Large steel structure micro-deformation detection method

Abstract

The invention provides a large-scale steel structure micro-deformation detection method, and belongs to the technical field of micro-deformation detection.According to the large-scale steel structure micro-deformation detection method, micro-vibration sensors are installed at key nodes to collect synchronous signals, a conduction matrix is constructed to reflect the conduction relation between the nodes, and an attenuation matrix is calculated to represent energy attenuation characteristics; frequency domain analysis is carried out to construct a resonance matrix and decompose the resonance matrix into a peak-valley matrix, a micro-deformation characteristic matrix is constructed based on the matrix fusion, a pre-training model is used to calculate a micro-deformation matrix and a key matrix, and finally, the deformation state of the steel structure is evaluated, and an over-limit region and severity are determined. According to the method, real-time and accurate detection of the micro-deformation of the large steel structure is realized, the core technical problem of high difficulty in real-time and accurate detection of the micro-deformation of the large steel structure in the prior art is solved, and a new technical means is provided for safety monitoring of the steel structure.

Description

Technical Field

[0001] The present invention belongs to the technical field of micro-deformation detection, and more specifically, relates to a method for detecting micro-deformations of large steel structures. Background Art

[0002] Due to their excellent strength-to-weight ratio, high construction efficiency, and strong spatial adaptability, large steel structures are widely used in important infrastructure such as stadiums, exhibition centers, high-rise buildings, and long-span bridges. With the continuous expansion of building scale and the complexity of structures, the safety monitoring of large steel structures has become particularly important. Traditional steel structure health monitoring mainly relies on technical means such as displacement monitoring, strain measurement, and vibration analysis. Static displacement monitoring usually uses measuring devices such as total stations and levels to regularly measure key points of the structure to obtain displacement data; strain measurement monitors the strain of the structure under load by installing strain gauges on the surface of the structure; vibration analysis technology mainly uses acceleration sensors to collect structural vibration signals and evaluates the structural state through methods such as modal analysis and spectrum analysis. These technologies have been widely used in engineering practice and provide a basic guarantee for the safe operation of steel structures.

[0003] However, with the increasing complexity of the application environment of steel structures and the improvement of safety requirements, traditional technologies have shown obvious deficiencies in the real-time and accurate detection of micro-deformations. First, static displacement monitoring is mainly accurate to the millimeter level, making it difficult to capture early micron-level deformations of the structure, and the measurement interval is relatively long, making continuous monitoring impossible; second, although strain gauge monitoring has high accuracy, the installation process is complex, the number of monitoring points is limited, and it is difficult to construct a global deformation image; third, traditional vibration analysis mainly focuses on changes in modal parameters in the low-frequency range, lacks sensitivity to high-frequency characteristics caused by micro-deformations, and it is difficult to establish an accurate correspondence between vibration signals and micro-deformations. In addition, affected by environmental factors such as temperature changes, humidity fluctuations, and external interference, the detection accuracy and reliability of traditional technologies are further reduced. Especially under complex load conditions, the deformations of steel structures often exhibit non-linear characteristics, and traditional linear analysis methods are difficult to accurately describe such complex deformation processes.

[0004] Micro-deformation of large steel structures is an early sign of structural performance degradation, and timely and accurate detection is crucial to preventing major safety accidents. Micro-deformation is usually at the micron level, hidden and slow to develop. Traditional detection methods can often only detect deformation after it develops to a dangerous level, losing the opportunity for early intervention. Existing technologies are difficult to meet the requirements of high precision, real-time performance and wide coverage at the same time, especially for large-span steel structures, which are more difficult to monitor comprehensively due to their large size and numerous nodes. In practical applications, factors such as environmental noise interference, signal acquisition frequency limitations and insufficient sensor layout density further aggravate the difficulty of accurate detection of micro-deformation. Therefore, how to overcome these limitations and realize real-time and accurate detection of micro-deformation of large steel structures has become a key technical problem that needs to be solved in the field of steel structure health monitoring. In other words, there is a technical problem in the existing technology that it is difficult to accurately detect micro-deformation of large steel structures in real time. Summary of the invention

[0005] In view of this, the present invention provides a method for detecting micro-deformation of large steel structures, which can solve the technical problem in the prior art that it is difficult to accurately detect micro-deformation of large steel structures in real time.

[0006] The present invention is implemented as follows: The present invention provides a method for detecting micro-deformation of a large steel structure, comprising the following steps: installing multiple micro-vibration sensors on the large steel structure, wherein the micro-vibration sensors are distributed at key nodes of the steel structure and are used to collect micro-vibration signals of the steel structure; using the micro-vibration sensors to synchronously collect micro-vibration signals of the steel structure at different time points to form a micro-vibration signal set; constructing a micro-vibration conduction matrix based on the micro-vibration signal set, wherein the micro-vibration conduction matrix characterizes the conduction relationship of micro-vibration signals between nodes of the steel structure; calculating a micro-vibration conduction attenuation matrix based on the micro-vibration conduction matrix, wherein the micro-vibration conduction attenuation matrix characterizes the transmission process of micro-vibration signals in the steel structure. energy attenuation characteristics; perform frequency domain analysis on the micro-vibration signal set to identify the resonance characteristics of the steel structure and construct a micro-vibration conduction resonance matrix; decompose the micro-vibration conduction resonance matrix into a resonance peak matrix and a resonance valley matrix; construct a micro-deformation fusion feature matrix of the steel structure based on the micro-vibration conduction matrix, the micro-vibration conduction attenuation matrix, the resonance peak matrix and the resonance valley matrix; use a pre-trained micro-vibration-micro-deformation model to process the micro-deformation fusion feature matrix to calculate the micro-deformation matrix; calculate the micro-deformation key matrix based on the micro-deformation matrix; evaluate the deformation state of the steel structure based on the micro-deformation key matrix to determine the micro-deformation over-limit area and severity.

[0007] Among them, the key nodes refer to the connection points in the long-span steel structure that have important influences on the overall structural stability and force transmission performance, including the connection points between the main steel beams and columns, the intersection points of the main and secondary beams, the support position points, and the position points of the cross-section with span change; the micro-vibration sensor has the ability to measure triaxial acceleration and can simultaneously collect micro-vibration signals in the x, y, and z directions; the micro-vibration sensor is firmly fixed on the surface of the steel structure by a rigid connection method to ensure that there is no attenuation in the vibration transmission between the sensor and the structure; for large steel structures, the sensor distribution density is to install 1 sensor per 50 - 100 square meters, and it is appropriately densified at the key nodes.

[0008] Among them, the synchronous collection of the micro-vibration signals of the steel structure by the micro-vibration sensor at different time points includes: under normal working conditions, the system performs a full-network synchronous collection with a duration of 10 minutes every 4 hours; under special working conditions, when the vibration acceleration detected by any sensor exceeds the preset threshold, a full-network synchronous collection is automatically triggered, and the duration is from 30 seconds before the event to 10 minutes after the event; the collected original micro-vibration signals are preprocessed, including signal denoising, baseline correction, and outlier processing; the preprocessed micro-vibration signals in each time window are organized according to the collection time and sensor position to form a four-dimensional micro-vibration signal set.

[0009] Among them, constructing the micro-vibration conduction matrix includes: performing cross-correlation analysis on the signals between each pair of sensor nodes in the preprocessed micro-vibration signal set and calculating the cross-correlation function; constructing an initial micro-vibration conduction time matrix based on the propagation time of all node pairs; reconstructing the conduction time matrix using the shortest path algorithm; calculating the micro-vibration propagation speed matrix in the structure based on the optimized conduction time matrix and combining the spatial distance between the sensors; calculating the stress transfer efficiency matrix through the relationship between the propagation speed matrix and the material properties of the structure; constructing the micro-vibration conduction matrix by integrating the characteristics of propagation time, propagation speed, and stress transfer efficiency.

[0010] Among them, calculating the micro-vibration conduction attenuation matrix includes: performing energy attenuation analysis on the micro-vibration signals between each pair of sensor nodes and calculating the energy attenuation rate; selecting clear micro-vibration events, and for sensors i and j, respectively calculating the energy spectral density functions of the micro-vibration signals at the two points; performing band integration on the energy spectral density functions to obtain the total energy value; calculating the energy attenuation ratio, which represents the proportion of micro-vibration energy transfer from node i to node j; introducing distance normalization processing and calculating the energy attenuation coefficient per unit distance; constructing an initial micro-vibration conduction attenuation matrix based on the attenuation coefficient; performing optimization processing on the initial attenuation matrix using the spatial smoothing filtering method; applying the structural symmetry constraint, for the parts with symmetric structures, forcing their attenuation characteristics to satisfy the corresponding symmetry.

[0011] Among them, the frequency-domain analysis of the micro-vibration signal set includes: converting the time-domain signal into the frequency domain by using the fast Fourier transform method; windowing the signal by using the Hanning window; smoothing the converted spectrum; identifying the frequency-domain characteristics of each measuring point based on the smoothed spectrum; for each measuring point, extracting the resonance frequency and the corresponding amplitude; analyzing the frequency-domain relationship between each measuring point by using the frequency response function method and calculating the transfer function; constructing a frequency correlation matrix based on the transfer function; performing eigenvalue decomposition on the frequency correlation matrix and extracting the main modal characteristics; constructing a micro-vibration conduction resonance matrix based on the modal analysis results and the frequency correlation matrix.

[0012] Among them, decomposing the micro-vibration conduction resonance matrix into a resonance peak matrix and a resonance valley matrix includes: setting a frequency response threshold; performing peak search and valley search on the frequency response curve of each measuring point; performing spatial clustering analysis on the original peak-valley characteristics, grouping the peak points with close frequencies and adjacent spatial positions into the same resonance mode, and grouping the valley points with close frequencies and adjacent spatial positions into the same suppression mode; constructing a resonance peak matrix and a resonance valley matrix based on the clustering results; performing feature extraction on the resonance peak matrix and the resonance valley matrix in the frequency dimension and calculating the weight coefficients of each frequency point; using the weighted summation method to reduce the three-dimensional resonance peak matrix and resonance valley matrix to a two-dimensional matrix.

[0013] Among them, constructing the micro-deformation fusion feature matrix of the steel structure includes: determining the weights of each feature matrix in the fusion process; normalizing each feature matrix to ensure that the dimensions and numerical ranges of different matrices are the same; performing weighted fusion on the normalized feature matrices, calculating the fusion feature values by using the linear weighted method to form a preliminary micro-deformation fusion feature matrix; applying spatial filtering to the preliminary fusion feature matrix by using a Gaussian filter; applying nonlinear enhancement processing to the filtered fusion matrix by using the power function enhancement method.

[0014] Among them, calculating the micro-deformation matrix includes: converting the micro-deformation fusion feature matrix into a format suitable for model input; normalizing the compressed feature data; inputting the processed feature data into a pre-trained micro-vibration - micro-deformation model; reorganizing the model output results into a micro-deformation matrix; introducing physical constraint conditions for post-processing, applying the structural deformation continuity constraint, applying the principle of minimum structural deformation energy, and applying the support constraint conditions; after physical constraint optimization, obtaining the final micro-deformation matrix.

[0015] Among them, calculating the key matrix of micro-deformation includes: extracting the deformation data of key nodes from the micro-deformation matrix; evaluating the importance of the deformation data of key nodes, and the evaluation indexes include the absolute value of the deformation amount, the relative value of the deformation amount, the consistency between the deformation direction and the structural stress direction, etc.; calculating the comprehensive risk coefficient based on multiple evaluation indexes; constructing the key matrix of micro-deformation according to the comprehensive risk coefficient; performing cluster analysis on the key matrix of micro-deformation, and classifying the key nodes with similar risk coefficients into multiple levels corresponding to different risk levels; determining the micro-deformation over-limit area includes: establishing a micro-deformation safety threshold system; comparing the deformation data in the key matrix of micro-deformation with the safety threshold to identify the over-limit area; performing spatial cluster analysis on the over-limit area; calculating the deformation characteristic parameters for each over-limit area; evaluating the severity of the over-limit based on the deformation characteristic parameters; generating a micro-deformation over-limit area distribution map and a micro-deformation over-limit development trend map.

[0016] Compared with the prior art, a method for detecting micro-deformation of a large steel structure provided by the present invention proposes a method for detecting micro-deformation of a large steel structure based on multi-sensor collaboration. By arranging micro-vibration sensors at key nodes of the steel structure, collecting structural micro-vibration signals, constructing multi-dimensional characteristics such as a micro-vibration conduction matrix, a conduction attenuation matrix, and a resonance characteristic matrix, and combining with a micro-vibration - micro-deformation model, real-time and accurate detection of micro-deformation of the steel structure is achieved. This method breaks through the limitations of traditional technologies, can achieve the detection of micro-deformations at the micron level, and provides a new technical means for the safety monitoring of steel structures.

[0017] The present invention solves the technical deficiencies of traditional technologies in the real-time and accurate detection of micro-deformations. First, by adopting a high-sensitivity micro-vibration sensor network, continuous and high-precision acquisition of steel structure micro-vibration signals is achieved, and the detection accuracy can reach the micron level, which is one order of magnitude higher than that of traditional static displacement monitoring; second, by constructing a micro-vibration conduction matrix, the internal stress transmission path of the structure is analyzed, avoiding the limitation that traditional methods can only obtain local information and realizing the inference of the global deformation state; third, by calculating the micro-vibration conduction attenuation matrix, the attenuation law of micro-vibration energy in the structure is revealed, providing a new dimension for indirectly reflecting the structure material and connection state; fourth, through frequency domain analysis and the construction of a resonance characteristic matrix, the internal connection between micro-vibration characteristics and structural micro-deformations is established. In particular, by decomposing the resonance peak matrix and the resonance valley matrix, the dangerous frequencies and safe frequencies of the structure are identified, providing frequency domain characteristics for accurate detection; fifth, through the fusion of multi-dimensional characteristic matrices, a micro-deformation fusion characteristic matrix is constructed, comprehensively considering the conduction characteristics, attenuation characteristics, and resonance characteristics, and improving the accuracy and anti-interference ability of detection.

[0018] The present invention solves the technical problem of the great difficulty in the real-time and accurate detection of micro-deformations of large steel structures, and realizes the early and accurate identification of micro-deformations. Compared with traditional technologies, the present invention has significant advantages: First, the detection accuracy is improved, and the detection of micro-meter-level deformations can be realized to meet the requirements of early warning; Second, the real-time monitoring is enhanced, and the real-time evaluation of the deformation state is realized by continuously collecting micro-vibration signals; Third, the spatial coverage is wider, and a global deformation image of the structure is constructed by arranging sensors at key nodes; Fourth, the feature dimension is richer, and the micro-vibration signals are analyzed from three dimensions of conduction, attenuation, and resonance to extract more complete deformation features; Fifth, the anti-interference ability is stronger, and the influence of environmental noise on the detection results is reduced through multi-dimensional feature fusion. In addition, the present invention also realizes the quantitative evaluation of the influence degree of the deformation of key nodes, providing a decision-making basis for structural safety management. These technical advantages enable the present invention to realize the real-time and accurate detection of micro-deformations of large steel structures in complex environments, providing reliable technical support for the whole life cycle management of steel structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] In order to make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0021] As Figure 1 shown, it is a flowchart of a method for detecting micro-deformations of a large steel structure provided by the present invention. The method includes the following steps:

[0022] S01. Install a plurality of micro-vibration sensors on the large steel structure. The micro-vibration sensors are distributed at key nodes of the steel structure and are used to collect micro-vibration signals of the steel structure;

[0023] S02. Use the micro-vibration sensors to synchronously collect micro-vibration signals of the steel structure at different time points to form a micro-vibration signal set;

[0024] S03. Based on the micro-vibration signal set, construct a micro-vibration conduction matrix, and the micro-vibration conduction matrix characterizes the conduction relationship of micro-vibration signals between nodes of the steel structure;

[0025] S04. Calculate a micro-vibration conduction attenuation matrix according to the micro-vibration conduction matrix, and the micro-vibration conduction attenuation matrix characterizes the energy attenuation characteristics of the micro-vibration signal during the transmission process in the steel structure;

[0026] S05. Perform frequency-domain analysis on the micro-vibration signal set, identify the resonance characteristics of the steel structure, and construct a micro-vibration conduction resonance matrix;

[0027] S06. Decompose the micro-vibration conduction resonance matrix into a resonance peak matrix and a resonance valley matrix. The resonance peak matrix characterizes the resonance enhancement characteristics of the steel structure at specific frequencies, and the resonance valley matrix characterizes the resonance suppression characteristics of the steel structure at specific frequencies;

[0028] S07. Based on the micro-vibration conduction matrix, the micro-vibration conduction attenuation matrix, the resonance peak matrix, and the resonance valley matrix, construct a micro-deformation fusion feature matrix of the steel structure;

[0029] S08. Use the pre-trained micro-vibration - micro-deformation model to process the micro-deformation fusion feature matrix, and calculate the micro-deformation matrix. The micro-deformation matrix characterizes the deformation amount and direction of each monitoring point of the steel structure;

[0030] S09. Calculate the micro-deformation key matrix according to the micro-deformation matrix. The micro-deformation key matrix characterizes the deformation amount of the key nodes of the steel structure and its influence degree on the structural safety;

[0031] S10. Based on the micro-deformation key matrix, evaluate the deformation state of the steel structure, and determine the micro-deformation over-limit area and severity;

[0032] Among them, the key nodes refer to the connection points in the long-span steel structure that have important influences on the overall stability and force transmission performance of the structure, including the connection points between the main beams and columns of the steel structure, the intersection points of the main and secondary beams, the support position points, and the position points of the cross-section with span change;

[0033] The micro-vibration conduction matrix represents the transmission relationship of micro-vibration signals between each monitoring point of the steel structure, and reflects the internal stress transmission path of the structure;

[0034] The micro-vibration conduction attenuation matrix represents the energy attenuation law of micro-vibration signals during the transmission process in the steel structure, and reflects the state of the structural materials and connection parts;

[0035] The micro-vibration conduction resonance matrix represents the resonance characteristics of the steel structure at different frequencies, and reflects the natural frequencies and modal characteristics of the structure;

[0036] The resonance peak matrix represents the characteristics of resonance enhancement of the steel structure at specific frequencies, corresponding to the dangerous frequencies of the structure;

[0037] The resonance valley matrix represents the characteristics of resonance suppression of the steel structure at specific frequencies, corresponding to the safe frequencies of the structure;

[0038] The micro-deformation matrix represents the deformation amount and direction of each monitoring point of the steel structure, and reflects the actual deformation state of the structure;

[0039] The key matrix of micro-deformation represents the deformation amount of the key nodes of the steel structure and its influence degree on the structural safety. It is a subset of the micro-deformation matrix and is used to evaluate the structural safety.

[0040] The micro-deformation over-limit area refers to the area where the micro-deformation amount of the key node exceeds the preset safety threshold.

[0041] The specific implementation manners of the above steps are described in detail below. The specific implementation manner of step S01 is to establish a structural micro-vibration signal acquisition network system. First, determine the distribution positions of the key nodes of the large steel structure, including the connection positions of main beams and columns, the intersection points of primary and secondary beams, the support position points, and the cross-section position points of span changes, etc. Select highly sensitive micro-vibration sensors, where the sensor accuracy is not less than 0.001g, the sampling frequency is not less than 1000Hz, and it has the ability to measure triaxial acceleration and can collect micro-vibration signals in three directions of x, y, and z at the same time. When installing, use a rigid connection method and use high-strength epoxy resin or a special metal fixing seat to firmly fix the sensor on the surface of the steel structure to ensure that there is no attenuation in the vibration transmission between the sensor and the structure. For large steel structures, it is recommended to install 1 sensor per 50 - 100 square meters, and the key nodes can be appropriately densified. The sensors are connected through a high-speed data transmission network, adopting a star or mesh topology structure to ensure the real-time and reliability of data transmission. Use the GPS or Beidou satellite positioning system to synchronize the time of each sensor, and the time synchronization accuracy should be better than 1 microsecond to ensure that the data collected by sensors at different positions has strict time consistency. After the sensor installation is completed, conduct an initial calibration test. By applying a known vibration source to the structure, check the response consistency and signal quality of each sensor. At the same time, establish a sensor health monitoring mechanism to monitor the working state of the sensors in real time, and mark or replace the sensors with abnormal signals in time. The function of this step is to establish a high-precision and high-reliability micro-vibration signal acquisition network and provide high-quality basic data for subsequent structural micro-deformation analysis.

[0042] The specific implementation of step S02 is to perform synchronous acquisition and preprocessing of micro-vibration signals with multiple time windows. First, a signal acquisition plan is formulated, including two modes: regular acquisition under normal working conditions and triggered acquisition under special working conditions (such as strong wind, dense crowd, etc.). Under normal working conditions, the system performs full-network synchronous acquisition every 4 hours for a duration of 10 minutes. Under special working conditions, when the vibration acceleration detected by any sensor exceeds a preset threshold (recommended to be 0.05g), full-network synchronous acquisition is automatically triggered, with a duration from 30 seconds before the event to 10 minutes after the event. During the acquisition process, each sensor maintains strict time synchronization, the sampling rate is uniformly set to 1000Hz, and the dynamic range is set to ±2g. The acquired original micro-vibration signals are preprocessed, including signal denoising, baseline correction, and outlier processing. For denoising, the wavelet decomposition and reconstruction method is used. The db4 wavelet basis is selected, and the signal is decomposed into 5 layers. After removing the high-frequency noise and low-frequency drift components, the signal is reconstructed. Baseline correction is performed using the polynomial fitting method, and a 3rd-order polynomial is used to fit and correct the signal baseline. Outlier processing is performed using the sliding median filtering method, and the sliding window size is set to 5 data points. The preprocessed micro-vibration signals of each time window are organized according to the acquisition time and sensor location to form a four-dimensional micro-vibration signal set, arranged according to the time window index, sensor index, time sampling point, and spatial dimension. At the same time, the quality of the signal set is evaluated, the signal-to-noise ratio is calculated, and the signal segments with a signal-to-noise ratio lower than 20dB are marked and given a lower weight or directly excluded in subsequent analysis. The role of this step is to obtain a high-quality signal data set reflecting the micro-vibration characteristics of the steel structure under different times and different working conditions, providing a reliable data basis for subsequent analysis of micro-vibration conduction characteristics.

[0043] The specific implementation of step S03 is to construct a micro-vibration conduction matrix that reflects the internal stress transmission path of the structure. First, perform cross-correlation analysis on the signals between each pair of sensor nodes in the preprocessed micro-vibration signal set, and calculate the cross-correlation function. For sensor i and sensor j, by calculating the correlation degree of the signals collected by the two sensors at different time delays, find the time delay with the maximum correlation, and determine the propagation time of the micro-vibration signal between the two points. Based on the propagation times of all node pairs, construct an initial micro-vibration conduction time matrix, where the matrix elements represent the micro-vibration propagation time from node i to node j. Considering that there may be multiple paths for micro-vibration propagation in the structure, optimize the initial conduction time matrix, and use the shortest path algorithm (such as Dijkstra's algorithm) to reconstruct the conduction time matrix to ensure that the matrix represents the main propagation path of the micro-vibration signal in the structure. Based on the optimized conduction time matrix, combined with the spatial distance between the sensors, calculate the propagation speed matrix of the micro-vibration in the structure. For steel structures, the normal range of micro-vibration propagation speed is 3000 - 6000 m / s, and a propagation speed significantly deviating from this range may indicate structural abnormalities. Through the relationship between the propagation speed matrix and the structural material properties (such as elastic modulus, density, etc.), calculate the stress transfer efficiency matrix, which characterizes the stress transfer efficiency between different parts of the structure. Finally, comprehensively consider the characteristics of propagation time, propagation speed, and stress transfer efficiency in three aspects to construct a micro-vibration conduction matrix. This matrix is a comprehensive matrix, and its value range is normalized to 0 - 1. The larger the value, the smoother the micro-vibration conduction between the two points and the better the structural connection state. The role of this step is to construct a conduction matrix that characterizes the internal stress transmission path of the structure by analyzing the propagation characteristics of the micro-vibration signal in the structure, providing an important basis for evaluating the structural connection state and overall stability.

[0044] The specific implementation of step S04 is to calculate the micro-vibration conduction attenuation matrix to characterize the structural energy dissipation characteristics. First, perform energy attenuation analysis on the micro-vibration signals between each pair of sensor nodes and calculate the energy attenuation rate. Select clear micro-vibration events (such as transient responses generated by human knocking, environmental excitation, etc.). For sensor i and sensor j, calculate the energy spectral density functions of the micro-vibration signals at the two points respectively, where the frequency is used as the independent variable. Integrate the energy spectral density function over the frequency band to obtain the total energy value. Calculate the energy attenuation ratio, which represents the proportion of micro-vibration energy transferred from node i to node j. Considering the relationship between energy attenuation and propagation distance, introduce distance normalization processing and calculate the energy attenuation coefficient per unit distance, that is, divide the natural logarithm of the energy attenuation ratio by the physical distance between the nodes. Based on the attenuation coefficient, construct the initial micro-vibration conduction attenuation matrix. Due to the existence of noise and interference in actual measurements, optimize the initial attenuation matrix. Adopt the spatial smoothing filtering method to correct the outliers in the matrix, and set the filter window size to 3×3. Apply the structural symmetry constraint. For parts with symmetric structures, force their attenuation characteristics to satisfy the corresponding symmetry. Combine the data of multiple time windows and use the exponentially weighted moving average method to update the attenuation matrix, giving higher weights to recent data, and set the weight decay coefficient to 0.9. Finally, obtain the optimized micro-vibration conduction attenuation matrix, which characterizes the attenuation characteristics of the micro-vibration energy during the transmission process in the structure. For steel structures, the normal range of the energy attenuation coefficient is -0.05 to -0.15 / m, and values outside this range may indicate abnormal structural connections or material damage. The role of this step is to construct a conduction attenuation matrix by analyzing the attenuation characteristics of micro-vibration energy in the structure. This matrix reflects the state of the structural materials and connection parts and can be used to identify potential structural defects or damage locations.

[0045] The specific implementation of step S05 is to perform frequency-domain analysis on the micro-vibration signal and construct a resonance characteristic matrix. First, perform spectral analysis on the preprocessed micro-vibration signal, and use the fast Fourier transform method to convert the time-domain signal into the frequency domain. To improve the accuracy of spectral analysis, apply a Hanning window to the signal, with the window length set to 4096 data points and the overlapping rate of adjacent windows being 50%. Smooth the obtained spectrum using the moving average method, with the smoothing window width being 5 frequency points. Based on the smoothed spectrum, identify the frequency-domain characteristics of each measurement point, including the main frequency components, amplitude distribution, and phase characteristics. For each measurement point, extract the resonance frequency and the corresponding amplitude to form the resonance feature vector of the measurement point. Usually, the main resonance frequency range of the steel structure is between 0.5 - 10 Hz, and this frequency band is analyzed emphatically. Use the frequency response function method to analyze the frequency-domain relationship between each measurement point, calculate the transfer function, which represents the frequency response ratio of node j to node i. Based on the transfer function, construct a frequency correlation matrix, which characterizes the response correlation degree of different measurement points at a specific frequency. Perform eigenvalue decomposition on the frequency correlation matrix to extract the main modal characteristics. Usually, take the first 5 - 10 main modes for analysis, and these modes can usually describe more than 90% of the dynamic response characteristics of the structure. Based on the modal analysis results and the frequency correlation matrix, construct a micro-vibration conduction resonance matrix. This matrix is a three-dimensional tensor that describes the resonance conduction characteristics between each node of the structure at different frequencies. For subsequent processing convenience, slice the resonance matrix at specific frequency points to form a series of two-dimensional matrices, where the specific frequency points usually select the main resonance frequency points of the structure. The function of this step is to identify the resonance characteristics of the steel structure through frequency-domain analysis, construct a resonance matrix representing the modal characteristics of the structure, provide frequency-domain information for subsequent micro-deformation feature fusion, and can also be used to evaluate the dynamic performance of the structure and identify potential structural weaknesses.

[0046] The specific implementation of step S06 is to decompose the micro-vibration conduction resonance matrix into a resonance peak matrix and a resonance valley matrix to distinguish the vibration characteristics of the structure at different frequencies. First, perform peak-valley separation processing on the micro-vibration conduction resonance matrix obtained in step S05. Set the frequency response threshold. Usually, 1.5 times the average frequency response amplitude of the structure is taken as the resonance peak identification threshold, and 0.6 times the average frequency response amplitude is taken as the resonance valley identification threshold. Conduct peak search on the frequency response curve of each measurement point, and use the curve fitting method to find local maximum points. It is required that the amplitude of the maximum point is higher than the set resonance peak threshold, and the curves on both sides of it satisfy the concavity and convexity changes. Similarly, conduct valley search on the frequency response curve to find local minimum points. It is required that the amplitude of the minimum point is lower than the set resonance valley threshold, and the curves on both sides of it satisfy the concavity and convexity changes. For the identified peak points, record their frequency positions and amplitudes to construct the original resonance peak feature dataset. For the identified valley points, record their frequency positions and amplitudes to construct the original resonance valley feature dataset. Considering the spatial consistency of the structure, perform spatial clustering analysis on the original peak-valley features. Use the density clustering algorithm to group the peak points with close frequencies and adjacent spatial positions into the same resonance mode, and group the valley points with close frequencies and adjacent spatial positions into the same suppression mode. Based on the clustering results, construct the resonance peak matrix. The elements of this matrix represent the degree of resonance enhancement from one node to another at a specific frequency. The degree of resonance enhancement is calculated by the deviation between the frequency response ratio of two points at this frequency and the reference response ratio. A positive value indicates resonance enhancement, and the larger the value, the more significant the enhancement. Similarly, construct the resonance valley matrix. The elements of this matrix represent the degree of resonance suppression from one node to another at a specific frequency. The degree of resonance suppression is calculated by the deviation between the frequency response ratio of two points at this frequency and the reference response ratio. A negative value indicates resonance suppression, and the larger the absolute value, the more significant the suppression. To simplify subsequent processing, perform feature extraction on the resonance peak matrix and the resonance valley matrix in the frequency dimension, calculate the weight coefficients of each frequency point, and the higher the frequency is closer to the main modal frequency of the structure, the higher the weight. Use the weighted summation method to reduce the three-dimensional resonance peak matrix and resonance valley matrix to two-dimensional matrices. The function of this step is to distinguish the resonance characteristics of the steel structure at different frequencies, respectively identify the dangerous frequencies (resonance enhancement frequencies) and safe frequencies (resonance suppression frequencies) of the structure, and provide a basis for evaluating the performance of the structure under dynamic excitation.

[0047] The specific implementation of step S07 is to construct a micro-deformation fusion feature matrix of the steel structure based on the aforementioned matrix. First, determine the weights of each feature matrix during the fusion process. The weight allocation needs to consider factors such as the structure type and monitoring objectives. For large steel structures, it is recommended that the weight of the micro-vibration conduction matrix be 0.35, the weight of the micro-vibration conduction attenuation matrix be 0.25, the weight of the resonance peak matrix be 0.25, and the weight of the resonance valley matrix be 0.15. Before fusion, each feature matrix needs to be normalized to ensure that the dimensions and numerical ranges of different matrices are consistent. For the micro-vibration conduction matrix, the maximum-minimum normalization method is used to map the matrix element values to the 0-1 interval; for the micro-vibration conduction attenuation matrix, since its original values are negative, first take the absolute value, then use the maximum-minimum normalization, and then subtract the normalization result from 1 to ensure that the larger the value, the better the conduction performance; for the resonance peak matrix, the maximum-minimum normalization method is used to normalize the positive part to the 0-1 interval and set the negative part to 0; for the resonance valley matrix, the maximum-minimum normalization method is used to take the absolute value of the negative part and normalize it to the 0-1 interval and set the positive part to 0. Perform weighted fusion on the normalized feature matrices, and use the linear weighted method to calculate the fusion feature values to form a preliminary micro-deformation fusion feature matrix. Considering the spatial correlation between matrix elements, apply spatial filtering to the preliminary fusion feature matrix. Use a Gaussian filter with a filtering radius set to 0.1 times the minimum feature size of the structure to eliminate the influence of local noise and outliers. To highlight key features, apply non-linear enhancement processing to the filtered fusion matrix. Use the power function enhancement method. For the region where the fusion value is higher than 0.8, apply a power function with an exponent of 0.8 to weaken the enhancement effect; for the region where the fusion value is between 0.5 and 0.8, keep it unchanged; for the region where the fusion value is lower than 0.5, apply a power function with an exponent of 1.2 to enhance the effect. The finally constructed micro-deformation fusion feature matrix characterizes the comprehensive conduction performance and deformation correlation between each monitoring point of the structure. The larger the matrix element value, the better the conduction performance and the stronger the deformation correlation between the two points. The role of this step is to construct a fusion feature matrix that comprehensively reflects the conduction performance and deformation characteristics of the steel structure through multi-source feature fusion, providing comprehensive feature input for subsequent micro-deformation calculations.

[0048] The specific implementation of step S08 is to use a pre-trained micro-vibration - micro-deformation model to process the fused feature matrix and calculate the micro-deformation matrix. First, the micro-deformation fused feature matrix is converted into a format suitable for model input. Considering the sparsity and symmetry of the feature matrix, a sparse matrix compression storage method is adopted, and only the non-zero elements and their indices in the upper triangular part of the matrix are retained. The compressed feature data is normalized to make the data distribution meet the input requirements of the model. The processed feature data is input into the pre-trained micro-vibration - micro-deformation model, which is a deep learning model with a graph convolutional neural network structure and includes multiple functional modules such as feature extraction, graph convolution, and non-linear mapping. The feature extraction module of the model extracts local and global features from the input fused feature matrix. The local features are extracted by the sliding window method with a window size of 3×3 and a step size of 1; the global features are extracted by the spectral decomposition method, and the first 10 main eigenvectors are retained. The data after feature extraction is sent to the graph convolution module, which contains 3 layers of graph convolutional layers, and the number of convolutional kernels in each layer is 32, 64, and 128 respectively. The activation function uses the ReLU function. The features after graph convolution are sent to the non-linear mapping module, which contains 2 layers of fully connected layers with the number of neurons being 256 and 128 respectively, and the activation function uses the Tanh function. The output layer of the model is a fully connected layer with the number of neurons being 3 times the number of sensors, corresponding to the deformation amounts in the x, y, and z directions at each sensor position. The model output result is reorganized into a micro-deformation matrix, which contains sensor indices and three spatial direction components of the deformation. To improve the accuracy of the deformation calculation result, physical constraint conditions are introduced for post-processing. The continuity constraint of structural deformation is applied to ensure that the deformation amounts of adjacent measurement points meet the continuity condition, and the deformation gradient should not exceed a preset threshold, usually 0.001; the principle of minimum structural deformation energy is applied to optimize the deformation field distribution by minimizing the overall structural deformation energy; the support constraint condition is applied. For the fixed support position, its deformation amount is restricted not to exceed the instrument measurement error range, usually 0.01 mm. After optimization by physical constraints, the final micro-deformation matrix is obtained, which characterizes the deformation amounts and directions of each monitoring point of the steel structure. The function of this step is to convert the micro-vibration features into structural deformation information through the micro-vibration - micro-deformation model, realizing high-precision calculation of the micro-deformation of the steel structure.

[0049] The specific implementation of step S09 is to calculate the micro-deformation key matrix based on the micro-deformation matrix to evaluate the impact of the deformation of key nodes on structural safety. First, extract the deformation data of key nodes from the micro-deformation matrix. Key nodes include positions that have an important impact on the overall stability of the structure, such as the connection between the main beam and the column, the intersection of the main and secondary beams, the support position points, and the position points of the span change section. Conduct an importance assessment of the deformation data of key nodes. The evaluation indicators include the absolute value of the deformation amount, the relative value of the deformation amount, the consistency between the deformation direction and the structural force direction, etc. The evaluation of the absolute value of the deformation amount uses a piecewise function, and different thresholds are set for different types of structural members. For example, for a main beam with a span of 30 meters, when the deformation amount is less than 5 mm, it is a low risk; when it is 5 - 15 mm, it is a medium risk; when it is greater than 15 mm, it is a high risk. The evaluation of the relative value of the deformation amount uses the ratio of the deformation amount to the member length. Usually, when the ratio of the deformation amount to the member length does not exceed 1 / 1000, it is taken as the safety standard. The evaluation of the deformation direction is to calculate the angle between the deformation vector and the main force direction of the structure. When the angle is less than 30 degrees, the risk coefficient is 1.0; when it is 30 - 60 degrees, the risk coefficient is 0.8; when it is greater than 60 degrees, the risk coefficient is 0.6. Calculate the comprehensive risk coefficient based on multiple evaluation indicators, using the weighted summation method. The weight of the absolute value of the deformation amount is 0.4, the weight of the relative value of the deformation amount is 0.4, and the weight of the deformation direction is 0.2. Construct a micro-deformation key matrix according to the comprehensive risk coefficient, which includes the key node index and the comprehensive risk coefficient of this node. Conduct a clustering analysis on the micro-deformation key matrix, using the K-means clustering algorithm, and divide the key nodes with similar risk coefficients into 3 - 5 levels, corresponding to different risk levels. To facilitate risk assessment and decision-making, generate a key node risk distribution map, using a heat map method. The color ranges from green to red, indicating that the risk coefficient increases from low to high, visually showing the risk distribution of the structure. At the same time, generate a micro-deformation time trend map, recording the change trend of the key node risk coefficient over time, and using the exponential smoothing method to predict the short-term risk development trend. The role of this step is to extract key node information from the overall micro-deformation matrix, evaluate the degree of influence of key node deformation on structural safety, and provide an intuitive basis for structural safety assessment and decision-making.

[0050] The specific implementation of step S10 is to evaluate the deformation state of the steel structure based on the micro-deformation key matrix, and determine the micro-deformation over-limit area and severity. First, a micro-deformation safety threshold system is established, including two types: absolute threshold and relative threshold. The absolute threshold sets the reference value according to the structure type, usage function, etc. For example, for the steel structure used in office buildings, the deflection of the main beam is controlled within L / 250 (L is the span), and the lateral displacement of the column is controlled within H / 500 (H is the storey height); the relative threshold is set according to the statistical characteristics of the monitoring historical data, and usually the statistical mean plus 3 times the standard deviation is taken as the abnormal judgment threshold. Compare the deformation data in the micro-deformation key matrix with the safety threshold to identify the over-limit area. For the area where the deformation amount exceeds 90% of the absolute threshold, it is marked as a yellow warning area; for the area where the deformation amount exceeds the absolute threshold, it is marked as an orange warning area; for the area where the deformation amount exceeds 120% of the absolute threshold, it is marked as a red warning area. Conduct spatial clustering analysis on the over-limit area, using the density-based spatial clustering algorithm to group the over-limit points that are spatially adjacent and have similar deformation characteristics into the same over-limit area. Calculate the deformation characteristic parameters for each over-limit area, including the maximum deformation amount, average deformation amount, deformation range, deformation derivative (deformation gradient), etc. Evaluate the over-limit severity based on the deformation characteristic parameters, using the fuzzy comprehensive evaluation method to divide the over-limit area into four levels: slight, medium, severe, and critical. Generate a micro-deformation over-limit area distribution map, using a three-dimensional solid graph method, with different colors representing different severities, intuitively showing the spatial distribution and severity of the over-limit area. At the same time, generate a micro-deformation over-limit development trend map, record the change trend of the over-limit area area and severity over time, and use the weighted moving average method to analyze the development trend, providing a basis for predicting the development of the over-limit area. According to the over-limit evaluation results, formulate corresponding disposal suggestions. For slightly over-limit areas, it is recommended to increase the monitoring frequency; for medium over-limit areas, it is recommended to conduct on-site inspections and formulate a monitoring plan; for severely over-limit areas, it is recommended to restrict use and conduct structural reinforcement evaluation; for critically over-limit areas, it is recommended to immediately stop using and conduct emergency reinforcement. The function of this step is to evaluate the deformation state of the steel structure by comparing the micro-deformation data with the safety threshold, determine the over-limit area and severity, and provide a scientific basis for structural safety management and decision-making.

[0051] The following are the detailed structure of the micro-vibration - micro-deformation model, the steps for establishing the training data set, and the specific implementation of the training steps: The specific implementation of the detailed structure of the micro-vibration - micro-deformation model is to construct a deep learning model that can infer the structural micro-deformation from the micro-vibration feature matrix. This model adopts a deep learning architecture based on graph convolutional networks, and its overall structure is divided into five main parts: a data preprocessing layer, a feature extraction layer, a graph convolutional layer, a non-linear mapping layer, and an output layer. In the data preprocessing layer, first, the input micro-deformation fusion feature matrix is normalized using the z-score normalization method to make the data mean 0 and the standard deviation 1, eliminating the influence of different feature dimensions and value ranges. Then, the normalized feature matrix is converted into a graph structure representation, where the nodes correspond to the sensor positions, the edges correspond to the micro-vibration conduction relationships between the nodes, and the weights of the edges are the values of the corresponding elements in the fusion feature matrix. In the feature extraction layer, first, a spatial convolution module is used to extract local features. A 3×3 convolution kernel is adopted, and a total of 32 convolution kernels are set to capture local spatial correlations. At the same time, a spectral decomposition module is used to extract global features. The feature matrix is subjected to singular value decomposition, and the first 16 main eigenvectors are retained as the global feature representation. The local features and global features are fused through an attention mechanism to form an enhanced feature representation. In the graph convolutional layer, a three-layer stacked graph convolutional network is adopted. Each layer of graph convolution is implemented using the Chebyshev polynomial expansion method, with the polynomial order set to 3. The numbers of graph convolution kernels are 64, 128, and 256 respectively. After each layer of graph convolution, a batch normalization layer and a ReLU activation function are connected. The graph convolutional layer can make full use of the topological relationship between the nodes to learn the context features of the nodes. In the non-linear mapping layer, a four-layer fully connected network is adopted. The numbers of neurons are 512, 256, 128, and 64 respectively. The activation function is LeakyReLU, with the slope coefficient set to 0.2. After each layer of full connection, a Dropout layer is connected, and the dropout rate is set to 0.3 to prevent overfitting. In the output layer, a fully connected network is used to map the features to the deformation prediction results. The output dimension is 3 times the number of sensors, corresponding to the deformation amounts in three spatial directions at each sensor position. At the same time, an auxiliary deformation validity judgment branch is introduced to output the confidence of each deformation value for evaluating the reliability of the prediction results. In addition, a physical constraint module is added to the model. By introducing constraint conditions based on physical laws, such as structural continuity, support constraints, etc., the physical rationality of the prediction results is improved. The model adopts a residual connection structure, and a skip connection is added after each layer of graph convolution to alleviate the problem of gradient disappearance and accelerate the convergence of the model. Finally, the model outputs a vector representing the micro-deformation amounts at each monitoring point, which is reorganized to form a micro-deformation matrix. The characteristics of this model are that it makes full use of the graph structure characteristics of the micro-vibration signal, fuses spatial local features and global structure features, and combines physical constraints to achieve a high-precision mapping from micro-vibration features to micro-deformations.

[0052] The specific implementation of the steps for establishing the training dataset of the micro-vibration and micro-deformation model is to construct a high-quality micro-vibration and micro-deformation paired training dataset through various means. First, laboratory calibration data is collected. In the laboratory environment, steel structure test platforms of different types and sizes are built, including basic components such as beams, columns, frames, and complex combined structures. A micro-vibration sensor network is arranged on the test platform, and the sensor layout density is higher than that of actual projects to obtain more comprehensive micro-vibration data. At the same time, a high-precision optical displacement measurement system (such as a laser rangefinder, an optical deformation measurement system, etc.) is used as a reference standard to measure the actual deformation of the structure. Different types and intensities of excitations are applied on the test platform, including point excitation, distributed excitation, random excitation, etc., to simulate various external actions that may cause structural deformation. The micro-vibration signals collected under each excitation condition are paired with the corresponding deformation data to form a basic calibration dataset. Second, numerical simulation data is generated. A refined finite element model of the steel structure is established, and the model parameters are calibrated through the test data of the actual structure to ensure that the model can accurately reflect the dynamic characteristics of the actual structure. Based on the calibrated finite element model, the dynamic responses of the structure under different excitation conditions are simulated, and the acceleration responses and displacement responses of each node are calculated. The acceleration response data is processed to extract the micro-vibration feature matrix; the displacement response data is processed to form a reference micro-deformation matrix. Through a large number of numerical simulations, micro-vibration and micro-deformation paired data covering various working conditions are generated. Then, actual project data is collected. On the actual steel structure where the health monitoring system has been arranged, a high-precision deformation monitoring device (such as a fiber Bragg grating sensor, a high-precision inclinometer, etc.) and a micro-vibration sensor network are installed simultaneously. Under normal use conditions and special working conditions (such as wind load, temperature change, human flow excitation, etc.), micro-vibration data and deformation data are collected synchronously. The collected raw data is preprocessed, including denoising, filtering, outlier processing, etc., to ensure the data quality. The micro-vibration feature matrix and the micro-deformation matrix are extracted from the preprocessed data to form an actual project paired dataset. Finally, data augmentation and synthesis are carried out. The existing calibration data and actual project data are subjected to data augmentation processing, including adding noise, adjusting the signal intensity, applying random perturbations, etc., to increase data diversity. Based on the physical model and the existing data, synthetic data is generated to fill the working condition gaps in the actual project data. The semi-supervised learning method is adopted, and the labeled data is used to guide the annotation of the unlabeled data to expand the scale of the training dataset. All source data is integrated and cleaned, and the obvious error or low-quality data samples are removed to ensure the quality of the training dataset. The finally constructed training dataset includes four parts: calibration data, numerical simulation data, actual project data, and synthetic data, with a total number of samples of not less than 10,000 groups, covering various structural types, sizes, and working condition conditions.

[0053] The specific implementation of the steps for micro-vibration and micro-deformation model training adopts a multi-stage and multi-objective training strategy to ensure that the model has high precision and good generalization ability. First, data preprocessing and segmentation are carried out. The constructed micro-vibration and micro-deformation paired data set is preprocessed, including outlier detection and processing, feature standardization, missing value filling, etc. The data set is divided into a training set, a validation set, and a test set according to the ratio of 7:2:1 to ensure that the three subsets have similar data distribution characteristics. Then, model initialization is carried out. The weights of the graph convolutional network are initialized, and the Xavier initialization method is adopted to make the initial weights satisfy the distribution with a mean of 0 and a variance of 2 / (input dimension + output dimension). The He initialization method is used for the weights of the fully connected layer to make the initial weights satisfy the distribution with a mean of 0 and a variance of 2 / input dimension. Next, the first stage of training, that is, the pre-training stage, is carried out. A subset composed of numerical simulation data and calibration data is used for training without adding physical constraints, and the focus is on optimizing the basic expression ability of the model. The mean square error is used as the main loss function, the learning rate is set to 0.001, the Adam optimizer is adopted, the batch size is set to 64, and the number of training epochs is 200. Then, the second stage of training, that is, the fine-tuning stage, is carried out. Based on the pre-trained model in the first stage, actual engineering data is introduced, and at the same time, physical constraint terms are added. The loss function is a weighted composite loss, including three parts: mean square error loss, physical constraint loss, and regularization loss, and the weights of each part are 0.6, 0.3, and 0.1 respectively. The learning rate is adjusted to 0.0005, and a learning rate decay strategy is adopted, with the learning rate decaying to 0.8 times the original every 50 epochs. The batch size is adjusted to 32 to adapt to more complex loss calculations. The number of training epochs is 300, or until the validation set loss does not significantly decrease for 20 consecutive epochs. Then, the third stage of training, that is, the ensemble optimization stage, is carried out. Multiple models with different structures or initializations are trained to form a model ensemble. Each sub-model is trained using different data subsets or different hyperparameter settings. The prediction results of multiple sub-models are combined through ensemble learning methods (such as Bagging, Stacking, etc.) to improve the overall prediction accuracy and stability. Finally, model evaluation and deployment are carried out. The model performance is evaluated on the test set, and the evaluation metrics include quantitative metrics such as root mean square error, mean absolute error, coefficient of determination, etc., and qualitative metrics such as the physical rationality of the deformation distribution. For the best-performing model or model ensemble, model compression and optimization are carried out, including techniques such as knowledge distillation and weight quantization, to reduce the model size and improve the inference speed. The optimized model is exported in a format suitable for deployment and integrated and verified in an actual monitoring system. During the training process, an early stopping strategy is adopted. When the validation set loss does not significantly decrease for 15 consecutive epochs, the current stage of training is terminated in advance. A model checkpoint saving strategy is adopted to save the model parameters with the best validation set performance. At the same time, to improve the training efficiency, GPU acceleration computing is adopted, and the batch size is adjusted according to the GPU memory capacity.

[0054] Through the micro-vibration and micro-deformation model structure design, training dataset establishment, and model training process described in detail above, a deep learning model that can accurately map the relationship between micro-vibration characteristics and structural micro-deformation can be constructed. This model fully utilizes the advantages of graph convolutional networks in processing structured data, combines physical constraints to improve prediction rationality, and ensures model performance and generalization ability through a multi-stage training strategy. The multi-source construction method of the training dataset not only guarantees the accuracy and representativeness of the data but also expands the coverage through data augmentation. Finally, based on the micro-vibration signal feature matrix, this model can accurately predict the micro-deformation state of steel structures, providing important technical support for the health monitoring and safety assessment of steel structures. The establishment and application process of the entire micro-vibration and micro-deformation model form a complete technical chain, with strong engineering practicability and popularization value.

[0055] The following details the mathematical models or calculation processes involved in the present invention.

[0056] The calculation of the micro-vibration conduction matrix is specifically expressed as follows:

[0057]

[0058] In the formula, C ij is the micro-vibration conduction coefficient from sensor node i to node j; v ij is the propagation speed of micro-vibration between node i and node j (m / s); v ref is the reference value of the theoretical propagation speed of micro-vibration in steel (5000 m / s); t ij is the time (s) for micro-vibration to propagate from node i to node j; t ref is the reference propagation time (s), usually taking t ref = d ij / v ref where d ij is the spatial distance (m) between nodes i and j; η ij is the stress transfer efficiency coefficient, with a value range of [0, 1]; α, β, and γ are weight coefficients, satisfying α + β + γ = 1, usually taking α = 0.4, β = 0.3, and γ = 0.3.

[0059] Among them, the parameter acquisition method is:

[0060] v ij is calculated through the following formula:

[0061]

[0062] t ij is obtained through cross-correlation analysis, and the specific calculation is as follows:

[0063] tij = argmax τ R ij (τ);

[0064] Wherein, R ij (τ) is the cross - correlation function of the micro - vibration signals of node i and node j, and the calculation formula is:

[0065]

[0066] Among them, x i (t) and x j (t) are the micro - vibration signals at node i and node j respectively, T is the length of the signal acquisition time window, and τ is the time delay.

[0067] η ij is calculated by the following formula:

[0068]

[0069] Wherein, E i and E j are the micro - vibration signal energies at node i and node j respectively, λ is the attenuation coefficient, and usually λ = 0.05.

[0070] The construction of the micro - vibration conduction matrix C considers three aspects: the propagation speed, propagation time, and energy transfer efficiency of the micro - vibration signal in the structure, reflecting the characteristics of the internal stress transfer path of the structure. The propagation speed term characterizes the elastic properties of the structural material, the propagation time term reflects the time efficiency of micro - vibration propagation, and the energy transfer efficiency term characterizes the ability of micro - vibration energy to be maintained during propagation. These three indicators together constitute a comprehensive characterization of the structural connection state and overall stability.

[0071] The calculation of the micro - vibration conduction attenuation matrix is specifically expressed as follows:

[0072]

[0073] Wherein, A ij is the micro - vibration energy attenuation coefficient (1 / meter) from sensor node i to node j; d ij is the spatial distance (meter) between node i and j; P i and P j are the micro - vibration signal energies at node i and node j respectively; δ ij is the attenuation correction term used to correct the energy attenuation deviation in non - uniform structures.

[0074] Among them, the parameter acquisition method is:

[0075] P i and P jCalculated by the following formula:

[0076]

[0077] Wherein, S i (f) and S j (f) are the energy spectral density functions of the micro-vibration signals at nodes i and j respectively, f 1 and f 2 are the frequency ranges of interest. For steel structures, usually f 1 = 0.5 Hz, f 2 = 10 Hz.

[0078] δ ij Calculated by the following formula:

[0079]

[0080] Wherein, is the theoretical attenuation coefficient, and its value range is [-0.15, -0.05]; is the initial attenuation coefficient obtained by measurement; ΔT ij is the temperature difference (in degrees Celsius) between nodes i and j; ξ ij is the connection state coefficient, and its value range is [-0.1, 0.1]; k 1 , k 2 , k 3 are the correction weights. Usually, k 1 = 0.6, k 2 = 0.3, k 3 = 0.1.

[0081] Perform spatial smoothing on the initial attenuation matrix:

[0082]

[0083] Wherein, is the smoothed attenuation coefficient; N i and N j are the neighborhood sets of nodes i and j respectively; w mn is the spatial weight coefficient, which is inversely proportional to the distance; is the weight normalization factor.

[0084] The micro-vibration conduction attenuation matrix A characterizes the attenuation characteristics of micro-vibration energy during the transmission process in the structure, reflecting the state of the structural materials and connection parts. The relationship between energy attenuation and propagation distance is in logarithmic form, conforming to the energy attenuation law in wave propagation. The correction term takes into account three factors: the deviation between the theoretical and measured values, the temperature influence, and the connection state, improving the accuracy of the attenuation characteristic representation. The spatial smoothing process utilizes the correlation of attenuation characteristics between adjacent parts in the structure, reducing the noise influence of discrete measurement points. By analyzing the attenuation characteristics of micro-vibration energy in the structure, potential structural defects or damage locations can be identified, providing an important basis for structural health monitoring.

[0085] The calculation of the micro-vibration conduction resonance matrix is specifically expressed as follows:

[0086]

[0087] In the formula, R ij (f) is the micro-vibration conduction resonance coefficient from sensor node i to node j at frequency f; H ij (f) is the frequency response function from node i to node j; H ref (f) is the reference frequency response function, usually taken as the frequency response function under ideal conditions; φ ij (f) is the phase consistency function, and its value range is [0, 1].

[0088] Among them, the parameter acquisition method is:

[0089] H ij (f) is calculated by the following formula:

[0090]

[0091] In the formula, S xi (f) and S xj (f) are the power spectral densities of the micro-vibration signals at nodes i and j respectively.

[0092] φ ij (f) is calculated by the following formula:

[0093]

[0094] In the formula, Δθ ij (f) is the phase difference of the micro-vibration signals at nodes i and j at frequency f.

[0095] For multiple frequency points, the micro-vibration conduction resonance matrix can be expressed as a three-dimensional tensor R, where R ij，k = R ij (f k ), f k is the kth frequency point.

[0096] For a specific frequency f k , a two-dimensional matrix R (k) can be obtained, where

[0097] The micro-vibration conduction resonance matrix R describes the resonance conduction characteristics between nodes of the structure at different frequencies. It is a three-dimensional tensor that characterizes the frequency-domain dynamic response characteristics of the structure. This matrix takes into account two factors: the ratio of frequency response functions and phase consistency. The ratio of frequency response functions reflects the vibration transfer characteristics of the structure at a specific frequency, while the phase consistency function characterizes the synchronism of vibration transfer. By analyzing this matrix, the natural frequencies and modal characteristics of the structure can be identified, the performance of the structure under dynamic excitation can be evaluated, and potential structural weaknesses can be identified. When the structure shows abnormalities or damage, its resonance characteristics will change, manifested as abnormalities in the values of the resonance matrix elements. Therefore, this matrix is an important indicator for structural health monitoring.

[0098] The calculation of the resonance peak matrix is specifically expressed as follows:

[0099]

[0100] In the formula, P ij is the comprehensive resonance peak value from sensor node i to node j; R ij (f k ) is the micro-vibration conduction resonance coefficient from node i to node j at frequency f k ; is the resonance peak identification threshold, usually taken as where is the average value of the resonance coefficients at all frequency points; is the weight coefficient of frequency f k , which is related to the importance of the corresponding mode at this frequency point; K is the number of frequency points considered.

[0101] The calculation of the resonance valley matrix is specifically expressed as follows:

[0102]

[0103] In the formula, V ij is the comprehensive resonance valley value from sensor node i to node j; is the resonance valley identification threshold, usually taken as is the weight coefficient of frequency f k , which is related to the importance of the corresponding mode at this frequency point.

[0104] Among them, the method for obtaining parameters is as follows:

[0105] The frequency point f k is obtained through the following steps:

[0106] 1. Perform peak search and valley search on the frequency response curves of each measurement point;

[0107] 2. Conduct spatial clustering on the identified peaks and valleys to obtain the main resonance frequencies and suppression frequencies of the structure;

[0108] 3. Select the frequency points with higher importance as f k , usually 5 - 10 main frequency points are taken.

[0109] Weight coefficient and are calculated by the following formula:

[0110]

[0111] In the formula, λ k is the eigenvalue corresponding to the frequency f k , representing the importance of the mode corresponding to this frequency.

[0112] The resonance peak matrix P and the resonance valley matrix V respectively characterize the resonance enhancement characteristics and resonance suppression characteristics of the structure at specific frequencies. The resonance peak matrix reflects the frequencies and transmission paths where resonance is likely to occur in the structure, corresponding to the dangerous frequencies of the structure; the resonance valley matrix reflects the frequencies and transmission paths where resonance is not likely to occur in the structure, corresponding to the safe frequencies of the structure. The calculation of these two matrices considers the peaks and valleys on the frequency response curve, and through threshold screening and weight weighting, highlights the frequency points that have an important impact on the dynamic behavior of the structure. Decomposing the resonance characteristics into peak and valley parts is conducive to more precisely characterizing the frequency-domain dynamic characteristics of the structure and providing a basis for the dynamic analysis and safety assessment of the structure.

[0113] The calculation of the micro - deformation fusion feature matrix is specifically expressed as follows:

[0114] F ij = w c ·C ij + w a ·(1 - |A ij | norm ) + w p ·P ij + w v ·V ij ;

[0115] In the formula, F ij is the micro - deformation fusion eigenvalue from sensor node i to node j; C ij is the element of the micro - vibration conduction matrix; |A ij | norm is the absolute value of the element of the normalized micro - vibration conduction attenuation matrix; P ijis the formant matrix element; V ij is the formant valley matrix element; w c , w a , w p , w v are the weight coefficients, satisfying w c + w a + w p + w v = 1. Usually, w c = 0.35, w a = 0.25, w p = 0.25, w v = 0.15.

[0116] Apply spatial filtering to the preliminary fusion feature matrix:

[0117]

[0118] where is the fused feature value after filtering; N i and N j are the neighborhood sets of nodes i and j respectively; G(d mn , σ) is the Gaussian kernel function, d mn is the distance between nodes m and n, and σ is the Gaussian kernel parameter. Usually, σ is taken as 0.1 times the minimum feature size of the structure.

[0119] Apply non - linear enhancement to the filtered fusion matrix:

[0120]

[0121] The micro - deformation fusion feature matrix F * Comprehensively characterizes the conduction performance and deformation correlation between monitoring points of the structure by weighted - fusing the micro - vibration conduction matrix, micro - vibration conduction attenuation matrix, formant matrix and formant valley matrix. The fusion adopts the linear weighted method, considering the contributions of four different feature matrices, and the weight distribution is based on the importance of each feature for deformation characterization. The spatial filtering process uses the Gaussian kernel function to smooth the feature values of adjacent nodes, reducing the influence of local noise and outliers. The non - linear enhancement process applies different power functions according to the size of the feature values, so that the high - value region is appropriately suppressed and the low - value region is appropriately enhanced, improving the contrast of the features. This multi - source feature fusion method makes full use of various characteristics of the micro - vibration signal in the time domain and frequency domain, constructs a fusion feature matrix that comprehensively reflects the conduction performance and deformation characteristics of the steel structure, and provides high - quality feature input for subsequent micro - deformation calculation.

[0122] The calculation of the micro - deformation matrix is realized through the micro - vibration - micro - deformation model, and its core calculation process is specifically expressed as follows:

[0123] D = f GCN (F * );

[0124] Where D is the micro - deformation matrix, representing the deformation amounts of each monitoring point in three spatial directions; F * is the micro - deformation fusion feature matrix; f GCN is the micro - vibration - micro - deformation mapping model based on the graph convolutional network.

[0125] The calculation formula of the graph convolutional layer of this model is:

[0126]

[0127] Where H (l) is the node feature matrix of the l - th layer; is the adjacency matrix with self - connection added; is the degree matrix of W (l) is the weight matrix of the l - th layer; σ is the activation function, usually the ReLU function.

[0128] Apply physical constraint conditions to optimize the deformation matrix:

[0129]

[0130] Where D * is the optimized micro - deformation matrix; ||·|| F is the Frobenius norm; E cont is the continuity constraint term; E energy is the deformation energy constraint term; E support is the support constraint term; λ 1 , λ 2 , λ 3 are the constraint weight coefficients.

[0131] The calculation formula of the continuity constraint term is:

[0132]

[0133] Where d i ′ is the deformation vector of node i; is the deformation gradient at node i; w ij is the spatial correlation weight between node i and node j.

[0134] The calculation formula of the deformation energy constraint term is:

[0135] E energy (D′) = ∑ i d i ′T K i d i ′;

[0136] In the formula, K i is the stiffness matrix at node i.

[0137] The calculation formula for the support constraint term is:

[0138] E support (D′)=∑ i∈s ||d i ′|| 2 ;

[0139] In the formula, S is the set of support nodes.

[0140] The micro-deformation matrix D * represents the deformation amount and direction of each monitoring point of the steel structure, and is calculated from the fused feature matrix through the micro-vibration - micro-deformation model. This model adopts a graph convolutional network structure, which can make full use of the topological relationship between nodes for feature learning. The calculation formula of the graph convolutional layer is based on spectral graph theory, and through the transformation of the adjacency matrix, the aggregation and update of node features are realized. Physical constraint optimization combines three physical laws: continuity constraint, deformation energy constraint, and support constraint, improving the physical rationality of the deformation calculation results. The continuity constraint ensures that the deformation amounts of adjacent measuring points satisfy smooth transition; the deformation energy constraint reflects the principle of minimum structural deformation energy; the support constraint reflects the displacement constraint conditions of the fixed support points of the structure. Through this comprehensive method based on deep learning and physical laws, a high-precision mapping from micro-vibration features to structural deformation is achieved.

[0141] The calculation of the micro-deformation key matrix is specifically expressed as follows:

[0142]

[0143] In the formula, K i is the micro-deformation risk coefficient of the key node i; is the optimized micro-deformation vector of node i; S abs is the absolute value evaluation function of the deformation amount; S rel is the relative value evaluation function of the deformation amount, L i is the characteristic length of the member where node i is located; S dir is the deformation direction evaluation function, f i is the main stress direction at node i; w abs 、w rel 、w dir are weight coefficients, satisfying w abs +w rel +w dir =1, usually taking w abs =0.4, wrel = 0.4, w dir = 0.2.

[0144] The calculation formula for the absolute value evaluation function of the deformation amount is:

[0145]

[0146] In the formula, ||d|| is the modulus length of the deformation vector; d low , d high , d crit are the deformation thresholds corresponding to low risk, high risk, and critical risk respectively. For the main beam with a span of 30 meters, usually take d low = 5 mm, d high = 15 mm, d crit = 30 mm.

[0147] The calculation formula for the relative value evaluation function of the deformation amount is:

[0148]

[0149] In the formula, t is the characteristic length of the component, and the value is the actual length of the component (mm).

[0150] The calculation formula for the deformation direction evaluation function is:

[0151]

[0152] In the formula, θ is the included angle between the deformation vector d and the main stress direction f, and the calculation formula is:

[0153]

[0154] The micro-deformation key matrix K represents the deformation amount of the key nodes of the steel structure and its influence degree on the structural safety. It is a subset of the micro-deformation matrix and is used to evaluate the structural safety. This matrix comprehensively evaluates the deformation of the key nodes through three evaluation indicators: the absolute value evaluation of the deformation amount reflects the relationship between the deformation amount and the preset safety threshold; the relative value evaluation of the deformation amount considers the proportional relationship between the deformation amount and the component size; the deformation direction evaluation considers the consistency between the deformation direction and the structural stress direction. The design of the piecewise function makes the evaluation results have different sensitivities in different deformation intervals, and the closer the score is to 1, the higher the risk. Through this multi-index comprehensive evaluation method, the influence of the deformation of the key nodes on the structural safety can be evaluated more comprehensively, providing a scientific basis for structural safety evaluation and decision-making.

[0155] The evaluation of the micro-deformation over-limit area is specifically expressed as follows:

[0156]

[0157] In the formula, Oi is the overrun degree of node i; K i is the micro-deformation risk coefficient of node i; K th is the safety threshold, usually taken as K th = 0.7.

[0158] The severity evaluation formula for the overrun area is:

[0159] S region = ∑ i∈R w i ·O i ;

[0160] In the formula, S region is the overall severity of the overrun area; R is the set of nodes in the overrun area; w i is the weight coefficient of node i, related to the structural importance of the node.

[0161] The spatial distribution characteristic evaluation formula for the overrun area is:

[0162]

[0163] In the formula, φ region is the spatial aggregation degree of the overrun area; n is the number of nodes in the overrun area; d ij is the distance between node i and node j; σ is the scale parameter, usually taken as 0.2 times the minimum characteristic size of the structure.

[0164] The development trend evaluation formula for the overrun area is:

[0165] T region (t) = α·S region (t) + (1 - α)·T region (t - 1);

[0166] In the formula, T region (t) is the overrun trend index at time t; S region (t) is the overrun severity at time t; α is the smoothing coefficient, usually taken as α = 0.3.

[0167] The comprehensive risk level evaluation formula for the overrun area is:

[0168]

[0169] In the formula, R level is the comprehensive risk level of the overrun area; is the change rate of the overrun trend; μ 1 、μ 2 、μ 3 are weight coefficients, satisfying μ 1 + μ 2+μ 3 = 1, usually take μ 1 = 0.5, μ 2 = 0.3, μ 3 = 0.2.

[0170] The micro - deformation over - limit area evaluation system comprehensively evaluates the over - limit area through multiple indicators. The over - limit degree O i represents the degree to which the node deformation risk coefficient exceeds the safety threshold, and uses normalization processing to make the evaluation value fall between 0 and 1. The overall severity S region of the over - limit area considers the contributions of all over - limit nodes in the area, and the node weight reflects the importance of different nodes to the structure. The spatial aggregation degree Φ region describes the aggregation characteristics of over - limit nodes in spatial distribution. High aggregation degree usually means local structural problems. The development trend evaluation uses the exponential smoothing method, which can smooth short - term fluctuations and reflect long - term trends. The comprehensive risk level evaluation comprehensively considers three aspects: severity, spatial distribution, and development trend, providing a comprehensive risk assessment for structural safety management. This multi - dimensional evaluation method makes the risk assessment of the structural over - limit area more objective and comprehensive, and helps to formulate targeted disposal measures.

[0171] Optionally, the core structure of the micro - vibration - micro - deformation model is specifically expressed as follows:

[0172] The calculation formula of the feature pre - processing layer is:

[0173]

[0174] In the formula, F norm is the normalized micro - deformation fusion feature matrix; μ F* and σ F* are the mean and standard deviation of the feature matrix respectively.

[0175] Optionally, the local feature extraction formula of the feature extraction layer is:

[0176] L = Conv2D(F norm , W conv );

[0177] In the formula, L is the local feature matrix; Conv2D is the two - dimensional convolution operation; W conv is the convolution kernel parameter matrix.

[0178] Optionally, the global feature extraction formula of the feature extraction layer is:

[0179]

[0180] In the formula, G is the global feature matrix; U, Σ, V are for F normThe result of singular value decomposition; k is the number of main features to be retained, usually k = 16.

[0181] Optionally, the calculation formula for the attention mechanism of feature fusion is:

[0182] A = softmax(W a ·[L||G]);

[0183] F enhanced = A⊙[L||G];

[0184] In the formula, A is the attention weight matrix; W a is the attention weight parameter; [L||G] is the concatenation of local features and global features; ⊙ is the element-wise product; F enhanced为 is the enhanced feature matrix.

[0185] Optionally, the calculation formula for the graph convolutional layer based on the Chebyshev polynomial expansion is:

[0186]

[0187] In the formula, H (l) is the node feature matrix of the l-th layer; T k is the k-th order Chebyshev polynomial; is the normalized Laplacian matrix, is the graph Laplacian matrix; is the parameter matrix of the k-th order polynomial of the l-th layer; σ is the activation function.

[0188] Optionally, the calculation formula for the non-linear mapping layer is:

[0189] Z (l+1) = LeakyReLU(W (l) Z (l) + b (l) );

[0190] In the formula, Z (l) is the feature matrix of the l-th layer; W (l) and b (l) are the weight matrix and bias vector of the l-th layer respectively;

[0191] LeakyReLU(x) = max(0.2x, x) is the ReLU activation function with leakage.

[0192] Optionally, the deformation prediction formula for the output layer is:

[0193] d = W out Z (L) + b out ;

[0194] where d is the predicted deformation vector; W out and b out are the weight matrix and bias vector of the output layer respectively; Z (L) is the output of the last non - linear mapping.

[0195] The calculation formula for the deformation validity judgment branch is:

[0196] c = sigmoid(W con fZ (L) + b conf );

[0197] where c is the confidence vector of deformation prediction; W conf and b conf are the weight matrix and bias vector of the confidence branch respectively; is the sigmoid activation function.

[0198] Optionally, the calculation formula of the physical constraint module is:

[0199]

[0200] where is the physical constraint loss; are the continuity constraint loss, energy constraint loss and support constraint loss respectively; λ 1 , λ 2 , λ 3 are the weight coefficients.

[0201] Optionally, the overall loss function of the model is:

[0202]

[0203] where is the total loss; is the mean square error loss; is the regularization loss; α and β are the weight coefficients.

[0204] Optionally, the micro-vibration - micro-deformation model adopts a deep learning architecture based on graph convolutional networks, which includes five main parts: feature preprocessing, feature extraction, graph convolution, non-linear mapping, and output. Feature preprocessing uses a standardization method to eliminate the influence of different feature dimensions and numerical ranges. The feature extraction layer extracts local features through two-dimensional convolution, extracts global features through singular value decomposition, and uses an attention mechanism for feature fusion. The graph convolution layer is implemented based on the Chebyshev polynomial expansion, which can make full use of the topological relationship between nodes for feature learning. The non-linear mapping layer uses a leaky ReLU activation function to enhance the model's ability to process negative-valued features. The output layer predicts the deformation of each monitoring point, and at the same time introduces a deformation effectiveness judgment branch to evaluate the reliability of the prediction result. The physical constraint module improves the physical rationality of the prediction result by introducing constraint conditions based on physical laws. The training of the model uses a multi-objective loss function, comprehensively considering three aspects: prediction accuracy, physical constraints, and regularization. This method combining deep learning and physical constraints realizes a high-precision mapping from micro-vibration features to structural micro-deformations.

[0205] Specifically, the principle of the present invention is as follows: The technical principle of the present invention is based on the comprehensive application of structural vibration theory, material mechanics, and signal processing theory. By performing multi-dimensional analysis on micro-vibration signals, an association model between micro-vibration characteristics and micro-deformations is established to achieve real-time and accurate detection of micro-deformations of large steel structures.

[0206] From a physical mechanism perspective, when a steel structure is subjected to an external force, it will produce small deformations. These deformations change the stiffness distribution and mass distribution of the structure, thereby affecting the dynamic characteristics of the structure. When the structure is subjected to environmental excitations (such as wind loads, traffic loads, etc.), it will generate micro-vibration responses, and these micro-vibration responses contain rich information about the structure state. There is a deterministic relationship between structural deformation and micro-vibration response: deformation causes changes in structural parameters, parameter changes affect vibration characteristics, and vibration characteristics are reflected in micro-vibration signals. Based on this physical relationship, the present invention infers the micro-deformation state of the structure by carefully analyzing the characteristic changes of micro-vibration signals.

[0207] The micro-vibration conduction matrix is one of the core concepts of the present invention, which describes the propagation law of micro-vibration signals among various nodes of the structure. According to wave theory, micro-vibrations propagate in the structure in the form of waves, and the propagation characteristics are affected by the structural geometry, material properties, and connection states. When the structure undergoes micro-deformations, these propagation characteristics will change, manifested as variations in the micro-vibration conduction matrix. The present invention constructs a micro-vibration conduction matrix by synchronously collecting multi-point micro-vibration signals with high precision, and realizes sensitive detection of micro-deformations by analyzing its changes. Compared with traditional methods, this analysis based on conduction characteristics can capture the global structural response changes caused by deformations, rather than being limited to local deformations.

[0208] The micro-vibration conduction attenuation matrix reflects the attenuation law of micro-vibration energy in the structure. According to the principle of energy conservation, micro-vibration energy will attenuate due to factors such as material damping and joint friction during the propagation process. When the structure undergoes micro-deformation, the local stiffness and damping characteristics will change, resulting in a change in the energy attenuation law, which can be accurately captured by the micro-vibration conduction attenuation matrix. Especially for micro-deformations at the connection parts, such as bolt loosening and welding cracks, they often lead to obvious changes in the energy attenuation characteristics, and the detection of such hidden deformations can be achieved through the analysis of the attenuation matrix.

[0209] The micro-vibration resonance characteristic matrix, as well as the resonance peak matrix and resonance valley matrix decomposed therefrom, reflect the frequency response characteristics of the structure. According to the theory of structural dynamics, the natural frequencies and modes of the structure are closely related to its stiffness and mass distribution. Micro-deformation will cause a change in local stiffness, which in turn leads to a small shift in the natural frequency of the structure and a change in the mode shape. Through fine frequency-domain analysis, the present invention identifies such small changes and provides frequency-domain features for micro-deformation detection. The resonance peak matrix reflects the characteristic of enhanced vibration of the structure at a specific frequency, corresponding to the dangerous frequency of the structure; the resonance valley matrix reflects the vibration suppression characteristic, corresponding to the safe frequency. Through the comprehensive analysis of these two matrices, the change in resonance characteristics caused by micro-deformation can be identified, improving the sensitivity of detection.

[0210] Another key innovation of the present invention is the multi-dimensional feature fusion mechanism. By fusing the micro-vibration conduction matrix, conduction attenuation matrix, resonance peak matrix, and resonance valley matrix, a micro-deformation fusion feature matrix is constructed, comprehensively considering the indication effects of conduction characteristics, attenuation characteristics, and resonance characteristics on micro-deformation. This multi-dimensional feature fusion method avoids the limitations of single features and improves the accuracy and anti-interference ability of detection. Based on the fusion feature matrix, using a pre-trained micro-vibration - micro-deformation model, the micro-vibration features are mapped into the micro-deformation space to achieve the quantitative calculation of micro-deformation. Compared with traditional methods, the present invention more comprehensively considers the multi-dimensional features of micro-vibration signals, establishes the internal connection between micro-vibration characteristics and micro-deformation, and makes the detection results more accurate and reliable. At the same time, the technical route adopted by the present invention is non-invasive, real-time, and anti-interference, and can achieve the precise detection of micro-deformation in a complex environment. Through the early detection of micro-deformation, potential risks of the structure can be discovered in time, major safety accidents can be prevented, and technical support is provided for the safe operation of steel structures.

[0211] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention.

Claims

1. A method for detecting micro-deformation of large steel structures, characterized in that: The following steps are involved: Install multiple micro-vibration sensors on the large steel structure, where the micro-vibration sensors are distributed at key nodes of the steel structure to collect micro-vibration signals of the steel structure; Using the micro-vibration sensor to synchronously collect micro-vibration signals of the steel structure at different time points to form a micro-vibration signal set; constructing a micro-vibration conduction matrix based on the micro-vibration signal set, wherein the micro-vibration conduction matrix represents the conduction relationship of micro-vibration signals between nodes of the steel structure; Calculating a micro-vibration conduction attenuation matrix according to the micro-vibration conduction matrix, wherein the micro-vibration conduction attenuation matrix represents the energy attenuation characteristics of the micro-vibration signal transmission process in the steel structure; Performing frequency domain analysis on the micro-vibration signal set, identifying the resonance characteristics of the steel structure, and constructing a micro-vibration conduction resonance matrix; decomposing the micro-vibration conduction resonance matrix into a resonance peak matrix and a resonance valley matrix; Based on the micro-vibration conduction matrix, the micro-vibration conduction attenuation matrix, the resonance peak matrix and the resonance valley matrix, a micro-deformation fusion feature matrix of the steel structure is constructed; The micro-deformation fusion feature matrix is ​​processed by using a pre-trained micro-vibration-micro-deformation model to calculate a micro-deformation matrix; Calculate a micro-deformation key matrix according to the micro-deformation matrix; Based on the micro-deformation key matrix, the deformation state of the steel structure is evaluated to determine the micro-deformation exceeding limit area and severity.

2. The method for detecting micro-deformation of large steel structures according to claim 1, characterized in that: The key nodes refer to the connection points in the large-span steel structure that have an important impact on the overall stability and force transmission performance of the structure, including the connection points between the main beam and the column of the steel structure, the intersection points of the main and secondary beams, the support position points, and the span change section position points; the micro-vibration sensor has a three-axis acceleration measurement capability and can simultaneously collect micro-vibration signals in the x, y, and z directions; The micro-vibration sensor is firmly fixed on the surface of the steel structure by a rigid connection method to ensure that the vibration transmission between the sensor and the structure is not attenuated; for large steel structures, the sensor distribution density is 1 sensor per 50-100 square meters, and the density is appropriately increased at key nodes.

3. The method for detecting micro-deformation of large steel structures according to claim 2, characterized in that: The use of the micro-vibration sensor to synchronously collect micro-vibration signals of steel structures at different time points includes: under normal working conditions, the system performs full-network synchronous collection for 10 minutes every 4 hours; under special working conditions, when the vibration acceleration detected by any sensor exceeds a preset threshold, the full-network synchronous collection is automatically triggered, and the duration is 30 seconds before the event to 10 minutes after the event; the collected original micro-vibration signals are preprocessed, including signal noise reduction, baseline correction and outlier processing; the preprocessed micro-vibration signals of each time window are organized according to the collection time and sensor position to form a four-dimensional micro-vibration signal set.

4. The large-scale steel structure micro-deformation detection method according to claim 3 is characterized in that: Constructing the microvibration conduction matrix includes: performing cross-correlation analysis on the signals between each pair of sensor nodes in the preprocessed microvibration signal set and calculating the cross-correlation function; constructing the initial microvibration conduction time matrix based on the propagation time of all node pairs; reconstructing the conduction time matrix using the shortest path algorithm; calculating the propagation velocity matrix of microvibrations in the structure based on the optimized conduction time matrix and the spatial distance between sensors; calculating the stress transfer efficiency matrix through the relationship between the propagation velocity matrix and the structural material properties; and constructing the microvibration conduction matrix by integrating the three characteristics of propagation time, propagation velocity and stress transfer efficiency.

5. The method for detecting micro-deformation of large steel structures according to claim 4, characterized in that: Calculating the microvibration conduction attenuation matrix includes: performing energy attenuation analysis on the microvibration signal between each pair of sensor nodes and calculating the energy attenuation rate; selecting clear microvibration events, and calculating the energy spectrum density function of the microvibration signal at two points for sensor i and sensor j respectively; performing frequency band integration on the energy spectrum density function to obtain the total energy value; calculating the energy attenuation ratio, which represents the proportion of microvibration energy transfer from node i to node j; introducing distance normalization processing to calculate the energy attenuation coefficient per unit distance; constructing the initial microvibration conduction attenuation matrix based on the attenuation coefficient; optimizing the initial attenuation matrix by using the spatial smoothing filtering method; applying structural symmetry constraints, and forcing the attenuation characteristics of parts with symmetrical structures to satisfy the corresponding symmetry.

6. The method for detecting micro-deformation of large steel structures according to claim 5, characterized in that: The frequency domain analysis of the micro-vibration signal set includes: converting the time domain signal into the frequency domain using the fast Fourier transform method; windowing the signal using the Hanning window; smoothing the converted spectrum; identifying the frequency domain characteristics of each measuring point based on the smoothed spectrum; extracting the resonant frequency and the corresponding amplitude for each measuring point; analyzing the frequency domain relationship between the measuring points using the frequency response function method and calculating the transfer function; constructing a frequency correlation matrix based on the transfer function; performing eigenvalue decomposition on the frequency correlation matrix and extracting the main modal features; and constructing a micro-vibration conduction resonance matrix based on the modal analysis results and the frequency correlation matrix.

7. The method for detecting micro-deformation of large steel structures according to claim 6, characterized in that: Decomposing the micro-vibration conduction resonance matrix into a resonance peak matrix and a resonance valley matrix includes: setting a frequency response threshold; performing peak search and valley search on the frequency response curve of each measuring point; performing spatial clustering analysis on the original peak and valley features, classifying peak points with similar frequencies and adjacent spatial positions as the same resonance mode, and classifying valley points with similar frequencies and adjacent spatial positions as the same suppression mode; constructing a resonance peak matrix and a resonance valley matrix based on the clustering results; performing feature extraction on the resonance peak matrix and the resonance valley matrix in the frequency dimension, and calculating the weight coefficient of each frequency point; and using a weighted summation method to reduce the three-dimensional resonance peak matrix and the resonance valley matrix to a two-dimensional matrix.

8. The method for detecting micro-deformation of large steel structures according to claim 7, characterized in that: Constructing the micro-deformation fusion feature matrix of the steel structure includes: determining the weight of each feature matrix in the fusion process; normalizing each feature matrix to ensure that the dimensions and numerical ranges of different matrices are consistent; weighted fusion of the normalized feature matrix, using a linear weighted method to calculate the fusion eigenvalues ​​to form a preliminary micro-deformation fusion feature matrix; applying spatial filtering to the preliminary fusion feature matrix using a Gaussian filter; applying nonlinear enhancement processing to the filtered fusion matrix using a power function enhancement method.

9. The method for detecting micro-deformation of large steel structures according to claim 8, characterized in that: The calculation of the micro-deformation matrix includes: converting the micro-deformation fusion feature matrix into a format suitable for model input; standardizing the compressed feature data; inputting the processed feature data into a pre-trained micro-vibration-micro-deformation model; reorganizing the model output results into a micro-deformation matrix; introducing physical constraints for post-processing, applying structural deformation continuity constraints, applying the principle of minimum structural deformation energy, and applying support constraints; after physical constraint optimization, the final micro-deformation matrix is ​​obtained.

10. The method for detecting micro-deformation of large steel structures according to claim 9, characterized in that: Calculating the micro-deformation key matrix includes: extracting deformation data of key nodes from the micro-deformation matrix; evaluating the importance of deformation data of key nodes, and the evaluation indicators include absolute value of deformation variable, relative value of deformation variable, consistency of deformation direction with structural force direction, etc.; calculating the comprehensive risk coefficient based on multiple evaluation indicators; constructing the micro-deformation key matrix according to the comprehensive risk coefficient; performing cluster analysis on the micro-deformation key matrix, and dividing key nodes with similar risk coefficients into multiple levels corresponding to different risk levels; determining the micro-deformation over-limit area includes: establishing a micro-deformation safety threshold system; comparing the deformation data in the micro-deformation key matrix with the safety threshold to identify the over-limit area; performing spatial cluster analysis on the over-limit area; calculating the deformation characteristic parameters for each over-limit area; evaluating the severity of the over-limit based on the deformation characteristic parameters; generating a micro-deformation over-limit area distribution map and a micro-deformation over-limit development trend map.

Images