrix and the actual observed damage distribution matrix, and converted this difference into an adjustment instruction for the feedback gain coefficient in the analog amplifier algorithm.

[0064] After approximately 10,000 weight iterations, the system found that introducing a correction coefficient reflecting the influence of the shear span ratio and integrating it into the inhibition function of the neural synaptic self-healing model could significantly improve prediction accuracy. When the calibrated model re-evaluated the same type of structure, its predicted damage morphology showed a high degree of consistency with the actual earthquake damage. This process demonstrates that the invention, by combining the simulation of biological neural mechanisms with learning from historical big data, can continuously evolve its evaluation logic.

[0065] Meanwhile, to address the convergence speed of the analog amplifier algorithm, an adaptive step-size strategy was introduced during the calibration phase. When the gradient modulus of the energy function is large, the algorithm uses a larger search step size to quickly approach the optimal value region; when entering the energy basin, the step size automatically decreases, and the minimum energy point is accurately located through fine-grained search. This strategy enables the comprehensive evaluation of seismic performance to be completed within minutes when dealing with ultra-large-scale complex structures, ensuring the engineering practicality of the method.

[0066] Example 5: In Example 5, the present invention provides specific implementation details for real-time monitoring and early warning of structures after earthquakes.

[0067] In the hours following the earthquake, the system continuously acquired high-frequency data using the sensor array deployed in step 1. At this point, the evaluation focus shifted to predicting the aftershock response. Using the simulated neural synaptic self-healing model established in step 3, the system was able to simulate the dynamic evolution behavior under aftershocks of different intensities based on the structural force path reconstruction observed during the main shock.

[0068] Specifically, the system simulated the displacement response of the damaged area of ​​the structure when a 5.5 magnitude aftershock occurred. The analog amplifier algorithm quickly searched for new energy minimum points and found that without temporary support, the residual tilt angle of the structure would increase by 15%, triggering the safety threshold. Based on this identification result, the evaluation report output in step 5 immediately provided a structural force path reconstruction scheme, recommending the installation of hydraulic temporary supports at the most severely damaged location on the third floor, and specifying that the support force should reach 1200 kN to balance the additional eccentric bending moment generated by the main shock.

[0069] Furthermore, the system's ensemble optimization strategy also includes a self-diagnostic logic for the sensors themselves. After strong vibrations, some sensors may malfunction or experience range drift. The system calculates the consistency of signals from adjacent sensors and uses spatial interpolation techniques to compensate for data at failed nodes, ensuring no data breaks occur in the construction of a multi-source heterogeneous feature space. This highly resilient acquisition and evaluation logic makes artificial intelligence an indispensable technical support in post-earthquake emergency rescue operations.

[0070] In the output interface of the assessment report, various technical indicators are displayed in a three-dimensional visualization. The damage level of the component is represented by color depth, and the redistribution of stress flow in the vital core area is represented by a streamline diagram. The aggregated optimization reinforcement recommendations are broken down into specific engineering lists, which clearly specify the numbering of the reinforced components, material specifications, and construction sequence, thereby improving the intuitiveness and guidance of decision-making.

[0071] The above embodiments, through different dimensions and application scenarios, comprehensively demonstrate the specific implementation path of the artificial intelligence-based structural seismic performance evaluation method of the present invention. Each step is interconnected, from the accurate acquisition of underlying data to the biological simulation of high-level logic, and then to the optimal evaluation of global energy, collectively forming a closed-loop, adaptive intelligent seismic performance evaluation system.

[0072] Example 6: In Example 6, this invention demonstrates its application in seismic assessment of historical buildings, particularly timber structures. Timber structures have a unique mortise and tenon joint mechanism, and their damage patterns differ significantly from those of concrete structures.

[0073] In step 1, the sensor array was carefully concealed and deployed at key connection points such as wooden beams, columns, and brackets. Strain acquisition employed non-contact digital image correlation technology, using high-speed cameras positioned opposite each other to capture the development of cracks on the wood surface. Ambient humidity and wood moisture content information were added to the multi-source heterogeneous response data, as the elastic modulus of wood is affected by moisture content.

[0074] Step 2 involves developing a damage evolution model specifically for slippage and dislodgement of mortise and tenon joints. Unlike damage in continuous media, damage to timber structures often manifests as tenon pull-out or cracking of joints. The model establishes a nonlinear path-dependent function based on displacement accumulation, accurately recording the relative displacement sequence between the tenon and mortise during earthquake vibrations. When the relative displacement exceeds 30% of the tenon length, the system marks it as severe irreversible damage.

[0075] Step 3 involves simulating a neural synaptic self-healing model, mapping the dougong (bracket set) system as a highly complex neural synaptic network. Dougong plays an excellent role in energy dissipation within the wooden structure, similar to a protective mechanism in the nervous system. When a dougong unit is damaged, the mechanical load is distributed among other dougong units. The simulation model identifies this redundancy and synergistic effect within the dougong system by adjusting synaptic weights.

[0076] In step 4, the analog amplifier algorithm considers the frictional energy dissipation of the wooden structure as a crucial component of the energy function when searching for the lowest energy point. By monitoring changes in the energy gradient, the system identifies the stable form of the ancient building after the earthquake. The core area of ​​vitality is identified as the building's central column and its connected lintel system, which constitute the structure's adaptive support framework.

[0077] Step 5 of the assessment report provides a non-destructive reinforcement solution for the ancient building. The ensemble optimization strategy recommends using high-performance fiber ropes to reinforce loose mortise and tenon joints and adding a flexible energy-dissipating pad at the base of the wooden columns without damaging the original appearance. This AI-based assessment method achieves a balance between cultural heritage protection and improved seismic safety.

[0078] Example 7: In Example 7, the present invention explores an integrated implementation scheme of the method in a smart city disaster prevention and mitigation system.

[0079] In this approach, the invention extends beyond individual buildings to encompass entire building complexes. The data sources for step 1 include not only sensor arrays from individual buildings but also data from urban strong-motion monitoring networks. Through a cloud computing platform, the system processes multi-source heterogeneous response data from hundreds or even thousands of buildings in parallel.

[0080] Steps 2 and 3 are performed on a distributed server array to create a digital twin and a corresponding simulated neural synapse self-healing model for each building. When the city experiences an earthquake, the system can extract irreversible damage features in batches based on the structural type and year of construction of different buildings.

[0081] Step 4, the analog amplifier algorithm, seeks the energy minimum point for resource allocation at the city scale. The system not only evaluates the stability of individual buildings but also identifies which critical infrastructures (such as hospitals and command centers) are within the coverage of the adaptive optimization path by analyzing the vitality distribution of building clusters.

[0082] The assessment report output in step 5 evolves into city-level disaster mitigation decision-making directives. The report not only provides individual reinforcement recommendations but also prioritizes reinforcement for the entire building complex, offering a scientific basis for the allocation of limited reinforcement funds and materials. Example 8: In Example 8, this invention demonstrates the details of adaptive optimization path identification under the complex dynamic characteristics of extremely tall buildings. For super high-rise buildings exceeding 500 meters in height, the influence of higher-order vibration modes on the structural response is significant due to their long natural periods.

[0083] In step 1, in addition to conventional sensors, a microelectromechanical system (MEMS) accelerometer was deployed to monitor extremely low-frequency vibrations. During clock synchronization, due to signal attenuation and phase deviation caused by long-distance cabling, the system employed a PTP protocol based on a fiber optic synchronization network, ensuring synchronization accuracy better than 10 nanoseconds over a cabling distance of several kilometers.

[0084] During step 4, when identifying the adaptive optimization path, the system discovered that the energy distribution of the outrigger truss system (strengthening layer) of the super high-rise building underwent drastic reconstruction after a strong earthquake. By monitoring subtle changes in the energy gradient, the analog amplifier algorithm found that although some outrigger components experienced local buckling, due to the long-range enhancement effect of the neural synapse model, the surrounding core tube wall formed a new vertically stressed core through stress reorganization.

[0085] The identified vital core region reveals that the junction between the reinforcing layer and the core tube is crucial for maintaining overall overturning stability. Based on this, the seismic performance assessment report output in step 5 proposes an innovative ensemble optimization strategy: introducing shape memory alloy dampers into the diagonal web members of the reinforcing layer. These dampers utilize the material's hyperelasticity to automatically generate restoring forces after a strong earthquake, assisting the structure in returning to its initial equilibrium state and mimicking the self-healing process of neural synapses, thereby enhancing the structure's dynamic recovery potential.

[0086] Through the detailed descriptions of the above embodiments, it is clear that this invention, through the deep integration of bio-neural hybrid modeling logic and artificial intelligence algorithms, provides an advanced technical solution with path adaptive recognition and vitality assessment capabilities for building structures of different types and scales. This method not only solves the problem of characterizing nonlinear damage accumulation at the theoretical level, but also provides highly operable assessment and reinforcement decision support at the engineering practice level.

[0087] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the seismic performance of structures based on artificial intelligence, characterized in that, Includes the following steps: Step 1: Acquire multi-source heterogeneous response data by deploying sensor arrays at key nodes of the building structure. The multi-source heterogeneous response data includes acceleration, displacement, tilt angle and strain information. Perform clock synchronization and noise removal processing on the multi-source heterogeneous response data to construct a high-dimensional structural operation state feature space. Step 2: Construct a damage evolution model with nonlinear path dependence characteristics, map the high-dimensional structural operating state feature space to the damage evolution path, and extract irreversible damage evolution characteristics of the structure at different stress stages by simulating the energy dissipation and cumulative damage process under continuous seismic excitation. Step 3: Establish a simulated neural synapse self-healing model based on the hybrid modeling logic of biological neural systems. The extracted irreversible damage evolution features are used as neuronal input excitations. By simulating the dynamic adjustment mechanism of synaptic weights in biological neural systems, the degree of physical damage to the structure is transformed into changes in the connection strength of neural synapses. Step 4: Use the analog amplifier algorithm to search for the lowest energy point of the system potential energy surface in the global parameter domain, and combine it with the plasticity evolution rules in the simulated neural synapse self-healing model to identify the adaptive optimization path and core area of ​​structural vitality during the post-earthquake stress redistribution process. Step 5: Based on the identified adaptive optimization path and structural vitality core area, comprehensively evaluate the remaining bearing capacity and dynamic recovery potential of the building structure, and output a seismic performance evaluation report that includes structural force path reconstruction scheme and ensemble optimization reinforcement recommendations.

2. The method for evaluating the seismic performance of structures based on artificial intelligence according to claim 1, characterized in that, The process of acquiring multi-source heterogeneous response data in step 1 specifically involves: arranging sensor modules in the foundation layer, intermediate layer and top layer of the building structure respectively; The sensor module converts the sensed physical vibration signal into a digital discrete sequence through an analog-to-digital converter. During clock synchronization, a preset time reference signal is used to align the sampling frequencies of each sensor module. The clock synchronization process adopts a master-slave synchronization architecture based on a temperature-controlled crystal oscillator. The master control unit periodically sends synchronization trigger pulses to the distributed acquisition nodes. After receiving the pulses, the acquisition nodes reset their internal counters and mark the acquired discrete digital sequences according to the timestamps allocated by the master control unit. Software compensation is performed by calculating the physical delay of pulse transmission to ensure that the data from sensors at different spatial locations have time consistency at the same sampling instant. For noise removal, an adaptive filtering logic based on empirical mode decomposition is used to decompose the original signal into intrinsic mode function components with different characteristic scales. The higher-order components representing high-frequency environmental noise and the residual components representing low-frequency baseline drift terms are identified. The effective frequency components reflecting the dynamic characteristics of the structure itself are extracted using the energy threshold criterion to form a stable structural operating state feature space.

3. The method for evaluating the seismic performance of structures based on artificial intelligence according to claim 1, characterized in that, The process of constructing a damage evolution model with nonlinear path-dependent characteristics in step 2 specifically involves: when dealing with the accumulation of damage caused by repeated aftershocks, establishing an evolution function based on hysteretic energy dissipation, and calculating the cumulative plastic energy dissipation of the building structure after it enters the nonlinear stage by tracking the stress and deformation relationship of the building structure throughout the earthquake process. The damage evolution model defines the current damage state as the sum of the damage amount at the previous time step and the damage increment at the current time step, wherein the damage increment depends on the current stress state and the cumulative energy consumption in the previous stage, thereby establishing a nonlinear mapping relationship between the damage variable and the cumulative energy consumption. To address the nonlinear degradation process of materials, a piecewise continuous analytical logic is adopted. When the strain amplitude is lower than the predetermined critical value, the damage evolution model operates under the linear elastic submodule, at which point the stiffness of the building structure remains constant. When the strain amplitude exceeds the predetermined critical value, the system switches to the plastic evolution mode based on the energy criterion. At this time, the stiffness attenuation coefficient is defined as a function of the accumulated dissipated energy. Its variation follows a nonlinear power law distribution. As the accumulated dissipated energy increases, the stiffness attenuation coefficient evolves according to a preset exponential law.

4. The method for evaluating the seismic performance of structures based on artificial intelligence according to claim 3, characterized in that, The process of extracting irreversible damage evolution features in step 2 specifically involves: performing eigenvalue decomposition on the instantaneous stiffness matrix of the building structure to extract the eigenvalue evolution sequence that reflects the overall stiffness degradation of the building structure; When a step drop occurs in the feature value evolution sequence, the corresponding irreversible damage evolution feature is automatically identified, and the mutation position of the feature value evolution sequence is marked as a key turning point on the damage path. The model identifies the dominant damage mode of a building structure during the stress process by calculating the changes in the eigenvalues ​​of the building structure stiffness matrix. In the construction of the feature space, cross-correlation terms of structural dynamic characteristics are also introduced. By calculating the correlation matrix between acceleration signals at different nodes, the overall coordinated vibration state of the building structure is characterized. When local damage to the building structure causes a shift in the phase difference and energy transfer function between nodes, the shift is added as a key feature to the feature vector to capture the minute damage signal in the early stage of the building structure damage.

5. The method for evaluating the seismic performance of structures based on artificial intelligence according to claim 1, characterized in that, The process of establishing a simulated neural synapse self-healing model based on bio-neural hybrid modeling logic in step 3 specifically involves: abstracting local damage points of building structures into neuron nodes in a neural network, and transforming the mechanical transmission logic between components into the synaptic connection strength between neurons. When irreversible damage evolution characteristics indicate that a component is damaged, the corresponding neural synaptic connection strength is attenuated according to a preset inhibition function; When a building structure exhibits a tendency for stress redistribution and has potentially redundant stress paths, a long-range enhancement effect is triggered in the simulation, increasing the weight of synaptic connections on the redundant stress paths. The simulated neural synapse self-healing model discovers the internal load-bearing potential of building structures by simulating the functional reorganization ability of biological systems when faced with damage, and predicts the dynamic evolution behavior of building structures under subsequent aftershocks after experiencing the first damage. The synaptic weight dynamic adjustment mechanism dynamically maps the physical damage to the building structure into changes in the connection of neural synapses, thereby capturing the evolution process of building structure damage.

6. The method for evaluating the seismic performance of structures based on artificial intelligence according to claim 1, characterized in that, The process of searching for the lowest energy point of the system potential energy surface in step 4 specifically involves: constructing an equivalent energy function, which includes a first term reflecting the elastic potential energy of the building structure, a second term reflecting the plastic dissipation energy, and a third term reflecting the adjustment and compensation energy of the neural synapse. Among them, the elastic potential energy term is positively correlated with the degree of deformation of the building structure, the plastic dissipation energy term reflects the irreversible damage that has occurred to the building structure, and the compensation energy term reflects the change in the system's internal energy caused by the weight adjustment in the simulated neural synapse self-healing model; The analog amplifier algorithm iterates in the multidimensional solution space based on the feedback suppression and gain adjustment principle of the analog operational amplifier. In each iteration, it calculates the gradient direction of the equivalent energy function and updates the system state variables along the gradient descent direction. The analog amplifier algorithm introduces a dynamic gain adjustment mechanism. When the energy decrease rate is detected to be slowing down and the convergence criterion is not met, the feedback gain coefficient is increased to give the search path momentum to cross the local minimum on the potential energy surface until the global energy minimum point representing the building structure reaching steady-state equilibrium is located.

7. The method for evaluating the seismic performance of structures based on artificial intelligence according to claim 6, characterized in that, Step 4 also involves introducing a dynamic constraint mechanism to simulate boundary limitations in a real physical environment: during the iterative search process, the range of change of all physical variables is limited to a predetermined mechanically reasonable range. The mechanically reasonable range includes the condition that the equivalent stiffness of the component is not negative and the displacement of the node does not exceed the geometric deformation limit of the building structure. The dynamic constraint mechanism is integrated into the control logic of the analog amplifier algorithm in the form of a penalty function. When the search path touches the physical boundary, a reverse adjustment signal is generated to guide the search process back to the feasible physical space. After the global minimum energy point is located, the system analyzes the redistribution of the weights of neural synaptic connections and defines the connection branches whose weights are enhanced or remain stable after damage as the maximum vitality force path of the structure. The process identifies redundant parts within a building structure that were not originally primary load-bearing components but now provide support after an earthquake, enabling a scientific prediction of the building structure's self-healing effect.

8. The method for evaluating the seismic performance of structures based on artificial intelligence according to claim 1, characterized in that, The process of comprehensively evaluating the remaining bearing capacity of the building structure in step 5 specifically involves: the system calculating the ratio of the ultimate load of the building structure in the current state to the initial design load to obtain the remaining bearing capacity coefficient; Based on the recovery potential index output by the simulated neural synapse self-healing model, the difficulty of repairing building structures and the effectiveness of reinforcement are quantitatively classified. The recovery potential index reflects the possibility that a building structure can improve its overall seismic performance through local reinforcement. The seismic performance assessment report outputs a set of optimized reinforcement recommendations, including: component-level reinforcement schemes for weak points in the structure and system-level adjustment recommendations for the overall stress distribution; The optimized reinforcement recommendations determine the reinforcement location and strength by comparing the differences between the initial force path and the post-earthquake adaptive force path. At the component level, specify the elements that require increased material ductility or improved cross-sectional stiffness; At the system level, it is recommended to adjust the structural layout by introducing damping devices to guide energy flow to the path of maximum vitality and achieve a synergistic improvement in the overall performance of the structure.

9. The method for evaluating the seismic performance of structures based on artificial intelligence according to claim 1, characterized in that, Before performing step 1, a model training and parameter calibration stage is also included: the baseline parameters of the simulated neural synapse self-healing model are calibrated using a pre-stored structural earthquake damage database. The structural earthquake damage database contains structural geometric parameters, material properties, earthquake waveform data, post-earthquake damage patterns, and reinforcement schemes from historical earthquake cases. During the calibration process, the difference between the predicted damage distribution matrix and the actual observed damage characteristics is calculated by comparing the damage morphology predicted by the model with the damage characteristics actually observed. The feedback gain coefficient in the analog amplifier algorithm is then adjusted using backpropagation logic. When the prediction error is reduced to within a preset threshold range, the model is determined to have the ability to evaluate the type of building structure. The calibration phase also introduces an adaptive step size strategy. When the gradient modulus of the energy function is large, a large search step size is used. When entering an energy basin, the step size is automatically reduced, and the global minimum energy point is located through fine-grained search.

10. The method for evaluating the seismic performance of structures based on artificial intelligence according to claim 1, characterized in that, The assessment method also includes the implementation process for real-time monitoring and early warning after the earthquake: within a predetermined time after the earthquake, the sensor array is used to continuously collect data, and the assessment focus shifts to the prediction of aftershock response. The simulated neural synapse self-healing model simulates the dynamic evolution behavior of building structures under aftershocks of different intensities based on the force path reconstruction of the building structure during the main earthquake. The system simulates the displacement response of the damaged area of ​​the building structure under subsequent aftershocks, and uses the analog amplifier algorithm to quickly search for new energy minimum points to identify whether the residual tilt angle of the building structure has reached the safety red line. Based on the identification results, a structural force path reconstruction scheme is output, specifying the physical location of temporary supports to be erected at the damaged parts and the required support force value, in order to balance the additional eccentric bending moment generated by the main shock. Meanwhile, the system calculates the consistency of signals from adjacent sensors and uses spatial interpolation technology to compensate for the data at the failure nodes, ensuring the continuity of data in the construction of a multi-source heterogeneous feature space. It also displays the damage level of the component and the redistribution of stress flow in the vital core area in a three-dimensional visualization.

Images