Intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology

Abstract

The application provides a steel structure deformation intelligent early warning monitoring system based on a laser measuring instrument technology, and relates to the technical field of building dynamic monitoring.The system comprises a dynamic laser ranging array, a spatial stiffness laser scanning unit, a time-space registration and matrix conversion module, a mechanical modal space reconstruction server and an energy state grading early warning terminal.The system realizes the unity of high-frequency time efficiency and full-field surface coverage through double-track asynchronous sampling, uses a generalized inverse matrix algorithm with regularization constraints to solve the problem of ill-conditioned matrix divergence under sparse measurement points, effectively suppresses measurement noise, compensates through a mechanical modal base fusion residual neural network, eliminates nonlinear distortion caused by material heterogeneity and welding residual stress, reconstructs a continuous three-dimensional deformation displacement field with high fidelity during a scanning window period, and finally realizes intelligent grading closed loop of overall instability, local buckling and rapid damage expansion based on time-space integration and differential solution of strain energy density.

Description

Technical Field

[0001] This invention relates to the field of building dynamic monitoring technology, specifically to an intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology. Background Technology

[0002] As my country's large public buildings, transportation hubs, and large-span industrial plants evolve towards ultra-large spans, complex spatial grid systems, and towering digital structures, higher demands are placed on the online monitoring and safety early warning of steel structure deformation during long-term service. In particular, all-weather real-time monitoring, adaptive damage identification, and reliable early warning of extreme instability under dynamic environments such as wind load, snow load, temperature stress, and uneven foundation settlement have become the focus of attention in the engineering community. For example, the large-span roof of high-speed railway station buildings requires highly timely dynamic displacement capture, large steel structure box girder bridges require continuous spatial contour deformation field reproduction, and the core tube of steel structure of super high-rise buildings requires real-time assessment of local buckling and overall instability risks. Currently, three-dimensional laser scanning, high-frequency laser ranging, and modern sensor technology have been gradually applied to the field of building structural health monitoring.

[0003] However, existing steel structure deformation monitoring technologies face the following key technical challenges in meeting these requirements:

[0004] 1. Insufficient spatiotemporal resolution and full-field coverage. Traditional manual inspections often use total stations, levels, or optical projection tools for periodic inspections. The accuracy of these inspections is greatly affected by human experience, making it difficult to quantify and impossible to achieve all-weather online monitoring. In terms of sensor monitoring, existing technologies mostly use fiber optic gratings, distributed optical fibers, or vibrating wire strain gauges. Although they can acquire local strain, the sensor deployment density is severely insufficient, and they can only monitor isolated local points. Another type of detection device, based on multi-view vision or single contact, not only requires contact sliding which can easily cause structural damage, but it can also only acquire discrete line trajectories or deformation of local feature points. When facing a large-span scanning spatiotemporal window, it cannot reconstruct a continuous three-dimensional deformation displacement field across the entire spatial domain at high frequency, making it very easy to miss microscopic early-stage defects at critical nodes.

[0005] 2. Limited engineering adaptability and algorithm robustness of inverse problem solving. When using sparse measurement points for inverse deformation field solution, the number of discrete measurement points in actual engineering is far less than the order of the structure's inherent modes, resulting in extremely sparse data input. Traditional least-squares generalized inverse matrix transformation is prone to severe numerical divergence when dealing with such overdetermined and ill-conditioned matrices. At the same time, the inherent high-frequency random noise of laser measurement equipment in complex industrial environments (such as industrial dust, air refractive index disturbances, and random environmental excitation) is infinitely amplified by singular eigenvalues ​​in the ill-conditioned inverse problem solution. Furthermore, existing technologies lack a physical mechanism and a data-driven adaptive fusion correction mechanism, which cannot effectively eliminate nonlinear deformation distortion caused by material heterogeneity and welding residual stress, leading to a high false alarm rate for system deformation reconstruction during cross-seasonal operation.

[0006] 3. The early warning mechanism lacks sufficient ability to identify the evolution of deep mechanical damage at the underlying level. Existing early warning systems mostly rely on whether a single geometric displacement (such as deflection or absolute displacement at a single point) exceeds the limit for qualitative alarm. This mechanism based on shallow geometric displacement cannot effectively reveal the complex stress redistribution and potential energy accumulation inside large-span spatial steel structures. It cannot distinguish between the risk of overall instability and collapse, local member buckling and fracture damage, and the chain-like rapid damage propagation stage caused by bolt shearing / weld cracking. It lacks a multi-track parallel energy field hierarchical evolution early warning logic, which is prone to false alarms or missed alarms, and is difficult to meet the second-level timeliness requirements for emergency command and automated plan triggering under extreme conditions.

[0007] To address the aforementioned issues, there is an urgent need for an intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology. This system would enable efficient registration of multi-source heterogeneous laser data, robust real-time reconstruction of the full-field continuous three-dimensional deformation field driven by sparse points, and intelligent dynamic closed-loop early warning of multi-level structural damage states based on the spatiotemporal evolution trajectory of strain energy density. Summary of the Invention

[0008] Technical problems to be solved

[0009] To address the shortcomings of existing technologies, this invention provides an intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology, which solves the problems mentioned in the background section.

[0010] Technical solution

[0011] To achieve the above objectives, the present invention provides the following technical solution: an intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology. The system includes a dynamic laser ranging array, a spatial stiffness laser scanning unit, a spatiotemporal registration and matrix transformation module, a mechanical mode spatial reconstruction server, and an energy state hierarchical early warning terminal, wherein: The dynamic laser ranging array includes multiple array-type high-speed laser rangefinders fixed on an external reference axis, used to continuously acquire the one-dimensional dynamic displacement response time sequence of multiple characteristic force nodes on the surface of the steel structure at a first sampling frequency; The spatial stiffness laser scanning unit includes a three-dimensional solid-state lidar, which is used to periodically acquire full-field spatial point cloud data of the steel structure at a second sampling frequency, wherein the first sampling frequency is greater than the second sampling frequency. The spatiotemporal registration and matrix transformation module is used to unify the one-dimensional dynamic displacement response time series and the full-field spatial point cloud data into the structural mechanics finite element spatial coordinate system of the steel structure. The mechanical modal space reconstruction server internally constructs the inherent modal space matrix of the steel structure, the physical information modal solution model, and the residual neural network compensation model. The mechanical modal space reconstruction server uses the full-field spatial point cloud data to extract the global spatial modal vibration mode basis of each order, and uses the one-dimensional dynamic displacement response time series to solve the dynamic dynamic weight coefficients in real time. The theoretical deformation field is obtained through modal synthesis, and the residual neural network compensation model is used to perform nonlinear correction on the theoretical deformation field to reconstruct the continuous three-dimensional deformation displacement field of the entire steel structure. The energy state hierarchical early warning terminal is used to calculate the full-field strain energy density distribution corresponding to the continuous three-dimensional deformation displacement field, and to provide early warning of slippage or instability based on the energy mutation rate.

[0012] Preferably, the placement points of each array-type high-speed laser rangefinder in the dynamic laser ranging array are determined based on the sensor placement optimization algorithm of the steel structure mechanical model. The placement points are located at the displacement maxima nodes or strain energy sensitive nodes of the first few natural mode vibration modes of the steel structure.

[0013] Preferably, the spatial stiffness laser scanning unit further includes multiple laser cooperative targets deployed at key weld points of the steel structure. The three-dimensional solid-state lidar extracts the three-dimensional coordinates of the center of the laser cooperative targets by identifying the differences in the reflection intensity of the laser cooperative targets, and introduces them into the mechanical modal space reconstruction server as spatial physical boundary conditions.

[0014] Preferably, the method for constructing the intrinsic mode space matrix in the mechanical mode space reconstruction server includes the following steps: Sp1: Based on the geometric drawings of the steel structure, the initial finite element stiffness matrix and mass matrix are established, and the theoretical spatial mode shapes are obtained through eigenvalue calculation. Sp2: During system initialization, the spatial stiffness laser scanning unit is used to acquire two sets of full-field spatial point cloud data under no-load and calibrated load conditions, and the difference between them is calculated to obtain the initial deformation field. Sp3: Using the initial deformation field, the sensitivity matrix of each order of the theoretical spatial mode is corrected to obtain a global spatial mode base that conforms to the physical properties of the solid steel structure.

[0015] Preferably, the physical information modal solution model in the mechanical modal space reconstruction server is embedded with elastic mechanical continuity differential equation constraints. The mechanical modal space reconstruction server uses a generalized inverse matrix algorithm with regularization constraints to globally suppress the ill-conditioned singularity of the spatial matrix and laser measurement noise by introducing a boundary energy penalty term. It solves in real time the sparse extraction matrix based on the global spatial modal vibration mode base at the deployment points of the dynamic laser ranging array, and the topological extension equation composed of the dynamic dynamic weight coefficient vector to be solved.

[0016] Preferably, the mechanical modal space reconstruction server performs matrix multiplication on the global spatial modal vibration mode base and the solved dynamic dynamic weight coefficient vector to synthesize a theoretical deformation displacement field that is continuous in the spatial domain and satisfies the constraints of the elasticity continuity differential equation.

[0017] Preferably, the construction and correction process of the residual neural network compensation model in the mechanical modal space reconstruction server includes the following steps: Sp4: At the periodic moment when the spatial stiffness laser scanning unit synchronously uploads the full-field spatial point cloud data, the actual point cloud deformation field corresponding to that moment is matrix subtracted from the theoretical deformation displacement field to extract the spatial nonlinear deformation residual matrix. Sp5: Using the one-dimensional dynamic displacement response time series as the input feature and the spatial nonlinear deformation residual matrix as the training label, supervised learning training is performed on the residual neural network compensation model. Sp6: During the scanning window, the real-time one-dimensional dynamic displacement response time sequence is input into the trained residual neural network compensation model to obtain the real-time predicted residual, and the real-time predicted residual is superimposed on the real-time theoretical deformation displacement field to output the corrected continuous three-dimensional deformation displacement field.

[0018] Preferably, the energy state classification early warning terminal calculates the current strain energy, local strain energy density gradient, and strain energy change rate of the entire steel structure based on the corrected continuous three-dimensional deformation displacement field integral, and performs slip or instability early warning according to the following logic: when the overall strain energy exceeds the preset elastic limit energy threshold, it is determined that the steel structure has an overall instability risk; when the local strain energy density gradient exceeds the preset threshold, it is determined that the corresponding area has a local damage or local buckling risk; when the strain energy change rate shows a step change, it is determined that the steel structure is in the rapid damage propagation stage, and triggers the corresponding level of early warning signal.

[0019] Beneficial effects

[0020] This invention provides an intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology. It has the following beneficial effects: 1. This invention utilizes a dynamic laser ranging array for continuous high-frequency linear trajectory capture, while simultaneously employing a spatial stiffness laser scanning unit for periodic low-frequency full-scene surface scanning. This constructs a dual-track asynchronous parallel sampling mechanism. Combined with spatiotemporal registration and matrix transformation modules, it efficiently unifies the high-frequency discrete one-dimensional temporal sequence with the low-frequency macroscopic three-dimensional point cloud in the spatiotemporal dimension. This design fills the monitoring gap of three-dimensional lidar during the scanning window period, successfully achieving real-time and continuous reconstruction of the entire spatial domain continuous three-dimensional deformation displacement field without blind spots at a high refresh rate, even without the need for continuous radar operation. This greatly improves the accuracy and timeliness of capturing the initial microscopic defects of key nodes in large-span steel structures.

[0021] 2. This invention utilizes a generalized inverse matrix algorithm with regularization constraints. By introducing a boundary energy penalty term, it imposes physical energy conservation constraints on the solution, globally suppressing the ill-conditioned singularity of the spatial matrix. This effectively filters and cancels high-frequency random measurement noise caused by dust and air turbulence in industrial settings. Simultaneously, by constructing a residual neural network compensation model, it uses nonlinear residuals extracted from periodically point clouds for supervised iterative training. During the window period, the microscopic residuals predicted by AI are accurately superimposed onto the theoretical deformation field that satisfies the constraints of the differential equation of elasticity. This closed-loop mechanism of mechanical skeleton integration with AI microscopic compensation effectively eliminates nonlinear deformation distortion caused by material non-uniformity and welding residual stress, significantly reducing the false alarm rate of deformation reconstruction during long-term operation across seasons.

[0022] 3. This invention abandons the traditional qualitative early warning mode that relies solely on a single geometric deflection or displacement exceeding the limit. Instead, it transforms the reconstructed high-fidelity continuous three-dimensional deformation field into a fundamental mechanical energy field. The energy state hierarchical early warning terminal extracts three core energy indicators along the spatial and temporal dimensions—the overall current strain energy, the local strain energy density gradient, and the strain energy change rate—by solving the spatiotemporal integral and differential of the strain energy density across the entire field. Through a three-track parallel intelligent discrimination logic, the system can achieve a second-level response and accurately distinguish between: the overall instability and collapse risk caused by large-area structural degradation, the risk of local plastic bending or buckling damage caused by a sudden energy accumulation in specific web members or nodes, and the chain-like rapid damage propagation stage caused by bolt fracture or weld cracking. This provides high-definition three-dimensional topological highlight rendering and hierarchical materialized alarm output, perfectly meeting the high-time-efficiency triggering requirements of emergency command and automated emergency plans under extreme disaster conditions. Attached Figure Description

[0023] Figure 1 This is a diagram of the overall system architecture of the present invention; Figure 2 This is a flowchart illustrating the overall system operation of the present invention. Figure 3 This is the system sensor status monitoring interface of the present invention; Figure 4 This is the system early warning log display interface of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1: Please see Figure 1 and Figure 2 As shown, a steel structure deformation intelligent early warning and monitoring system based on laser measuring instrument technology is mainly composed of the following five core units in terms of physical space and logical architecture: dynamic laser ranging array, spatial stiffness laser scanning unit, spatiotemporal registration and matrix transformation module, mechanical modal space reconstruction server, and energy state hierarchical early warning terminal.

[0026] The dynamic laser ranging array consists of multiple independently deployed high-speed laser rangefinders. These rangefinders are fixed to an external reference axis away from the main body of the steel structure to be measured by bolts or rigid bases (such as a rigid shear wall of the opposite building or a specially cast independent rigid observation column away from the vibration source) to ensure that the measurement reference itself is not subject to displacement coupling interference due to the dynamic vibration of the steel structure. The array is responsible for continuously emitting collimated laser beams at a high-frequency first sampling frequency (e.g., 20Hz to 50Hz) to accurately capture the instantaneous one-dimensional axial dynamic displacement response time sequence of multiple pre-set key characteristic force nodes on the surface of the steel structure.

[0027] The spatial stiffness laser scanning unit includes at least one large field-of-view three-dimensional solid-state lidar, which is deployed on the opposite side of a fixed field of view that can completely cover the stress surface of the steel structure under test. This unit is used to perform a large-scale spatial point cloud scan of the entire steel structure at a lower second sampling frequency (e.g., triggered periodically every 24 hours, or triggered by system commands when encountering extreme external load changes such as typhoons or strong earthquakes), thereby obtaining full-field spatial point cloud data that reflects the macroscopic full-field spatial contour characteristics of the steel structure. Among them, due to the different requirements of timeliness and data volume for dynamic monitoring and macroscopic scanning, the first sampling frequency is much higher than the second sampling frequency.

[0028] The spatiotemporal registration and matrix transformation module, the mechanical modal space reconstruction server, and the energy state classification and early warning terminal are logically embedded or deployed in the backend computing control center or edge computing gateway. These three components establish bidirectional data interaction with the frontend dynamic laser ranging array and spatial stiffness laser scanning unit through data transmission links based on industrial-grade fiber optics, local area networks, or wireless communication protocols such as MQTT and 5G. The spatiotemporal registration and matrix transformation module is responsible for receiving asynchronous, multi-source laser data uploaded from the frontend and using spatial physical mapping relationships to uniformly project the one-dimensional time sequence of the rangefinder and the three-dimensional point cloud of the radar onto the pre-set structural mechanics finite element spatial coordinate system of the steel structure. The mechanical modal space reconstruction server serves as the computing core of the system, responsible for using the registered data to run its internally integrated modal matrix and residual network compensation model to reconstruct a continuous and high-precision three-dimensional deformation displacement field in the spatial domain. Finally, the energy state classification and early warning terminal is responsible for assessing physical stiffness degradation and outputting dynamic alarm commands.

[0029] This monitoring system overcomes the shortcomings of traditional monitoring technologies, which can either only monitor isolated points or cannot monitor continuously in real time. It constructs a data interaction and collaborative working mechanism based on physical mechanisms, asymmetric dual-track sampling, and AI residual correction. The entire system operates in a dual-track asynchronous parallel state: on the one hand, the dynamic laser ranging array continuously tracks the instantaneous dynamic changes of sparse characteristic force nodes on a millisecond time scale; on the other hand, the spatial stiffness laser scanning unit captures the macroscopic spatial stiffness profile of the structure on a large-cycle time scale.

[0030] Once the data is aggregated at the mechanical modal space reconstruction server, the system does not directly use ordinary mathematical black-box algorithms for blind interpolation. Instead, it first relies on the extracted global spatial modal vibration mode base (which strictly follows the constraints of the elasticity continuity differential equation in the spatial domain) and uses a generalized inverse matrix algorithm with regularization constraints to solve the high-frequency, sparse one-dimensional dynamic displacement response time series in real time into dynamic dynamic weight coefficient vectors of each mode. Then, it synthesizes the theoretical deformation displacement field through matrix multiplication. On this basis, the system further calls the residual neural network compensation model to perform real-time superposition correction on nonlinear micro residuals that cannot be exhausted by mechanical formulas due to material inhomogeneity or welding residual stress.

[0031] Through this collaborative solution chain that uses "line" data to drive "surface" reconstruction and integrates mechanical trends with AI micro-compensation, the system successfully achieves real-time reconstruction and output of a continuous three-dimensional deformation displacement field that fully conforms to the physical continuity law of elasticity, even during the scanning window when the lidar is not in use. Ultimately, these high-precision deformation fields are pushed to the energy state classification early warning terminal in real time, where they are converted into the structural energy evolution trajectory (i.e., overall strain energy, local strain energy density gradient, and strain energy change rate), thus completing the intelligent closed-loop monitoring from geometric displacement capture to deep mechanical instability risk assessment.

[0032] Please see Figure 3 As shown, the system provides a sensor status monitoring interface. The interface includes a sensor status display area to show the real-time operating status information of multiple laser rangefinders in the dynamic laser ranging array, the 3D solid-state lidar in the spatial stiffness laser scanning unit, and multiple laser cooperative targets. It also includes a device location display area to show the spatial coordinates of each sensor device within the steel structure monitoring scenario. Specifically, the dynamic laser ranging array, as the system's high-frequency dynamic data source, is physically implemented using multiple independently operating phase-type high-speed laser rangefinders. This is to continuously acquire the one-dimensional dynamic displacement response of multiple characteristic force nodes on the steel structure surface. In terms of timing, each high-speed laser rangefinder integrates a high-frequency laser emitting diode, an avalanche photodetector, and a high-speed digital signal processor. The specific sampling process is as follows: The high-speed laser rangefinder emits a collimated continuous laser beam, modulated by a high-frequency amplitude, at a preset first sampling frequency (set to 50Hz, i.e., sampling once every 20 milliseconds) towards a specific characteristic stress node on the steel structure surface. Upon reaching the characteristic stress node (or the retroreflective patch attached to the node), the laser beam undergoes diffuse reflection. The reflected light signal is received by the avalanche photodetector. The internal high-speed digital signal processor measures the phase difference between the emitted and received signals. Combined with the wavelength of the modulated light wave The absolute one-dimensional axial distance between the rangefinder and the force-bearing node at the current moment is calculated using the following formula. : ; in, Given the integer wavelength of the light wave along its transmission path, the rangefinder continuously performs the above calculations to output a one-dimensional dynamic displacement response time series that evolves continuously with the time axis. ,in This is the reference axial distance of the structure under initial no-load conditions.

[0033] To ensure that the displacement response timing is not mixed with the displacement noise of the monitoring equipment itself, each high-speed laser rangefinder is rigidly fixed to the external reference axis (such as the rigid foundation on the opposite side of the station building, an independent reinforced concrete observation column, or the main stable wall of an adjacent building that is not affected by the load of the steel structure itself) through pre-embedded steel plates and anti-vibration bolts, so that the reference coordinates of the rangefinder are completely stationary in space.

[0034] Furthermore, the specific locations of the instruments in the dynamic laser ranging array are not randomly distributed, but determined by a sensor placement optimization algorithm (such as the Effective Independence Method, Efi) based on the steel structure mechanics model. The core physical logic is to maximize the capture of the overall dynamic characteristics of the structure by sparse locations. In practice, a finite element model of the steel structure is first established in the computer, and the inherent mode shape matrices of the first few orders (e.g., the first 1 to 5 dominant modes) are obtained through eigenvalue calculation. Each column in the modal shape matrix represents the spatial deformation profile of the structure at a certain natural frequency. To ensure the highest signal-to-noise ratio of the calculated dynamic weighting coefficients when the number of measurement points is limited, the algorithm constructs the modal center matrix or calculates the basis orthogonal projection of the target modal shape matrix, and calculates the effective independence index vector of each candidate node using the following formula. : ; in, This means taking the diagonal elements of the matrix and solving for... The value of each element in the vector represents the contribution of that node to maintaining the linear independence of the mode shapes. The system selects... The largest value Each node (i.e., the displacement maximum node located in the dominant spatial mode or the bending strain maximum node most sensitive to strain energy) is used as the final characteristic force node. Then, the laser axis of each high-speed laser rangefinder is aligned with the normal direction of these characteristic force nodes, thereby continuously capturing the dynamic deformation response of the structure with the highest geometric sensitivity.

[0035] The spatial stiffness laser scanning unit includes at least one large-field-of-view 3D solid-state lidar (LiDAR). To periodically acquire high-density full-field spatial cloud data of the steel structure within a low-frequency, long-cycle period at a second sampling frequency (e.g., once every 24 hours), this 3D solid-state lidar employs a microelectromechanical system (MEMS) micromirror scanning architecture internally. The internal laser pulse generator produces nanometer-scale high-power pulsed lasers. Through the two-dimensional coordinated deflection of two one-dimensional MEMS micromirrors (one fast, one slow), the laser beam is controlled to perform a full-field scan of the entire steel structure surface in a very short time according to a preset matrix grid, without any mechanical rotation. The lidar measures the round-trip time of each pulse using the time-of-flight (ToF) method and records the deflection angle of the micromirrors at the corresponding moments. This allows for real-time calculation and output of a massive set of grid point clouds with absolute spatial coordinates, which is the full-field spatial point cloud data.

[0036] To eliminate system geometric errors caused by minor deformations of the radar casing due to environmental temperature drift and wind vibration during long-term monitoring, multiple laser cooperation targets with known spatial physical locations are pre-welded or fixed by strong magnetic adsorption in rigid areas where microscopic local cracking does not occur, such as key welded ball joints of the steel roof, the sides of high-strength bolt groups, and the middle of the compression main chord. The surfaces of the laser cooperation targets are coated with specially made high retroreflectivity materials (such as micro glass bead-type or micro prism-type reflective films), making their reflectivity much higher than that of ordinary steel surfaces.

[0037] When a 3D solid-state lidar performs a matrix-style full-field spatial scan of a steel structure, its laser beam sweeps across the cooperative target area. The energy of the reflected light pulse received by the radar detector will produce a steep jump (i.e., the reflection intensity will produce obvious local peaks). The image processing device in the spatial stiffness laser scanning unit utilizes this physical characteristic and uses an intensity threshold segmentation algorithm to extract the point cloud data of the cooperative target from the point cloud of ordinary structural steel components. Furthermore, using a 3D sphere or 3D plane fitting algorithm, the 3D world coordinates of the geometric center of each laser cooperative target are accurately calculated. The mechanical modal space reconstruction server uses these high-precision real-time center coordinates as "invariants." On the one hand, they are used to calibrate and eliminate the spatial installation displacement drift of the radar itself. On the other hand, they are directly used as initial boundary conditions and elastic constraints input into the physical information modal solution model, locking the physical spatial boundary of the structure and preventing numerical distortion of the subsequent continuous deformation field in the edge region.

[0038] Since the dynamic laser ranging array acquires one-dimensional axial relative displacement and uses high-frequency non-periodic time-series signals, while the spatial stiffness laser scanning unit acquires three-dimensional absolute point clouds of a large-span space and uses low-frequency periodic still frames, there are serious asymmetries between the two in both time and space. The spatiotemporal registration and matrix transformation module is used to complete the fusion and normalization of these two types of multi-source heterogeneous data. The specific operation process is as follows: First, a time alignment process is performed: Each high-speed laser rangefinder and solid-state lidar in the dynamic laser ranging array has a built-in or external Network Time Protocol (NTP) client, which maintains high-precision synchronization with the GPS / BeiDou satellite clock source embedded in the system via gigabit Ethernet. The synchronization accuracy reaches the microsecond level. When the spatiotemporal registration and matrix transformation module receives one-dimensional dynamic displacement data and full-field point cloud data, it first reads the absolute timestamp built into its hardware and uses a time linear interpolation algorithm to align the asynchronously acquired data to a unified time axis coordinate. The spatial matrix transformation process is then performed: At the initial deployment, the system establishes a unified spatial rectangular coordinate system, namely the structural mechanics finite element spatial coordinate system, with the center of the steel structure in the design drawings as the origin. (Indicated), the point cloud coordinates output by the spatial stiffness laser scanning unit are usually in the radar's own independent instrument coordinate system (indicated by...). (Indicated by) the measured three-dimensional coordinates of the extracted laser cooperative target in the radar coordinate system. And the design or calibration of the target's three-dimensional coordinates in the finite element coordinate system. By solving the classic absolute orientation problem, a system is constructed using... Rotation matrix and Translation vector The spatial affine transformation matrix formed follows the physical transformation equations as follows: ; The spatiotemporal registration and matrix transformation module uses this transformation matrix to perform global spatial projection on the input radar full-field point cloud data, so that each point cloud feature corresponds to a physical node of the finite element mesh. Similarly, for the dynamic laser ranging array, since its installation reference axis is stationary relative to the finite element spatial coordinate system, the module uses the world coordinates of the rangefinder's installation space and the spatial direction cosine vector of the collimated laser beam. The one-dimensional axial relative displacement sequence By using scalar projection, the displacement components of the three-dimensional dynamic response in the finite element spatial coordinate system are converted, realizing the complete unification of multi-source heterogeneous laser data in the spatiotemporal dimension, and transmitting it to the backend mechanical model space reconstruction server in real time.

[0039] As the core computing node of the system, the mechanical modal space reconstruction service constructs an intrinsic modal space matrix that forms the mechanical foundation for subsequent point-to-surface spatial reconstruction. Due to the unavoidable problems encountered in the actual construction of large-span steel structures, such as welding residual stress, uneven preload of high-strength bolts, microscopic material inhomogeneity, and discrepancies between the actual stiffness and ideal state of support boundary conditions, purely theoretical models generated directly by computers cannot accurately reflect the dynamic deformation characteristics of the actual structure. Therefore, this system uses the mechanical modal space reconstruction server to execute steps, specifically the following pipelined physical and computational operations (Sp1 to Sp3), to construct an intrinsic modal space matrix that closely matches the actual physical properties of the actual steel structure. Sp1: Establishing the Initial Matrix and Eigenvalue Solving (Theoretical Finite Element Modeling): During the system initialization and startup phase, the system reads the as-built geometric drawings of the solid steel structure from the computer-aided design (CAD) or building information model (BIM), obtains the physical cross-sectional geometric features of all load-bearing members, the three-dimensional spatial topological connection relationships of nodes, and the constitutive parameters of the steel (such as the elastic modulus). shear modulus and material density The finite element discretization module in the mechanical modal space reconstruction server, based on the variational principle of energy functional theory, discretizes the large-span continuous steel structure grid into a finite element spatial mesh composed of a large number of beam elements, shell elements, and nodes. By performing global coordinate transformation and topological superposition on the spatial stiffness matrices of each element, the global theoretical initial finite element stiffness matrix of the system is constructed in memory. Similarly, by projecting and superimposing the masses of each unit in a centralized or uniform manner, the global theoretical initial finite element mass matrix of the system is constructed. Subsequently, the model solution module, based on the classical mechanical equations of undamped free vibration, performs eigenvalue calculations on the aforementioned initial matrix. Its physical governing equations are expressed as follows: ; The server reverse-engineers the system's front-end by using a multi-step subspace iteration method or the Lanzos algorithm for solving large sparse matrices eigenvalues. The natural angular frequencies of the first order (the lowest dominant model with the highest proportion of structural energy, such as the first 1 to 10 orders). And correspondingly, the theoretical spatial mode shape vectors of each order, which reflect the spatial deformation trend of the structure at a specific frequency. By assembling the calculated mode shape vectors column by column, the mode shape matrix under the initial theoretical state can be obtained. The theoretical modal foundation has been completed.

[0040] Sp2: Acquisition of Differential Deformation Field (Point Cloud Measured Deformation Calibration): Its core physical logic is to acquire the actual geometric deformation response of the solid steel structure under known state changes through the spatial stiffness laser scanning unit. The specific operation process is as follows: During the initialization window when the system is first powered on, the three-dimensional solid-state lidar in the spatial stiffness laser scanning unit is first controlled to perform a full-field holographic matrix scan on the steel structure under the reference state of no live load, no wind load, and constant temperature, to obtain the first set of reference spatial point cloud sets. Subsequently, a set of known physical calibration loads (e.g., precise construction counterweight suspension under the space frame, or parking a calibration vehicle of a specific tonnage on the main beam of the bridge) are applied to the designated characteristic stress grid area of ​​the steel structure. After the structure reaches static equilibrium, the three-dimensional solid-state lidar is controlled again to perform a full-field spatial scan with the same grid density to obtain a second set of load state spatial point cloud sets. After receiving these two sets of point cloud data, the mechanical modal space reconstruction server first calls the mesh alignment module, using the three-dimensional world coordinates of the laser cooperative target as rigid constraint points, to perform the mesh alignment. and Accurate spatial rigid body registration is performed. After registration, the nearest point iteration (ICP) algorithm based on mesh neighborhood search or the spatial difference voxel projection algorithm is used to calculate the 3D triaxial displacement scalar difference of each voxel mesh between the two sets of point clouds. The calculation is as follows: ; Therefore, the actual geometric profile variation data of the solid steel structure under standard load can be directly extracted and derived in the entire physical space, that is, the initial measured deformation field.

[0041] Sp3: Sensitivity Matrix Correction (Modal Basis Measured Tuning): Its core operation is to perform inverse closed-loop correction on the theoretical spatial mode shapes through the initial measured deformation field, transforming them from an "idealized model" into a "digital twin entity." The specific solution process is as follows: Since the maximum error between theory and reality usually comes from the uncertainty of support boundary conditions and the local stiffness degradation of connection nodes, the server uses the rotational stiffness and translational stiffness of each support in the finite element model as the design variable parameter vector to be corrected. The mechanical modal space reconstruction server first converts the stiffness matrix in Sp1. Regarding design variables Taylor expansion is used to calculate the first-order partial derivative matrix of deformation with respect to parameter changes, thus constructing the spatial deformation sensitivity matrix [S]: ; The physical meaning of this sensitivity matrix [S] is that it quantitatively reflects how much geometric displacement change the theoretical model will produce at the radar sampling point for every unit of fine adjustment in the stiffness of the support or node. Next, the server uses the initial measured deformation field obtained from Sp2. As the target, a weighted least squares optimization objective function is constructed to solve for the parameter correction increment that minimizes the sum of squares of the differences between the theoretically calculated deformation and the radar-measured deformation. : ; in, The server uses the calculated correction increments to update the design variables, resulting in a stiffness matrix that conforms to the physical properties of the solid structure, which is the measurement point weight coefficient matrix. Finally, the server utilizes the corrected stiffness matrix. and quality matrix The eigenvalues ​​are recalculated, and the resulting mode shape matrix is ​​the final global spatial mode shape basis that closely matches the physical properties of the solid steel structure. The base, which serves as a completely orthogonal physical and mechanical coordinate system in the spatial domain, is locked within the server's intrinsic modal space matrix to support the subsequent high-frequency dynamic reconfiguration chain.

[0042] After the intrinsic modal space matrix is ​​constructed, the system enters the real-time online monitoring stage. The mechanical modal space reconstruction server uses its embedded physical information modal solution model to start performing high-frequency one-dimensional point data to three-dimensional full-field data continuous spatial reconstruction during the window period of lidar scanning. The core of this process is to ensure the physical rationality of the solution results and the robustness against physical noise under sparse measurement point conditions.

[0043] In practical engineering applications, the number of high-speed laser rangefinders included in a dynamic laser ranging array (i.e., the number of spatially discrete measuring points) Limited by installation space and equipment costs, the number of modalities is often much smaller than the order (i.e., modal number) of the global spatial modal prototype basis for system construction. ),Right now Meanwhile, when high-speed laser rangefinders perform high-frequency data acquisition, they are inevitably subject to physical noise interference caused by environmental dust, air refractive index disturbances, wind load vibrations, etc. If traditional structural mechanics inverse transformation is directly used, the one-dimensional dynamic displacement response time series of characteristic force nodes acquired by the dynamic laser ranging array and spatiotemporally registered will be affected. , and the dynamic weight coefficient vector to be solved In mathematics, this forms a typical highly underdetermined and ill-conditioned spatial topological extension equation: ; in, It manifests as a The non-positive definite sparse extraction matrix of order is obtained by using a global space modal prototype basis. In the middle, only extract the content related to... The rectangular matrix formed by the row vectors corresponding to the actual deployment points of each high-speed laser rangefinder exhibits severe ill-conditioned singularity due to its extremely high condition number. If the generalized inverse matrix is ​​directly calculated, the tiny random high-frequency noise contained in the laser ranging time series will be infinitely amplified by the singular eigenvalues, leading to an imbalance in the calculated weight coefficient vector. The numerical values ​​diverge drastically, and the reconstructed deformation field will exhibit tearing or distortions that do not conform to the laws of mechanics.

[0044] To overcome this technical bottleneck, this invention introduces the elasticity continuity differential equation constraint into the physical information modal solution model, and introduces a generalized inverse matrix algorithm with regularization constraints (using the Tikhonov regularization architecture). This algorithm, by introducing a boundary energy penalty term, transforms the originally unsolvable ill-conditioned topological extension equation into a functional extremum optimization problem under physical boundary energy constraints. Its objective optimization governing equation is expressed as: ; In the above formula, The L2 norm is represented by the first term of the formula, which is the sum of squared residuals used to constrain the solution result to approximate the dynamic displacement measured by the laser rangefinder; the second term... This refers to the introduced boundary energy penalty term, where It is the Tikhonov regularization matrix (using the identity matrix or structural stiffness operator matrix). As a regularization hyperparameter greater than zero, the physical significance of this boundary energy penalty term in mechanics is that it imposes a priori physical constraints on the deformation energy of the structure, limits the energy upper limit of each mode (especially the ultra-high-order modes that are highly susceptible to noise excitation) participating in the synthesis, and fundamentally suppresses the amplification effect of high-frequency noise in the ill-conditioned inverse problem.

[0045] The mechanical modal space reconstruction server at each time step (That is, every 20 milliseconds), by finding the variational extremum of the above objective optimization equation, the generalized inverse matrix operator with regularization constraints is derived and solved. Its final closed-loop analytical solution equation is expressed as: ; During operation, the server reads the input in real time. Multiply by a pre-computed generalized inverse matrix operator with regularization constraints in memory. This allows for the extremely stable and efficient inverse solution of the dynamic weight coefficient vectors of each spatial mode shape under the constraint of physical energy conservation at the current moment. The algorithm consumes very little computing power when running online, effectively ensuring the robustness and high real-time performance of the system's data processing.

[0046] Successfully retrieved the dynamic weight coefficient vector in real time Subsequently, the mechanical modal space reconstruction server calls its internal modal space synthesis module to perform the macroscopic deformation field restoration operation in the above steps. The specific operation process is as follows: the server uses the global spatial modal vibration mode bases of all orders, which are solidified in Sp3 within the inherent modal space matrix and completely conform to the physical properties of the solid steel structure, to perform the restoration operation. (It contains a spatially continuous finite element deformation base vector), and the dynamic weight coefficient vector just calculated at the current time step. The physical superposition equation for high-concurrency matrix product synthesis is expressed as follows: ; At the specific calculation level, the modal base It contains the spatial coordinate variation components of tens of thousands of finite element mesh nodes across the entire field, and is compared with a vector containing instantaneous amplitude information. Multiplication instantly maps and extends the dynamic features of the "line" obtained from sparse measurement points to the spatial features of the "surface" of all nodes in the field.

[0047] Due to its role as a synthetic substrate Strictly derived from the stiffness and mass matrices of the finite element method, it naturally satisfies the square law of elasticity, the geometric continuity differential equation, and the boundary constraints. Therefore, the synthesized theoretical deformation displacement field... It exhibits a completely continuous, smooth, and stiffness-distribution-compliant three-dimensional geometric surface across the entire spatial domain. It will never exhibit any non-physical abrupt changes or tearing phenomena that often occur in mathematical interpolation methods (such as spline interpolation and Kriging interpolation). As a result, the system can perfectly complete the high-frequency continuous reconstruction of the theoretical deformation displacement field during the radar scanning window without the need for continuous scanning by the lidar, thanks to the time-series drive provided by the dynamic ranging array.

[0048] To overcome the engineering challenges of traditional pure mechanical reconstruction, which cannot fully fit steel structures due to non-uniform material heterogeneity, initial defects in welding, and nonlinear elastic distortion caused by microscopic slippage at nodes, this invention specifically constructs and embeds a residual neural network compensation model within the mechanical modal space reconstruction server. This model works in parallel and collaboratively with the aforementioned physical information modal solution model. By executing the following steps, the model is trained and corrected online using periodically obtained full-field high-density laser scanning point cloud data as real labels, thereby providing real-time nonlinear deformation correction during the scanning window when the radar is turned off.

[0049] Sp4: Extraction of Spatial Nonlinear Deformation Residual Matrix: The construction and correction of the entire residual model begins at this step. Its core purpose is to accurately extract the microscopic deformation error that cannot be quantitatively analyzed by the mechanical finite element model. The specific operation process is as follows: During the daily operation of the system, the three-dimensional solid-state lidar in the spatial stiffness laser scanning unit periodically triggers the acquisition of the full-field spatial point cloud according to the second sampling frequency. At the absolute synchronization period when the lidar uploads the latest frame of the full-field spatial point cloud data... The mechanical modal space reconstruction server first controls its internal spatiotemporal registration and matrix transformation module to project and transform the point cloud of that frame into a unified structural mechanics finite element space coordinate system. Then, through 3D mesh surface reconstruction, it transforms the point cloud into a true point cloud deformation field with continuous spatial characteristics, reflecting the actual load profile of the steel structure at that moment, denoted in matrix form as follows: .

[0050] At the same time, the server uses the physical information modal solution model to read the same timestamp. Discrete node displacement data acquired by a dynamic laser ranging array are used to solve for the dynamic mechanical weight coefficient vector using a regularized generalized inverse matrix algorithm. This vector is then multiplied by the global spatial mode shape basis to synthesize the continuous theoretical deformation displacement field in the spatial domain at that moment online. .

[0051] Subsequently, the server's residual extraction module performs matrix subtraction on the two sets of spatial field matrices, which are in the same coordinate system and have the same timestamp, using all matrix elements. The difference extraction equation is expressed as follows: ; The resulting difference matrix is ​​the spatial nonlinear deformation residual matrix. The residual matrix physically represents all the microscopic local nonlinear responses accumulated in the manufacturing, installation and long-term fatigue process of the solid steel structure that cannot be fully expressed by the differential equations of traditional continuum mechanics, thus pointing out the target for the subsequent learning of the neural network.

[0052] Sp5: Efficient Supervised Training of Residual Neural Network Compensation Model: Its core operation lies in establishing a nonlinear high-dimensional mapping network between "discrete point temporal features" and "continuous surface residual features across the entire field." The specific implementation process is as follows: In this embodiment, the embedded residual neural network compensation model adopts a spatial-temporal fusion architecture combining deep convolutional neural networks (CNN) and long short-term memory networks (LSTM). The server uses the one-dimensional dynamic displacement response time series of each feature force node, after spatiotemporal registration, acquired by the dynamic laser ranging array within a sliding window time before the current synchronization sampling period. As input features to the network, these features undergo standardization and smoothing to improve the network's generalization ability. Simultaneously, the server uses the spatial nonlinear deformation residual matrix, calculated in Sp4 and corresponding to the current time step. The discretized scalar values ​​of the full-field spatial grid are used as the unique supervised training labels corresponding to the input features.

[0053] Within a separate training thread on the server, the system constructs a loss function based on mean squared error (MSE) to quantitatively evaluate the network's current prediction error: ; in, The nonlinear forward propagation mapping function representing a neural network. and These are the connection weight matrix and bias vector of the network, respectively. The server uses a stochastic gradient descent algorithm with adaptive moment estimation (Adam optimizer) to calculate the partial derivatives of the loss function with respect to the weights and biases online, backpropagate along the opposite direction of the gradient, and dynamically iteratively update the internal network parameters of the neural network.

[0054] When the loss function converges to below the preset minimum convergence threshold, or when the number of iterations reaches the set upper limit, the server immediately determines that the training of the residual neural network in this cycle is complete. At this time, the network has deeply mastered the inherent physical coupling law between the dynamic vibration mode of the sparse feature force nodes and the microscopic local nonlinear residual variation of the overall structure.

[0055] Sp6: Real-time Dual-Drive Fusion Correction During the Scanning Window: Finally, during the long scanning window when the LiDAR is off and no point cloud scanning is performed, the system fully initiates this step to perform real-time dual-drive fusion correction calculations. The entire operation process of its high-frequency online operation is as follows: The dynamic laser ranging array outputs the real-time one-dimensional dynamic displacement response time sequence of the characteristic force nodes at a high frequency and continuously at a first sampling frequency of 50Hz. After entering the server, this time sequence is synchronously split and input into two parallel calculation channels in terms of timing logic: In the first channel, the physical information modal solution model uses a generalized inverse matrix algorithm with regularization constraints to calculate the dynamic weights based on the ranging time series, and then multiplies them with the intrinsic modal basis matrix to output the macroscopic theoretical deformation displacement field at that instant at a high frequency. ; In the second channel, the aforementioned real-time dynamic displacement response timing is synchronously fed into the trained residual neural network compensation model. The network performs pure forward propagation calculations, instantly inferring and outputting the instantaneous spatial microscopic predicted residual across tens of thousands of finite element mesh nodes throughout the field. Subsequently, the server's data fusion and correction module performs physical fusion of the theoretical deformation displacement field output from the first channel and the real-time prediction residual inferred from the second channel, using spatially corresponding nodes. The correction and synthesis equation is expressed as follows: ; The spatial superposition of the two perfectly achieves the complementary advantages of mechanical foundation to ensure that the physical framework does not collapse and AI residual compensation to ensure that the microscopic accuracy is not distorted.

[0056] The corrected continuous three-dimensional deformation displacement field output Not only does the overall deformation trend strictly obey the boundary conditions and stiffness continuity distribution of elasticity mechanics, but it also incorporates the disturbances caused by the initial defects of the structural entity and nonlinear stress concentration at the local micro level, resulting in extremely high geometric fidelity and physical reliability. This high-precision form is continuously pushed to the energy state classification early warning terminal at a high refresh rate of 50Hz, providing an extremely reliable digital twin foundation for subsequent in-depth mechanical instability state evolution assessment.

[0057] As the core of the entire monitoring system for safety decision-making and visual presentation, the energy state classification early warning terminal is responsible for transforming geometric deformation data into underlying mechanical damage state and outputting visual early warning commands. Simple geometric displacement deformation (such as deflection at a certain point) cannot fully reveal the stress redistribution and instability potential inside a large-span spatial steel structure. Therefore, the terminal achieves the physical leap from geometric displacement capture to energy state classification early warning by executing the following complete set of operation procedures.

[0058] First, the terminal performs a physical mapping from the full-field deformation field to the energy field. The energy state hierarchical early warning terminal receives in real time a continuous three-dimensional deformation displacement field, corrected by a neural network, pushed at a high frequency of 50Hz by the mechanical modal space reconstruction server. The specific operation process is as follows: The mechanical analysis module inside the terminal, based on the continuous three-dimensional deformation displacement field, uses geometric equations to analyze the spatial coordinates... Calculate the first-order spatial partial derivatives and the spatial six-component strain tensor at any infinitesimal element in the entire field of the steel structure. Subsequently, according to the generalized Hooke's law for steel in the elastic stage (based on the elastic modulus) Compared to Poisson Constructed isotropic elasticity matrix ), calculate the corresponding stress tensor Based on the energy functional principle of elasticity, the strain energy density at each spatial micro-element in the entire field is calculated. Its physical governing equations are expressed as follows: ; strain energy density This reflects the elastic energy density accumulated within the structural micro-element due to geometric distortion caused by external loads. Based on this, the final step utilizes the spatial three-dimensional Gaussian integral method to calculate the physical domain occupied by the entire steel structure. By performing global three-dimensional spatial integration, the current strain energy of the entire steel structure at the current moment can be calculated. : ; Through the above calculation of the mechanical matrix driven by point cloud and ranging, the terminal completely materializes the abstract spatial geometric surface into an energy index with a clear mechanical evolution trajectory, laying a solid data foundation for the subsequent three-track parallel early warning logic.

[0059] Please see Figure 4 As shown, it provides a system warning log display interface, which includes a warning statistics area and a warning log display area. The warning statistics area is used to classify and count the number of normal events, warning events, and dangerous events. The warning log display area is used to display various alarm records generated by the monitoring system, including warning level, anomaly description information, occurrence time, and corresponding anomaly location. Specifically, the energy state classification warning terminal relies on the calculated current strain energy... Based on its spatial and temporal differential characteristics, and strictly following the following three parallel physical logics, the slippage, plastic damage, local buckling, and overall instability of steel structures are classified and materialized for output: Logic 1: Overall instability risk assessment and alarm based on overall strain energy: During operation, the terminal continuously integrates the current strain energy obtained from global spatial integration. With respect to the system's intrinsic preset elastic limit energy threshold Perform online comparison; this threshold During the system calibration phase, through finite element destructive full-process tracking analysis, the total critical strain energy that the large-span steel structure can withstand when entering the global elastic critical buckling state was calculated. This was done when monitoring... If it is determined that the main steel structure has exceeded the physical limit of elastic bearing capacity due to severe overloading or large-scale degradation of structural stiffness, and there is an extreme high risk of overall instability and collapse, the terminal will immediately issue the highest level, namely the first-level red alarm: the system will instantly trigger multiple high-decibel electric sound and light alarms at the monitoring site through the relay output module, and at the same time send a forced flashing interception pop-up to the owner's emergency command center terminal through the 5G network, and automatically activate the automated emergency plan (such as closing the passage door and issuing personnel evacuation broadcast signals).

[0060] Logic 2: Local damage or buckling risk assessment and alarm based on local strain energy density gradient: To detect the hidden defects of sudden local brittle fracture or local buckling of a web member or welded ball joint in a large-span space frame due to local compression, the terminal uses a spatial differential operator to calculate the maximum directional derivative of the strain energy density in the three-dimensional spatial direction, thus constructing the local strain energy density gradient. When the structure normally conforms to overall deformation, this gradient exhibits a continuous and smooth variation; however, once a member buckles locally due to a defect or the weld cracks locally, the energy balance between that member and its adjacent members is disrupted, and the local strain energy density gradient at their interface changes. An extremely steep spatial jump will occur, with the value rapidly exceeding the preset local gradient threshold. At this point, the terminal determines that there is a risk of local damage or buckling in the corresponding area and immediately issues a Level 2 yellow alert. In terms of physical presentation, the terminal renders a three-dimensional topological mesh model of the steel structure in real time on a three-dimensional digital twin display screen and uses color mapping technology to visualize the energy field. Once this alarm is triggered, the system will automatically lock the spatial coordinates of the abnormal local member or node and render the damaged member in the three-dimensional model with a bright yellow or red high-frequency flashing, prompting maintenance personnel to conduct targeted and precise inspection.

[0061] Logic 3: Determination and Alarm of Rapid Damage Propagation Stage Based on Sudden Changes in Strain Energy Rate of Change: In addition to monitoring in the spatial domain, the terminal also performs real-time differentiation of energy evolution along the time axis, and calculates the strain energy change rate at the current moment through differential calculation. When a steel structure encounters strong wind loads, snow loads, or uneven foundation settlement, and internal high-strength bolts undergo continuous shearing or tensile welds crack, the structural stiffness will experience a step-like sudden drop within microseconds. This will inevitably manifest as a step-like instantaneous change in the strain energy change rate on the energy axis (i.e., a pulse peak appears in the first derivative of time). When the terminal detects such a step-like change in the strain energy change rate within a set number of consecutive time steps, it determines that the steel structure is no longer in a stable, slow deformation period, but in an extremely dangerous rapid damage propagation stage (i.e., internal structural damage is spreading like a domino effect). The terminal then triggers a corresponding level of emergency continuous siren warning signal, notifying on-site workers that the structure is likely to experience chain-like failure within seconds, and that the current construction work must be immediately terminated or emergency evacuation measures must be taken on-site, thereby ensuring the absolute safety of personnel and property to the greatest extent possible.

[0062] In summary, this specific implementation presents a digital twin early warning and monitoring system that integrates structural mechanics physical mechanisms, asymmetric multi-source laser precision measurement, and data-driven deep learning. By physically solidifying the hardware architecture and eliminating vibration interference, the system ensures the purity and reliability of the data source. In the core calculation chain, the system abandons the black-box drawbacks of traditional pure mathematical interpolation algorithms, pioneering a strategy based on physical mechanisms. It utilizes the global spatial modal vibration modes of each order after sensitivity matrix correction as the framework, and overcomes the technical bottleneck of ill-conditioned matrix divergence caused by sparse measurement points and laser noise in practical engineering through a generalized inverse matrix algorithm with regularization constraints. This achieves stable reconstruction from discrete "line" data to a continuous "surface" theoretical deformation field. Furthermore, the system integrates an AI residual neural network compensation mechanism, using low-frequency periodic point cloud data as supervised labels for online iteration, and real-time superposition of material inhomogeneities, welding residual stress, and other nonlinear micro-defect residuals that cannot be exhausted by mechanical formulas. In terms of theoretical framework, this dual-track fusion architecture, which provides global trends through physical drive and compensates for microscopic defects through data drive, enables the system to output a continuous three-dimensional deformation displacement field that is 100% matched with the solid steel structure with extremely high sampling refresh rate, high precision, continuous and smooth output even during the window period of lidar scanning. Finally, through the energy state hierarchical early warning terminal, the geometric field depth is mapped into a full-field strain energy density field. Based on the three core mechanical indicators of overall strain energy, local strain energy density gradient and strain energy change rate, the system performs zero-delay, visualized hierarchical judgment and physical alarm output for three different levels of physical disease evolution stages: "overall instability, local buckling and rapid damage propagation". The entire implementation, from the high-frequency capture of physical displacement at the front end, to the high-precision full-field continuous reconstruction by the dual engines of mechanics + AI in the middle, and then to the intelligent safety decision-making of the deep energy field state at the back end, truly realizes the all-weather, highly robust intelligent monitoring and dynamic safety early warning closed loop of large-span spatial steel structures under long-term and complex environmental loads.

[0063] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0064] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart early warning and monitoring system for steel structure deformation based on laser measuring instrument technology, characterized in that: The system includes a dynamic laser ranging array, a spatial stiffness laser scanning unit, a spatiotemporal registration and matrix transformation module, a mechanical mode space reconstruction server, and an energy state hierarchical early warning terminal, wherein: The dynamic laser ranging array includes multiple array-type high-speed laser rangefinders fixed on an external reference axis, used to continuously acquire the one-dimensional dynamic displacement response time sequence of multiple characteristic force nodes on the surface of the steel structure at a first sampling frequency; The spatial stiffness laser scanning unit includes a three-dimensional solid-state lidar, which is used to periodically acquire full-field spatial point cloud data of the steel structure at a second sampling frequency, wherein the first sampling frequency is greater than the second sampling frequency. The spatiotemporal registration and matrix transformation module is used to unify the one-dimensional dynamic displacement response time series and the full-field spatial point cloud data into the structural mechanics finite element spatial coordinate system of the steel structure. The mechanical modal space reconstruction server internally constructs the inherent modal space matrix of the steel structure, the physical information modal solution model, and the residual neural network compensation model. The mechanical modal space reconstruction server uses the full-field spatial point cloud data to extract the global spatial modal vibration mode basis of each order, and uses the one-dimensional dynamic displacement response time series to solve the dynamic dynamic weight coefficients in real time. The theoretical deformation field is obtained through modal synthesis, and the residual neural network compensation model is used to perform nonlinear correction on the theoretical deformation field to reconstruct the continuous three-dimensional deformation displacement field of the entire steel structure. The energy state hierarchical early warning terminal is used to calculate the full-field strain energy density distribution corresponding to the continuous three-dimensional deformation displacement field, and to provide early warning of slippage or instability based on the energy mutation rate.

2. The intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology according to claim 1, characterized in that: The placement points of each array-type high-speed laser rangefinder in the dynamic laser ranging array are determined based on the sensor placement optimization algorithm of the steel structure mechanical model. The placement points are located at the displacement maxima nodes or strain energy sensitive nodes of the first few natural mode vibration modes of the steel structure.

3. The intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology according to claim 1, characterized in that: The spatial stiffness laser scanning unit also includes multiple laser cooperative targets deployed at key weld points of the steel structure. The three-dimensional solid-state lidar extracts the three-dimensional coordinates of the center of the laser cooperative targets by identifying the differences in the reflection intensity of the laser cooperative targets, and introduces them into the mechanical modal space reconstruction server as spatial physical boundary conditions.

4. The intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology according to claim 1, characterized in that: The method for constructing the intrinsic mode space matrix in the mechanical mode space reconstruction server includes the following steps: Sp1: Based on the geometric drawings of the steel structure, the initial finite element stiffness matrix and mass matrix are established, and the theoretical spatial mode shapes are obtained through eigenvalue calculation. Sp2: During system initialization, the spatial stiffness laser scanning unit is used to acquire two sets of full-field spatial point cloud data under no-load and calibrated load conditions, and the difference between them is calculated to obtain the initial deformation field. Sp3: Using the initial deformation field, the sensitivity matrix of each order of the theoretical spatial mode is corrected to obtain a global spatial mode base that conforms to the physical properties of the solid steel structure.

5. The intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology according to claim 4, characterized in that: The physical information modal solution model in the mechanical modal space reconstruction server is embedded with elastic mechanical continuity differential equation constraints. The mechanical modal space reconstruction server uses a generalized inverse matrix algorithm with regularization constraints to globally suppress the ill-conditioned singularity of the spatial matrix and laser measurement noise by introducing a boundary energy penalty term. It solves in real time the sparse extraction matrix based on the global spatial modal vibration mode base at the deployment points of the dynamic laser ranging array, as well as the topological extension equation composed of the dynamic dynamic weight coefficient vector to be solved.

6. The intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology according to claim 5, characterized in that: The mechanical modal space reconstruction server synthesizes the global spatial modal vibration mode base and the solved dynamic dynamic weight coefficient vector by matrix multiplication to obtain a theoretical deformation displacement field that is continuous in the spatial domain and satisfies the constraints of the elasticity continuity differential equation.

7. The intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology according to claim 1, characterized in that: The construction and correction process of the residual neural network compensation model in the mechanical modal space reconstruction server includes the following steps: Sp4: At the periodic moment when the spatial stiffness laser scanning unit synchronously uploads the full-field spatial point cloud data, the actual point cloud deformation field corresponding to that moment is matrix subtracted from the theoretical deformation displacement field to extract the spatial nonlinear deformation residual matrix. Sp5: Using the one-dimensional dynamic displacement response time series as the input feature and the spatial nonlinear deformation residual matrix as the training label, supervised learning training is performed on the residual neural network compensation model. Sp6: During the scanning window, the real-time one-dimensional dynamic displacement response time sequence is input into the trained residual neural network compensation model to obtain the real-time predicted residual, and the real-time predicted residual is superimposed on the real-time theoretical deformation displacement field to output the corrected continuous three-dimensional deformation displacement field.

8. The intelligent early warning and monitoring system for steel structure deformation based on laser measuring instrument technology according to claim 1, characterized in that: The energy state classification early warning terminal calculates the current strain energy, local strain energy density gradient, and strain energy change rate of the entire steel structure based on the corrected continuous three-dimensional deformation displacement field integral, and performs slip or instability early warning according to the following logic: when the overall strain energy exceeds the preset elastic limit energy threshold, it is determined that the steel structure has an overall instability risk; when the local strain energy density gradient exceeds the preset threshold, it is determined that the corresponding area has a local damage or local buckling risk; when the strain energy change rate shows a step change, it is determined that the steel structure is in the rapid damage propagation stage and triggers the corresponding level of early warning signal.

Images