Modular self-resetting connection and method for global perception and early warning of structures

Abstract

This invention discloses a modular self-resetting connection and a method for global structural perception and early warning. Modular self-resetting energy-dissipating beam-column connection nodes are set up, and a graded energy dissipation mechanism is constructed based on the yield priority of components to achieve controllable energy dissipation, post-earthquake self-resetting, and low-damage control of the main structure. A finite element multi-scale model with embedded node hysteretic constitutive structures is constructed, and electromagnetic state sensing units are deployed at each node's SMA prestressed component to collect internal force and deformation signals. A full-structure monitoring matrix is ​​constructed through temperature compensation and spatiotemporal fusion of multi-source signals. The LightGBM-FCN fusion algorithm is used to identify structural conditions, inter-story damage distribution, and node cluster health status. An LSTM-Transformer network is used to predict the dynamic response, inter-story deformation, and residual deformation evolution of the structure. An over-limit index is calculated based on multi-dimensional safety thresholds, triggering four levels of safety early warning and providing operation and maintenance repair assessment conclusions. This invention can improve the shortcomings of insufficient monitoring coverage, delayed early warning, large post-earthquake residual deformation, and high repair difficulty in prefabricated buildings.

Description

Technical Field

[0001] This invention belongs to the field of prefabricated structure earthquake resistance and disaster prevention, health monitoring and numerical simulation technology, specifically involving a modular self-resetting connection and a method for global structural perception and early warning. Background Technology

[0002] The seismic performance of prefabricated frame structures is highly dependent on the connection nodes of structural components. The energy dissipation capacity, deformation recovery capacity, and monitorability of these nodes directly determine the overall seismic safety and post-earthquake repair efficiency of the structure. Traditional prefabricated structure connection nodes generally suffer from many drawbacks, such as a single energy dissipation mechanism, large residual deformation after an earthquake, concentrated damage, high repair difficulty, and high operation and maintenance costs.

[0003] Existing shape memory alloy (SMA) self-resetting node technology can reduce residual structural deformation to a certain extent. Monitoring technology based on electromagnetic sensing has the advantages of high durability and non-contact measurement of SMA internal forces. However, current research on related technologies is mostly limited to single optimization at the component level and node level, which can only achieve single-point monitoring of local nodes. It has not built an integrated system covering the health monitoring, condition identification, advanced prediction and hierarchical early warning of the entire structure, and cannot meet the needs of full-domain safety management of engineering structures.

[0004] Meanwhile, traditional general-purpose technologies have many technical problems, specifically as follows: Traditional finite element algorithms are prone to difficulties in computational convergence and low iteration efficiency when solving highly nonlinear conditions such as large structural deformation, interface friction slip, and hyperelastic phase transition of SMA materials, and cannot support the engineering requirements of real-time structural monitoring and rapid early warning; conventional structural monitoring systems mostly use external single-point sensors, which have problems such as cumbersome measurement point layout, complex wiring, high equipment failure rate, and poor overall stability; existing intelligent monitoring algorithms have not achieved deep coupling between multi-scale physical finite element modeling and deep learning algorithms, and the data-driven model lacks physical mechanism constraints, resulting in defects such as low recognition accuracy, insufficient reliability of prediction results, delayed early warning, and weak scene generalization ability.

[0005] In summary, there is currently a lack of a complete set of technical solutions that integrate node construction innovation, multi-scale accurate simulation, global state perception, intelligent damage identification, dynamic response prediction, and structural hierarchical early warning, making it difficult to achieve seismic resistance, disaster prevention, and health and safety management throughout the entire life cycle of prefabricated structures. Based on this, this invention proposes a modular self-resetting connection and a method for global structural perception and early warning. Summary of the Invention

[0006] In order to overcome the shortcomings of the existing technology, the present invention aims to provide a modular self-resetting connection and global structural perception and early warning method to achieve the integration of node-level construction innovation, structural multi-scale modeling, global perception, intelligent identification, advanced prediction and overall early warning.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] The first aspect of the present invention provides a modular self-resetting connection and a method for global structural perception and early warning, comprising the following steps: S1. Construct an integrated modular beam-column connection node that integrates energy dissipation, friction self-resetting, SMA superelastic reset, independent shear resistance and electromagnetic sensing. By presetting the yield priority of components, multi-level energy dissipation and post-earthquake self-resetting control of the structure under different seismic conditions can be achieved. S2. Based on vector finite element theory, a three-level multi-scale VFIFE model including structural, component, and node scales is constructed. The nonlinear hysteretic constitutive models corresponding to modular nodes are embedded into this multi-scale model system. The virtual inverse motion method is used to remove rigid body displacements during the model calculation process, and the explicit central difference algorithm is used to solve the structural dynamic response, thereby realizing the simulation analysis of the strongly nonlinear mechanical behavior of the structure. Based on this model, multi-condition numerical calculations are carried out to generate a multi-dimensional structural response simulation database in batches, providing data support for the training of subsequent intelligent recognition models. S3. Electromagnetic state sensing units are deployed in the SMA prestressed bars at each node. The internal forces and deformations of the SMA components are monitored non-contactly through coaxial coupling electromagnetic sensing. The multi-source sensor signals are fused in time and space and corrected for temperature to eliminate detection interference caused by ambient temperature and to build a standardized structural monitoring matrix covering the entire structure. S4. The LightGBM-FCN deep and shallow feature fusion algorithm is adopted, and the standardized structural monitoring matrix is ​​used as input to globally identify the structural working status, inter-layer damage distribution and node cluster health failure level. S5. An LSTM-Transformer hybrid time-series network is adopted. Relying on the dual advantages of LSTM capturing long-term time-series dependencies and Transformer focusing on key response features, a mapping relationship between load and structural dynamic response is constructed. The measured monitoring data and simulation time-series data are integrated to predict the evolution of structural dynamic response, inter-story deformation and residual deformation in advance. S6. Pre-set multi-dimensional structural safety thresholds such as inter-story drift angle, residual deformation, yield ratio of node clusters, SMA stress, and structural stiffness degradation rate. Calculate a normalized exceedance index based on this multi-dimensional structural safety threshold system to determine the overall risk level of the structure. Trigger a four-level safety warning according to the risk level classification. At the same time, use multi-parameter weighted performance indicators to quantitatively evaluate the overall seismic performance of the structure, output structural operation and maintenance and repair assessment results, and provide decision-making basis for structural safety management, operation and maintenance reinforcement, and emergency response.

[0009] Preferably, in step S1, the modular self-resetting beam-column connector serves as a standard seismic-resistant unit for the overall structure. It is integrated with four functional modules and a sensing system to achieve graded energy dissipation control based on different working states corresponding to earthquake levels: under minor earthquake conditions, the connection as a whole remains elastic; under moderate earthquake conditions, the low-yield soft steel energy dissipation module yields and dissipates energy preferentially; under major earthquake conditions, the self-resetting rotational friction energy dissipation module activates slippage energy dissipation; under extremely severe earthquake conditions, the SMA module enters the hyperelastic phase transition stage, simultaneously providing structural reset capability and energy dissipation capability.

[0010] Preferably, in step S2, modular nodes are embedded into the overall structure to construct a three-scale VFIFE model of structure-component-node. The mapping rules between each functional module and numerical element are as follows: the low-yield soft steel energy dissipation module corresponds to the nonlinear axial flexible rod element, the self-resetting rotational friction energy dissipation module corresponds to the nonlinear rotational spring element, the SMA self-resetting module corresponds to the flag-shaped constitutive spring element, and the shear-resistant element corresponds to the shear spring element. Explicit integration using central difference is employed. ,in, , , Let be the position vector of the particle. For time step, For quality, For the first The net force acting on each step.

[0011] Preferably, in step S3, the sensing module of the electromagnetic state sensing unit includes an excitation coil, a coaxial coil sensing element, a temperature sensor, a multi-channel acquisition instrument, and a signal demodulation and control instrument. It utilizes a coaxially coupled electromagnetic sensing structure to perform non-contact monitoring of the internal force and deformation of the SMA component. During operation, the excitation coil generates an alternating magnetic field. When the SMA component is subjected to force, its permeability changes, leading to changes in magnetic flux and induced voltage. The induced voltage satisfies the following formula: ,in, For induced voltage, The number of coil turns. For magnetic flux, For time; eliminating the influence of environment and temperature rise, the compensation formula is: ,in, This is the temperature-compensated magnetic characteristic signal. Original measurement signal, This is the temperature compensation coefficient. For real-time temperature, This is the reference temperature.

[0012] Preferably, in step S5, the input timing matrix... Output definition ,in, For external load time history, To monitor the response in real time, For SMA stress, For multi-scale simulation labels, This is the predicted value of the top-level displacement. This is the predicted value of the inter-story drift angle. This represents the predicted value of residual deformation. For the predicted value of the node's turning angle, This represents the trend of damage evolution.

[0013] Preferably, in step S6, the threshold exceeding the limit is uniformly determined using a normalized exceedance index. ,in, This is the overall structural safety exceedance index; the larger the value, the more dangerous the situation. The numerator is the real-time / predicted response value, and the denominator is the design limit. The threshold for the proportion of nodes yielding. The overall stiffness degradation rate of the structure; the overall seismic performance of the structure is comprehensively evaluated using the performance index PI. ,in, , , , These are the weighting coefficients, and their sum is 1. The total energy consumption of the structure, Energy consumption to meet design goals.

[0014] Preferably, step S1 includes the following sub-steps: S11, Detailed parameters and structural design of each functional module: The self-resetting rotating friction energy dissipation module, located at the neutral axis of the node, consists of two single-sided friction plates, a double-sided friction plate, a shear-resistant connection unit, and SMA bolts. The SMA bolts provide preload and unloading restoring force, and their friction activation bending moment and activation rotation angle satisfy… ,in, Friction activates bending moment, The distance from the bolt position to the center of rotation. For bolt preload, The coefficient of friction, The equivalent gradient of the friction plate, To activate the corner, Tangential stiffness; Low-yield mild steel energy dissipation modules are arranged at the upper and lower edges of the connection nodes. They utilize topology-optimized perforated mild steel plates and are connected to the end plates via bolts, allowing for quick replacement. Under minor to moderate earthquakes, they preferentially yield and dissipate energy, forming a "structural fuse." Their mechanical parameters meet the following requirements. , , ,in, The yield moment of the energy-dissipating mild steel. For mild steel yield strength, For the cross-sectional area of ​​mild steel, As the lever arm, For the bending stiffness of energy-dissipating mild steel, For the elastic modulus of mild steel, The effective length of the working segment. For the yield angle of energy-consuming mild steel; The SMA self-resetting module, composed of multiple prestressed Ni-Ti alloy rods, is arranged on both sides of the node to provide superelastic restoring capability and suppress residual structural deformation. Its axial strain and restoring bending moment satisfy... The elastic phase satisfies During the hyperelastic phase transition stage, the following conditions are met: ,in, For the axial strain of the SMA rod. For stress arm, The effective length of the working section of the SMA bar. The elastic modulus of SMA The cross-sectional area of ​​the SMA rod is... To restore the bending moment during the elastic phase, For the corner, and The equivalent activation bending moment and activation rotation are for the hyperelastic phase transition stage.

[0015] An independent shear connection unit is arranged at the center of the connection node. The independent shear connection unit includes a shear pin, and the outer wall of the pin is fitted with a polytetrafluoroethylene wear-resistant bushing. The independent shear connection unit independently bears all the vertical shear force of the precast concrete beam and column. The node rotational friction slip occurs in the neutral axis friction plate area, and the shear force transmission path is decoupled from the node bending slip and energy dissipation mechanism. Based on the hierarchical activation mechanism of each module, each module works sequentially according to the preset yield-activation order, and the overall bending moment-rotation relationship of the connection node conforms to a piecewise function:

[0016] in, This is the stiffness hardening coefficient of mild steel. The equivalent rotational stiffness of the self-resetting rotating friction energy dissipation module.

[0017] S12. Factory prefabrication and on-site dry assembly: steel end plates are pre-embedded at the ends of precast concrete beams and precast concrete columns, and module positioning holes are reserved; each functional module is independently processed, and the friction module and sensor kit are pre-assembled; on-site dry assembly is carried out in sequence, including temporary column fixing, beam alignment, shear pin installation, sensor kit layout, SMA rod tensioning and anchoring, friction module assembly, and soft steel module fixing, and protective sensor cables are stored. S13. Multi-earthquake collaborative operation: Under minor earthquakes, none of the modules are activated, and the nodes are in an elastic working state; under moderate earthquakes, the soft steel modules yield and dissipate energy, while the other modules remain elastic; under major earthquakes, the soft steel modules continue to dissipate energy, and the friction modules slip to assist in energy dissipation; under extremely severe earthquakes, the SMA modules undergo hyperelastic phase transformation, suppressing residual deformation of the nodes.

[0018] Preferably, step S3 includes the following sub-steps: S31. Sensing Hardware Composition and Deployment Scheme The sensing system includes a ring electromagnetic sensing kit, a patch temperature sensor, a multi-channel synchronous acquisition instrument, a signal demodulation control host, an industrial edge gateway, and protective accessories. Each node SMA rod is independently equipped with a ring electromagnetic sensing kit to form a sensing array with full coverage, and a backup sensing device is reserved to ensure continuous monitoring. The excitation coil and the induction coil are isolated by an insulating frame to ensure a stable distribution of the alternating magnetic field.

[0019] S32, Signal Acquisition and Temperature Compensation Algorithm An alternating signal is applied to the excitation coil to form a stable magnetic field. The SMA component is deformed by force, resulting in a change in magnetic permeability, which causes fluctuations in the induced voltage signal. A temperature compensation model is established, and the original monitoring signal is corrected by combining the real-time temperature and the reference temperature. By calibrating the fitting compensation coefficient and embedding it into the acquisition unit, real-time dynamic temperature compensation of the monitoring signal is achieved.

[0020] S33, Signal Inversion and Monitoring Matrix Construction By combining simulation analysis and on-site tension calibration, a linear inversion model is established to achieve non-contact and accurate identification of internal forces in SMA components. The stress, compensation voltage, temperature, and position coding features of each node are integrated to construct a single-node feature vector. The matrix is ​​then spliced ​​across the entire domain to generate a real-time monitoring matrix of the entire structure, which is then input into the AI ​​recognition model. Shielding, filtering, and single-point grounding anti-interference measures are adopted to ensure the stability of monitoring signals under complex working conditions.

[0021] Preferably, step S4 includes the following sub-steps: S41. Model Input / Output Definition Model input feature vector: Output state vector: , For SMA axial stress, Encode the node position. This refers to the overall working state of the structure. This is the yield state of mild steel. The friction module is in an active state. For the SMA working phase, The output represents the structural damage level. The input time series matrix has dimensions of [800 time steps, 18 nodes, 4 features], and the output includes engineering quantitative indicators such as the overall structural working level, module activation status, and hierarchical damage index.

[0022] S42, Model Training and Feature Fusion Mechanism To construct a gradient boosting tree, the LightGBM module uses a multi-class log loss function: log loss. ,in, For the number of state categories, To predict probabilities, Provides true labels; quickly extracts shallow correlation features from monitoring data and outputs a shallow preliminary state probability vector. ,in, The initial state probability vector is used; the one-dimensional FCN module performs convolutional feature extraction on the time series sequence. ,in, For convolution operations, For convolution kernel, For bias, The ReLU activation function is used; the deep temporal feature vector is obtained by global average pooling. , For deep temporal feature vectors, This is a time-series monitoring sequence. For global average pooling; the shallow initial state probability vector is concatenated and fused with the deep temporal feature vector: Finally, the global state is output through a fully connected layer and a softmax activation: ,in, For fully connected layer mapping, This is the Softmax activation function.

[0023] S43, Model Robustness Optimization and Fault-Tolerant Deployment The model generalization optimization was completed by adding gradient Gaussian noise and supplementing with on-site measured data. After pruning, quantization, and lightweighting, it was deployed on the edge gateway, supporting independent inference offline. It has single-node sensor fault tolerance capability, and relies on the spatiotemporal correlation features of the whole domain to ensure that the recognition process is not interrupted, and refreshes the structural status results every 0.2 seconds. S44. After the LightGBM-FCN fusion model is trained, the test set evaluation results will be used to verify the performance of the trained LightGBM-FCN fusion model on the test set samples, which include samples of frequent, fortified, rare, and extremely rare earthquake conditions.

[0024] Preferably, step S5 includes the following sub-steps: S51, LSTM temporal coding with long dependency capture, and a forget gate. The input gate is Candidate storage units are The cell status is updated to The output gate is The output signal is ,in, For the Sigmoid function, It is the hyperbolic tangent function. , , , These represent the input gate, forget gate, output gate, and storage unit at time [time]. activation vector, , , , These represent the model parameters for the input gate, forget gate, output gate, and storage unit input vectors, respectively. , , , These represent the model parameters for the input gate, forget gate, output gate, and hidden layer vectors of the storage units, respectively. S52, Transformer multi-head attention key feature focusing, scaled dot product attention. ,in, For query vector, For key vectors, For value vectors, For vector dimensions; S53, Multi-head Attention Fusion , ,in, , , A learnable weight matrix, used to... , , Projecting onto different subspaces allows each head to focus on different features. The outputs of multiple attention heads are concatenated to obtain a vector that integrates multi-dimensional key features. To output the projection matrix, the concatenated features are mapped to the final attention output; S54, fully connected mapping. ,in, Predict the structural response sequence for a future period of time. The features are extracted by LSTM (temporal features) and Transformer (key features). As a fully connected layer, it compresses and maps high-dimensional features into future multi-step structural responses. It can predict the next 5 to 20 time steps to achieve early warning.

[0025] Preferably, step S6 includes the following sub-steps: S61, establish a multi-scale threshold system for the overall structure consisting of inter-story drift angle, residual deformation, nodal yield ratio, SMA stress limit, and stiffness degradation rate; S62 calculates the structural safety exceedance index, compares real-time monitoring and future prediction responses with thresholds to determine exceedances; S63 outputs four levels of structural warnings: green, yellow, orange, and red, based on the degree of exceeding limits. S64 uses a weighted performance index formula to quantitatively evaluate the seismic performance of the structure; S65 outputs early warning levels, dangerous locations, damage status, and emergency decision-making basis, providing support for structural safety control.

[0026] A second aspect of the present invention provides a structural seismic resistance and health monitoring system, comprising cascaded functional units: a modular self-resetting node cluster unit, a multi-scale vector finite element modeling unit, a multi-node electromagnetic sensing array unit, a LightGBM-FCN structural condition identification unit, an LSTM-Transformer structural response prediction unit, and a graded early warning and assessment decision unit. Each unit operates collaboratively, sequentially completing tasks such as node load-bearing capacity assessment, numerical simulation modeling, on-site signal acquisition, structural condition identification, response trend prediction, safety graded early warning, and operation and maintenance assessment decision-making.

[0027] The third aspect of the present invention provides a prefabricated frame structure in which beam-column joints are assembled through modular self-resetting energy-dissipating connections; the prefabricated frame relies on the above-mentioned monitoring, evaluation and early warning technologies to carry out structural health monitoring, seismic performance evaluation and safety early warning.

[0028] Beneficial effects (1) This invention achieves the following through a three-level energy dissipation mechanism: yielding of soft steel, frictional slip, and superelasticity of SMA. It can also achieve the following: no damage in small earthquakes, repairability in moderate earthquakes, and no collapse in large earthquakes. The self-resetting capability of SMA can significantly reduce the residual deformation of the structure. The low-yield soft steel energy dissipation module can be quickly replaced, giving the structure the high toughness characteristics of rapid post-earthquake reset, rapid repair, and rapid commissioning. (2) The present invention adopts a three-scale VFIFE modeling method of structure, component and node, which does not require the formation of an overall stiffness matrix. It can efficiently handle complex mechanical behaviors such as large structural deformation, node friction slip, graded energy dissipation triggering and material nonlinear coupling, and support real-time analysis and rapid decision-making of the overall structure. (3) This invention relies on single-point electromagnetic sensing of SMA components to realize the global inversion of the overall structural working state, damage distribution, and node cluster performance. The system has low complexity, low cost, strong durability, and high long-term stability, solving the pain points of traditional structures with many monitoring points, complex wiring, and high failure rate. (4) The present invention uses the LightGBM-FCN fusion algorithm to achieve high-precision identification of structural state, with strong anti-noise ability and good generalization of small samples; it uses the LSTM-Transformer hybrid algorithm to achieve forward prediction of future multi-step response, and can automatically focus on the peak ground motion, weak layer and key node; the two work together to achieve accurate identification, fast prediction and early warning, and the overall performance is better than a single machine learning model. (5) The present invention constructs a closed-loop control system covering electromagnetic sensing, status recognition, response prediction, hierarchical early warning, performance evaluation, and replaceable repair. It can monitor the health status of the structure in real time, provide early warning of disaster risks, accurately locate the damage location, and provide a basis for repair decisions, so as to truly realize the intelligent, long-term, and resilient full life cycle operation and maintenance of prefabricated structures. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method steps of the present invention; Figure 2 This is a schematic diagram of the overall layout of the monitoring system of the present invention; Figure 3 This is a schematic diagram of the modular self-resetting energy-dissipating connection node assembly structure of the present invention; Figure 4 This is an exploded disassembly diagram of the modular self-resetting energy-dissipating connection node of the present invention; Figure 5 This is a schematic diagram of the multi-stage bending moment-rotation relationship of the present invention; Figure 6 This is a schematic diagram showing the self-reset node of the present invention and the corresponding VFIFE equivalent unit; Reference numerals: 1. Self-resetting rotating friction energy dissipation module; 2. Low yield soft steel energy dissipation module; 3. SMA self-resetting module; 4. Electromagnetic state sensing unit; 5. Left connecting plate; 6. Right connecting plate; 7. Multi-channel acquisition instrument; 8. Single-sided fixed friction plate; 9. Double-sided friction plate; 10. SMA bolt; 11. Shear-resistant connection unit; 12. Single-sided friction plate. Detailed Implementation

[0030] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The described embodiments are only some embodiments of the present invention, and not all embodiments. The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] like Figure 1As shown, this embodiment discloses a modular self-resetting connection and structural global perception and early warning method. Taking a three-story prefabricated concrete frame structure as the application object, it realizes the whole life cycle health monitoring, performance evaluation and safety early warning of the prefabricated frame structure through a complete set of technical solutions including modular node prefabrication, multi-scale nonlinear simulation, electromagnetic global perception, AI fusion recognition, time-series advanced prediction and hierarchical safety early warning. The specific implementation steps are as follows: S1. Construct an integrated modular beam-column connection node that integrates energy dissipation, friction self-resetting, SMA superelastic reset, independent shear resistance and electromagnetic sensing. By presetting the yield priority of components, multi-level energy dissipation and post-earthquake self-resetting control of the structure under different seismic conditions can be achieved. Modular self-resetting energy-dissipating beam-column connections are implemented through factory prefabrication and on-site dry assembly. They are sequentially configured with low-yield soft steel energy-dissipating modules, self-resetting rotational friction energy-dissipating modules, SMA self-resetting modules, and independent shear connection units. Nodes integrate electromagnetic sensing modules, relying on the yield rotation angle thresholds of each component to achieve graded energy dissipation and post-earthquake self-resetting damage control. This invention's modular self-resetting energy-dissipating beam-column connection is an integrated standardized node, integrating multi-level energy dissipation, hyperelastic self-resetting, independent shear resistance, and non-contact electromagnetic sensing functions. Through strict module yield priority design, it achieves a graded working mechanism of elasticity in minor earthquakes, energy dissipation in moderate earthquakes, multi-level energy dissipation in major earthquakes, and damage control during extremely severe earthquakes. The activation angle of each functional module strictly meets the following requirements: The nodes are manufactured using standardized prefabrication in the factory and dry assembly on site, eliminating wet work on site. This results in high assembly precision, fast construction speed, and rapid repair after an earthquake.

[0032] S11. Detailed parameters and structural design of each functional module (1) Self-resetting rotating friction energy dissipation module The self-resetting rotating friction energy dissipation module 1 is located at the neutral axis of the node and is the core sliding energy dissipation component. It mainly consists of two single-sided friction plates 12 on both sides, a double-sided friction plate 9 in the middle, SMA bolts 10, and a shear-resistant connection unit 11. The base plate is made of Q355B hot-rolled steel plate, and the thickness of both single-sided and double-sided friction plates is 14mm. The friction contact surface is sprayed with a 4mm thick copper-based powder metallurgy wear-resistant coating, with a designed friction coefficient μ=0.3 and a wear resistance cycle of not less than 2000 cycles. The pre-tightening component uses 4 Φ14NiTi-SMA bolts, with a designed pre-tightening force of 40kN per bolt, and the distance from the bolt center to the node rotation center is 190mm. The friction plate contact surface is set with a 1:120 micro equivalent gradient to effectively avoid reciprocating rotational sliding jamming. The factory adopts a three-stage pre-tightening process, with each stage holding the load for 30 minutes. After assembly, the pre-tightening force loss is re-measured after 24 hours of static rest and then re-tensioned to the design value. The friction activation angle of this module is set to 0.0075 rad to ensure no slippage under moderate earthquake conditions and to initiate slippage energy dissipation under rare large earthquake conditions. The SMA bolts provide preload and unloading recovery force, and their friction activation bending moment and activation angle satisfy the following formula:

[0033] In the formula: Friction activates bending moment, The distance from the center of the bolt to the center of rotation. For bolt preload, The coefficient of friction, The equivalent gradient of the friction plate, Activate the rotation angle for the friction module. This refers to the tangential stiffness.

[0034] (2) Low yield steel energy dissipation module Low-yield mild steel energy dissipation module 2 is arranged at the upper and lower edges of the connection node, serving as a seismic "fuse" for the structure. It uses Q100 ultra-low yield mild steel plate with a thickness of 10mm. Through topology optimization design, it features an elongated elliptical perforated structure with a major axis of 70mm and a minor axis of 25mm. Each mild steel plate has 6 rows of symmetrical perforations, achieving preferential yielding and energy dissipation through cross-sectional weakening. The effective working length of each mild steel plate is 280mm, and the lever arm is 380mm. It is connected to the precast beam and column embedded steel end plates using detachable M18 8.8 grade high-strength bolts, with a 3mm installation gap reserved for the bolts. After an earthquake, there is no need to remove the beams and columns or cut the concrete; each mild steel plate can be disassembled and replaced within 10 minutes. According to the factory cyclic loading test calibration, this module is the first energy dissipation module to be activated in the entire node system, and the only energy dissipation module under small to medium earthquake conditions, ensuring that the precast concrete beams and columns remain elastic and free from crack damage throughout the entire process.

[0035] The mechanical parameters of this module satisfy the following calculation formula:

[0036] In the formula: For the yield moment of energy-dissipating mild steel, For mild steel yield strength, For the cross-sectional area of ​​mild steel, It is a soft steel lever arm. For the bending stiffness of mild steel, For the elastic modulus of mild steel, This is the effective working length of the mild steel. This refers to the yield angle of mild steel.

[0037] (3) SMA self-reset module The SMA self-resetting module 3 is symmetrically arranged on both sides of the node, consisting of multiple superelastic prestressed NiTi shape memory alloy rods. Each SMA rod has a diameter of 20mm, an effective working length of 950mm, and a lever arm of 400mm. Four SMA rods are symmetrically arranged at a single node. The material's elastic modulus is 42GPa, and the austenite-to-martensite phase transformation initiation stress is 380MPa. At the factory, it is pre-tensioned and locked with 0.75% prestress, and after unloading, it possesses complete superelastic reset capability. The module's phase transformation initiation rotation angle is 0.015rad, entering the phase transformation energy dissipation stage only under extremely severe earthquake conditions. The superelastic restoring force offsets residual structural deformation, achieving near-zero residual rotation angle at the node after the earthquake.

[0038] The axial strain and recovering bending moment response of this module satisfy... The elastic phase satisfies During the hyperelastic phase transition stage, the following conditions are met: ,in, For the axial strain of the SMA rod. For stress arm, The effective length of the working section of the SMA bar. The elastic modulus of SMA The cross-sectional area of ​​the SMA rod is... To restore the bending moment during the elastic phase, For the corner, and The equivalent activation bending moment and activation rotation are for the hyperelastic phase transition stage.

[0039] (4) Independent shear connection unit This precast concrete beam-column connection joint is provided with an independent shear connection unit, which is arranged at the center of the connection joint and adopts a shear pin shaft structure. The outer wall of the pin shaft is sleeved with a polytetrafluoroethylene wear-resistant bush, which can independently bear all the vertical shear force of the precast concrete beam and column. The rotational frictional slip of the joint only occurs in the friction plate area at the neutral axis, and the shear force transfer path is completely decoupled from the bending slip and energy dissipation mechanism of the joint. The ultimate shear bearing capacity of the joint can reach 530kN, which can fully meet the requirements of vertical loads of three-story concrete frames and the vertical component action of earthquakes, and effectively prevent the shear cracking problem at the joints of precast beams and columns.

[0040] The joint is internally integrated with a low-yield mild steel energy dissipation module, a self-centering rotational friction energy dissipation module and an SMA self-centering module. The multi-modules work in a coordinated manner at different levels, so that the yielding and activation sequence of each module of the joint is controllable, and the overall moment-rotation relationship of the joint presents piecewise variation characteristics:

[0041] wherein, b is the stiffness hardening coefficient of mild steel, is the equivalent rotational stiffness of the self-centering rotational friction energy dissipation module.

[0042] S12, Factory prefabrication and on-site dry assembly Refer to Figure 4 and Figure 5 in the processing stage of the precast component factory, the PC beams and PC columns are first cast and formed, 20mm thick Q355B steel end plates are embedded at the ends of the beams and columns, and the positioning holes of each module are precisely reserved; then each functional module is independently processed, mechanical calibration, anti-corrosion coating and insulation detection are completed, and the friction modules and sensor kits are pre-assembled before being packaged and transported as a complete set. The construction site adopts a full dry assembly process, which sequentially completes the temporary fixing of the column body, the alignment of the beam body, the installation of the shear pin shaft, the layout of the sensor kits, the SMA anchoring, the fixing of the self-centering rotational friction energy dissipation module and the low-yield mild steel energy dissipation module as well as the layout of the sensor kits, and finally the sensor cables are collected and protected in a unified manner. After completion, 10% of the joints are randomly selected for cyclic rotation loading test. The joint is qualified if the residual rotation of the joint after unloading is ≤0.0005rad and there are no microcracks on the end plate.

[0043] S13, Coordinated operation under multiple seismic conditions Refer to Figure 5 under the frequent minor earthquake condition, the entire joint is in the elastic range, all modules are not activated, and the main frame is completely elastic without cracks; under the design moderate earthquake condition, only the low-yield mild steel energy dissipation module yields and dissipates energy, and the remaining modules remain elastic, so that the joint performance can be restored by only replacing the mild steel after the earthquake; under the rare major earthquake condition, the mild steel continuously dissipates energy, the friction module starts to slip to supplement energy dissipation, and only closed microcracks are generated in the concrete; under the extremely rare super-major earthquake condition, the SMA module enters the superelastic phase transformation stage, which relies on its excellent self-centering capacity to restrain the residual deformation of the structure and ensure that the main structure has no structural through damage.

[0044] S14. Sensing hardware deployment and electromagnetic coupling structure The nodes are equipped with an intelligent sensing and monitoring system. The sensing hardware mainly includes coil-type sensing elements, excitation coils, insulating frames, patch temperature sensors, signal demodulation and control hosts, multi-channel synchronous acquisition instruments, 5G industrial edge gateways, and supporting protective accessories. The excitation coils are arranged in a ring around the outer periphery of the SMA rod, and the coil-type sensing elements and excitation coils are arranged coaxially in a ring, forming an electromagnetic coupling structure. The excitation coils and coil-type sensing elements are isolated by an epoxy fiberglass insulating frame, effectively ensuring a uniform and stable distribution of the alternating magnetic field. Patch temperature sensors are deployed near the surface of the SMA rod, simultaneously collecting ambient temperature and local temperature rise signals of the component, providing basic data for subsequent temperature compensation and correction. Each SMA rod at each node is independently equipped with an electromagnetic sensing kit, forming a structurally comprehensive sensing array, while backup sensing equipment is reserved to ensure the continuity and reliability of the overall monitoring operation.

[0045] S15, Signal excitation acquisition and temperature compensation processing During system operation, the signal demodulation control host generates pulse signals or AC excitation signals to drive the excitation coil to produce a stable alternating magnetic field. The deformation of the SMA component under stress causes changes in its permeability, leading to fluctuations in the induced voltage of the coil-type sensing element. The multi-channel synchronous acquisition instrument collects the magnetic characteristic electrical signals output by the sensing element in real time. To eliminate monitoring interference caused by ambient temperature rise and component heating, the system constructs a temperature compensation model. It dynamically corrects the original measured signal by combining the real-time monitored temperature with the reference calibration temperature. The compensation coefficient is obtained through gradient temperature calibration experiments and embedded into the acquisition instrument's microcontroller, achieving real-time dynamic temperature compensation of the monitoring signal. Simultaneously, anti-interference measures such as cable shielding, hardware filtering, and single-point grounding ensure stable signal acquisition under complex construction and service conditions.

[0046] S16. Stress Inversion Modeling and Intelligent Nodal State Recognition During the rotational deformation of nodes, the axial tensile or compressive deformation of SMA rods is clearly coupled with the node connection rotation angle. This system combines VFIFE simulation theoretical stress with on-site tension calibration tests, establishing a linear stress inversion model through experimental calibration. This allows for non-contact and accurate identification of internal forces in SMA components without the need for attached strain gauges. By analyzing the time history characteristics of SMA axial stress, the system can accurately determine the actual working state and deformation development trend of nodes. Furthermore, by integrating multi-dimensional features of SMA stress, compensated voltage, real-time temperature, and position coding for each node, a four-dimensional feature vector for each node is constructed. This vector is then fully spliced ​​and refreshed in real time across the entire structure to generate a standardized monitoring matrix for the overall structure. This matrix is ​​continuously input into the AI ​​recognition model, enabling comprehensive intelligent monitoring and identification of the working state of precast beam-column connection nodes.

[0047] S2. Multi-scale VFIFE modeling and batch simulation implementation of a three-story prefabricated concrete frame. A three-level VFIFE multi-scale model was constructed, comprising nodes, components, and the structure. Hysteretic constitutive models were assigned to the nodes, and nonlinear constitutive models to the component materials. The overall coupled solution was then used to solve the structural dynamic time-history response under seismic loading. (Refer to...) Figure 6 This invention uses vector finite element method (VFIFE) to establish a three-level coupled multi-scale model of structure-component-node, without the need to assemble the overall stiffness matrix. It can accurately adapt to multiple strongly nonlinear coupled working conditions such as concrete damage, node friction slip, and SMA phase transformation, and meets the deformation coordination and force balance conditions throughout the process.

[0048] S21. Multi-scale unit mapping and solution algorithm The structural scale uses lumped mass point elements, effectively converting floor loads to nodal mass points. Rigid body displacements are stripped away using a virtual inverse motion algorithm, retaining only the pure deformation response of the structure. The component scale uses fiber beam-column elements, defining plastic damage constitutive models for concrete and kinematic hardening constitutive models for steel reinforcement. At the nodal scale, each functional module is equivalent to a standardized spring element: the low-yield soft steel energy dissipation module 2 corresponds to a nonlinear axial flexible rod element, the self-resetting rotational friction energy dissipation module corresponds to a nonlinear rotational spring element, the SMA self-resetting module corresponds to a flag-shaped hysteretic spring element, and the shear element corresponds to an independent shear spring element. Each spring element is coupled to the beam-column fiber element through end rigid arms. Specifically, the following sub-steps are included: The dynamic solution employs an explicit central difference integral algorithm, with the following iterative formula:

[0049] In the formula: Let h be the position vector of the particle at different times, h be the time step, and m be the mass of the particle. The net force on the particle in the nth step is calculated using a time step of 0.002s in this embodiment.

[0050] S22. Batch generation and preprocessing of simulation datasets Based on the response spectrum of a Class II site at 8 degrees and 0.2g, 100 natural seismic waves were selected, with peak ground acceleration gradients covering 0.08g to 0.8g, fully covering seismic conditions of magnitude 4. Batch simulation outputs a complete set of characteristic parameters, including top-floor displacement, inter-story drift angle, node rotation angle, SMA stress, concrete damage index, residual deformation, stiffness degradation rate, and total structural energy dissipation. After data cleaning and removal of outliers, 11,800 valid time-history samples were obtained, divided into training, validation, and test sets in a 7:2:1 ratio. The time-series sampling frequency was uniformly set to 50Hz, perfectly aligned with the field sensor acquisition frequency, achieving seamless integration of simulation and measured data. Parallel computing using dual-socket RTX5090 GPUs improved simulation efficiency by more than 10 times compared to traditional finite element methods, enabling continuous batch generation of AI training samples.

[0051] S3. Node electromagnetic sensing array deployment, signal acquisition, and temperature compensation calibration. Electromagnetic sensors are deployed on SMA (Structured Aluminum Mullion) components to collect stress, magnetic signals, and temperature data. Temperature compensation correction and spatiotemporal fusion of multi-source signals are performed, and the data from all nodes are stitched together to generate a standardized overall monitoring matrix. This invention employs non-contact coaxial electromagnetic sensing technology, relying on electromagnetic state sensing units deployed throughout the entire area to achieve real-time monitoring of the internal forces and node states of SMA components. A temperature compensation mechanism is included to eliminate environmental interference, and a standardized full-structure monitoring matrix is ​​constructed. Specifically, the following sub-steps are included: S31. Sensing Hardware Composition and Deployment Scheme The sensing system hardware includes a ring-shaped electromagnetic sensing kit, a PT100 patch temperature sensor, a 24-channel synchronous acquisition unit, a signal demodulation and control host, a 5G industrial edge gateway, and supporting protective accessories. The three-layer frame comprises 18 standardized nodes, each with four SMA rods fitted with an independent electromagnetic sensing kit, forming a full-coverage sensing array. Two additional backup sensing devices are reserved in the weakest layer to ensure continuous monitoring. The sensing coils use 0.7mm enameled copper wire, with 100 turns in the excitation coil and 220 turns in the induction coil, isolated by an epoxy fiberglass insulation skeleton to ensure a stable alternating magnetic field distribution.

[0052] S32, Signal Acquisition and Temperature Compensation Algorithm During operation, the excitation coil outputs a 50Hz, 12V alternating voltage to generate a stable magnetic field. The deformation of the SMA component under stress causes a change in magnetic permeability, resulting in fluctuations in the induced voltage. The induced voltage satisfies the following: .

[0053] To eliminate the interference of ambient temperature rise and component heating, a temperature compensation model is established:

[0054] In the formula: This is a temperature-compensated, standardized magnetic signal. The original measured signal is given, k is the temperature compensation coefficient, and T is the real-time monitored temperature. The reference calibration temperature is used. The compensation coefficient is obtained through a temperature gradient calibration test from 10℃ to 35℃, and the microcontroller of the data acquisition instrument is implanted to realize real-time dynamic compensation.

[0055] S33, Signal Inversion and Monitoring Matrix Construction By combining VFIFE simulation theoretical stress with on-site tension calibration tests, a system was established. The linear inversion model enables non-contact and accurate identification of SMA internal forces without the need for attached strain gauges. It integrates SMA stress, compensation voltage, real-time temperature, and location coding features at each node to form a single-node four-dimensional feature vector. These 18 node vectors are then fully spliced ​​together, generating a standardized monitoring matrix for the entire structure every second, which is then input into the AI ​​recognition model in real time. Simultaneously, anti-interference measures such as cable shielding, hardware filtering, and single-point grounding ensure a signal-to-noise ratio higher than 35dB even under complex construction conditions.

[0056] Deployment and Application of S4 and LightGBM-FCN Fusion Structure Global State Recognition Model LightGBM mines shallow statistical features, and FCN extracts temporal deep features. The two types of features are weighted and fused to output the overall structural operating status, inter-layer damage distribution, and node cluster health. This invention employs the LightGBM-FCN shallow and deep feature fusion algorithm to achieve globally accurate identification of structural operating status, damage distribution, and node cluster failure levels based on a global monitoring matrix. Specifically, it includes the following sub-steps: S41. Model Input / Output Definition Model input feature vector: Output state vector: , For SMA axial stress, Encode the node position. This refers to the overall working state of the structure. This is the yield state of mild steel. The friction module is in an active state. For the SMA working phase, The output represents the structural damage level. The input time series matrix has dimensions of [800 time steps, 18 nodes, 4 features], and the output includes engineering quantitative indicators such as the overall structural working level, module activation status, and hierarchical damage index.

[0057] S42, Model Training and Feature Fusion Mechanism To construct a gradient boosting tree, the LightGBM module uses a multi-class log loss function: log loss. ,in, For the number of state categories, To predict probabilities, Provides true labels; quickly extracts shallow correlation features from monitoring data and outputs a shallow preliminary state probability vector. ,in, The initial state probability vector is used; the one-dimensional FCN module performs convolutional feature extraction on the time series sequence. ,in, For convolution operations, For convolution kernel, For bias, The ReLU activation function is used; the deep temporal feature vector is obtained by global average pooling. , For deep temporal feature vectors, This is a time-series monitoring sequence. For global average pooling; the shallow initial state probability vector is concatenated and fused with the deep temporal feature vector: Finally, the global state is output through a fully connected layer and a softmax activation: ,in, For fully connected layer mapping, This is the Softmax activation function.

[0058] S43, Model Robustness Optimization and Fault-Tolerant Deployment The model generalization optimization was completed by adding gradient Gaussian noise and supplementing with on-site measured data. After pruning, quantization, and lightweighting, it was deployed on the edge gateway, supporting independent inference offline. It has single-node sensor fault tolerance capability, and relies on the spatiotemporal correlation features of the whole domain to ensure that the recognition process is not interrupted, and refreshes the structural status results every 0.2 seconds. Evaluation results of the S44 and LightGBM-FCN fusion models on the test set after training. The trained LightGBM-FCN fusion model was validated on a test set, which included samples of frequent, fortified, rare, and extremely rare earthquake conditions. The model achieved an average absolute error ≤0.03 for identifying the concrete layered damage index; a node damage level identification accuracy ≥94.2%; and a single-frame inference time ≤80ms, meeting the real-time requirement of a 0.2s refresh output for the edge gateway. Compared to using the LightGBM or one-dimensional FCN baseline model alone, this fusion model improved the accuracy of identifying complex multi-node cluster failure conditions by more than 7%. Even under simulated sensor disconnection and electromagnetic noise interference conditions, it maintained an identification accuracy of over 90%, demonstrating single-point fault tolerance.

[0059] S5 and LSTM-Transformer hybrid time series prediction model for early warning implementation This invention employs a hybrid temporal network, combining LSTM to capture long-term temporal dependencies and Transformer multi-head attention to pinpoint critical risk periods. This network predicts future dynamic responses such as inter-story displacement and residual deformation.

[0060] S51, Network Core Calculation Formula LSTM units capture long-term dependencies through multiple gating mechanisms: The Gate of Oblivion The input gate is Candidate storage units are The cell status is updated to The output gate is The output signal is ,in, For the Sigmoid function, It is the hyperbolic tangent function. , , , These represent the input gate, forget gate, output gate, and storage unit at time [time]. activation vector, , , , These represent the model parameters for the input gate, forget gate, output gate, and storage unit input vectors, respectively. , , , These represent the model parameters for the input gate, forget gate, output gate, and hidden layer vectors of the storage units, respectively.

[0061] Transformer multi-head attention focuses on key features by scaling the dot product: Multi-head feature fusion: , Finally, the fully connected layer maps and outputs the future multi-step response sequence: It can predict the structural dynamic response in 5 to 20 steps ahead.

[0062] S52, Model Engineering Adaptation Strategy Four attention heads are set up to focus on the peak seismic intensity, weak layer signal, stress exceeding the limit node, and the period of sudden damage change, respectively, to accurately adapt to the delayed degradation characteristics of concrete structures. Every 1 second, the prediction deviation is corrected by combining VFIFE simulation data to avoid the nonlinear distortion problem of pure data-driven models, realize the advanced prediction of structural response, and reserve a window of opportunity for early warning and response.

[0063] The S53 and LSTM-Transformer hybrid temporal prediction model test set evaluation results were used to validate the trained LSTM-Transformer hybrid temporal prediction model using time-series samples from the test set. For the advanced prediction of structural response in the future 0.1~0.4s, the average relative error of the inter-story drift angle prediction is ≤4.8%, and the average absolute error of the top floor residual displacement prediction is ≤0.8mm. Compared with the single LSTM baseline model, the prediction error is reduced by about 32% under strong nonlinear earthquake conditions. One round of prediction output is completed every 1s, which can meet the temporal requirements for advanced early warning of structures under earthquake action.

[0064] S6. Implementation of Multi-Threshold Graded Early Warning and Structural Seismic Performance Assessment This invention constructs a multi-dimensional safety threshold system, calculates a global over-limit index to classify four levels of early warning, quantifies the seismic performance index PI, and outputs operation and maintenance plans and post-earthquake damage repair suggestions. The invention constructs a multi-dimensional structural safety threshold system, determines the risk level by normalizing the over-limit index, and combines weighted performance indicators to quantitatively evaluate seismic performance, thereby achieving four-level graded early warning and accurate performance evaluation.

[0065] S61. Safety Threshold and Over-Limit Judgment Formula Formula for calculating the overall structural safety exceedance index:

[0066] In the formula: ξ is the overall over-limit index. The larger the value, the higher the structural risk. ξ corresponds to the ratio of the real-time / predicted value to the limit value of the five core indicators: inter-story drift angle, residual deformation, SMA stress, nodal yield ratio, and stiffness degradation rate.

[0067] S62, Level 4 Early Warning Judgment Logic The over-limit index is calculated synchronously based on measured data and predicted data, and the maximum value is used as the basis for early warning judgment: ξ≤0.6 is a green safe state, only data is stored and there is no alarm; 0.6<ξ≤1.0 is a yellow alert state, and inspection reminders are pushed; 1.0<ξ≤1.5 is an orange warning state, the monitoring screen of the weak layer is highlighted and evacuation prompts are pushed; ξ>1.5 is a red emergency state, and audible and visual alarms and emergency broadcasts are activated.

[0068] S63, Seismic Performance Quantitative Evaluation Model Formula for calculating the overall seismic performance index PI of a structure:

[0069] Where: weight The total weight is 1; The total energy consumption of the structure, The system is designed to meet energy consumption targets. It automatically generates daily operation and maintenance reports and outputs a special assessment report immediately after an earthquake, including a damage list, repair recommendations, and long-term operation and maintenance forecasts.

[0070] Based on the above-mentioned modular self-resetting connection and structural global perception and early warning method, the present invention also provides a seismic health monitoring system that can be implemented using the method, as well as a prefabricated frame structure built using the self-resetting node.

[0071] This embodiment also provides a structural seismic resistance and health monitoring system. This system can operate the aforementioned modular self-resetting connection and global structural perception and early warning methods. Internally, the system comprises cascaded and collaborative functional units: a modular self-resetting node cluster unit, a multi-scale vector finite element modeling unit, a multi-node electromagnetic sensing array unit, a LightGBM-FCN overall structural state identification unit, an LSTM-Transformer overall structural response prediction unit, and a structural-level hierarchical early warning and assessment decision-making unit. These units work collaboratively to sequentially achieve a complete closed loop encompassing structural node seismic load-bearing capacity, refined numerical simulation modeling, on-site sensor signal acquisition, structural damage and operational status identification, structural deformation development trend prediction, safety-level early warning, and operation and maintenance plan decision-making.

[0072] In addition, this embodiment also provides a prefabricated frame structure in which all beam-column joints are assembled using the modular self-resetting energy-dissipating connection structure described above. This prefabricated frame structure can be equipped with the aforementioned complete set of sensing and early warning methods to carry out overall structural health monitoring, seismic performance quantitative assessment, and graded safety early warning and control throughout its entire life cycle. Relying on the excellent seismic self-resetting performance of the self-resetting nodes themselves, combined with an integrated intelligent monitoring and early warning system, the overall seismic toughness and long-term operation and maintenance safety management capabilities of the prefabricated frame are effectively improved.

[0073] The above embodiments are merely preferred embodiments of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A modular self-resetting connection and structural global awareness and early warning method, characterized in that, Includes the following steps: S1. Construct an integrated modular beam-column connection node that integrates energy dissipation, friction self-resetting, SMA superelastic reset, independent shear resistance and electromagnetic sensing. By presetting the yield priority of components, the structure can achieve graded controllable energy dissipation and low-damage self-resetting after earthquake. S2. Construct a three-level vector finite element multi-scale model of structure, components, and nodes with embedded nodal hysteretic constitutive models. Use the virtual inverse motion method to strip rigid body displacements and combine it with the explicit central difference algorithm to solve the structural dynamic response and complete the simulation of the strong nonlinear mechanical behavior of the structure. S3. Electromagnetic state sensing units are installed in the SMA prestressed bars at each node to monitor the internal forces and deformations of the components in a non-contact manner through coaxial coupling electromagnetic sensing. Spatiotemporal fusion and temperature compensation correction are performed on multi-source sensor signals to construct a standardized structural monitoring matrix covering the entire structure. S4. The LightGBM-FCN deep and shallow feature fusion algorithm is adopted, and the standardized structural monitoring matrix is ​​used as input to globally identify the structural working status, inter-layer damage distribution and node cluster health failure level. S5. An LSTM-Transformer hybrid time-series network is adopted to integrate measured monitoring data and simulated time-series data to predict the dynamic response, interlayer deformation and residual deformation evolution of the structure in advance. S6. Calculate the over-limit index based on the multi-dimensional structural safety threshold system, trigger four-level safety warnings in stages, quantitatively evaluate the overall seismic performance of the structure, and output the structural operation and maintenance and repair assessment results. The integrated modular beam-column connection node in S1, which integrates energy dissipation, friction self-resetting, SMA superelastic reset, independent shear resistance, and electromagnetic sensing, is an integrated standard node. The node integrates the following functional modules: Low-yield mild steel energy dissipation modules are deployed at the upper and lower edges of the connection, prioritizing yielding and dissipating seismic energy. Self-resetting rotational friction energy dissipation modules are deployed near the neutral axis of the node, assembled from friction plates and SMA bolts, providing frictional energy dissipation and structural restoring force. The SMA self-resetting module is composed of prestressed Ni-Ti shape memory alloy rods, providing structural reset capability through hyperelasticity and suppressing residual structural deformation. Independent shear connection units are deployed at the connection center, constructing independent shear force transmission paths to decouple the shear mechanism from node friction slip and bending energy dissipation mechanisms. Electromagnetic state sensing units are non-contactly fitted onto the outer periphery of the SMA rods to collect internal force and deformation monitoring signals of the SMA components. Each functional module is activated sequentially according to different seismic conditions, with the mild steel yield rotation angle, friction activation rotation angle, and SMA phase transformation activation rotation angle increasing in sequence. The overall connection exhibits piecewise linear bending moment-rotation response characteristics, achieving graded energy dissipation and reset control. The modular beam-column connection node is equipped with an orderly graded yield energy dissipation mechanism. The activation order is as follows: the low yield soft steel energy dissipation module yields and dissipates energy first, then the self-resetting rotational friction energy dissipation module activates the sliding energy dissipation module, and finally the SMA module enters the hyperelastic phase transformation stage, realizing graded energy dissipation of the structure and low damage control of the main structure.

2. The modular self-resetting connection and structural global awareness and early warning method of claim 1, wherein, The vector finite element multi-scale model includes a three-level coupled system of structural scale, component scale, and node scale. Each scale model is a standardized and callable unit module that maintains deformation coordination and force balance. The modular element mapping method is used to match the mechanical constitutive model of each functional module. The vector finite element does not require the assembly of the overall stiffness matrix. The dynamic response of the structure under large deformation, strong nonlinearity, and multi-node coupling is solved by explicit central difference integration.

3. The modular self-resetting connection and structural global awareness and early warning method of claim 1, wherein, The electromagnetic state sensing unit described in S3 adopts a coaxial coupling arrangement of excitation coil and sensing coil. Combined with temperature synchronous acquisition and temperature compensation correction mechanism, it eliminates the interference of ambient temperature rise on the magnetoelectric signal. Combined with spatiotemporal fusion processing of multi-source sensing signals, it constructs a standardized monitoring matrix covering the entire structure. The model training dataset is generated by multi-scale vector finite element batch simulation, covering the structural time history response, deformation, stress, damage, energy consumption, and earthquake and wind load time history parameters.

4. The modular self-resetting connection and global structural perception and early warning method according to claim 1, characterized in that, The LightGBM-FCN deep and shallow feature fusion algorithm adopts a deep and shallow feature layered fusion recognition mechanism, extracts shallow correlation features and temporal spatial depth features from monitoring data and weights them together to globally identify the structural working status, inter-layer damage distribution and node cluster health failure level.

5. The modular self-resetting connection and global structural perception and early warning method according to claim 1, characterized in that, The LSTM-Transformer hybrid temporal network combines the long temporal dependency capture capability of LSTM with the multi-head attention focusing capability of Transformer to output the structural response prediction results for the next 5 to 20 steps, thereby achieving advanced prediction of the structural dynamic response.

6. The modular self-resetting connection and global structural perception and early warning method according to claim 1, characterized in that, The multi-dimensional structural safety threshold system includes inter-story drift angle, residual deformation, yield ratio of node clusters, SMA stress, and structural stiffness degradation rate limits. It uses a multi-index normalized exceedance index to determine the overall structural risk, outputs four-level early warnings based on the degree of exceedance, and quantitatively evaluates the overall seismic performance of the structure through multi-parameter weighted performance indicators to support early warning issuance and performance report output.

Images