A machine learning based method and system for monitoring the health of a building structure

Abstract

The application relates to a kind of building structure health monitoring method and system based on machine learning, specifically as follows: first, the three-dimensional numerical model of target building is constructed by finite element simulation simulation, generates simulation signal and injects Gaussian white noise, adjusts model parameters to constitute data set with health category label;Then, adaptive wavelet denoising is used in combination with dynamic normalization for data enhancement, and then adaptive window short-time Fourier transform and adaptive band enhancement are used to extract and strengthen high-resolution time-frequency features;Subsequently, a neural network model is constructed, which fuses the physical prior guidance of structure and the multiscale space-time interaction, modulates, fuses and encodes the feature matrix to obtain a refined feature vector, and realizes the output of damage state probability distribution;And the model training is constrained by the regularization term of feature consistency and prediction smoothness, and finally the trained model is deployed to realize building structure health state evaluation and safety warning, improve monitoring accuracy and reliability, and provide effective technical support for building safety guarantee.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building health intelligent monitoring, and in particular to a building structure health monitoring method and system based on machine learning. BACKGROUND

[0002] With the increasingly large and complex of modern building structures, the safety and durability during long-term service have become a major issue of social concern. Structure health monitoring technology aims to realize early identification and safety warning of damage by real-time sensing and analyzing of structural response data, which is a key means to ensure the safety of major infrastructure throughout its life cycle. In recent years, data-driven methods represented by deep learning have brought new light to structure health monitoring. Such methods can automatically learn the complex mapping relationship between damage and response from massive monitoring data, and exhibit strong feature abstraction ability.

[0003] However, the prior art has the following disadvantages: the data preprocessing method is rigid, and the conventional minimum-maximum normalization or fixed threshold wavelet denoising cannot adaptively handle the contradiction between simulation noise and local feature preservation, which easily leads to feature distortion; the time-frequency feature extraction is limited by fixed window function or fixed scale transformation, and the time-frequency resolution is insufficient, and there is a lack of focusing mechanism for key damage frequency bands, and the feature discrimination is low; the mainstream neural network model is mostly a general architecture, which is difficult to integrate the physical prior knowledge in the field of structure health monitoring, and the modeling ability of multi-scale time evolution characteristics and cross-channel coupling relationship of damage response is limited; the model training is usually only oriented to classification accuracy, and lacks regularization constraints to guide the features to comply with the laws of structural mechanics, and the generalization ability and prediction smoothness are insufficient in the scenes of class imbalance and continuous change of damage state.

[0004] Therefore, the present application proposes a building structure health monitoring method and system based on machine learning to solve the above problems. SUMMARY

[0005] The present application is aimed at the deficiencies of the prior art, and develops a building structure health monitoring method and system based on machine learning. The present application can improve the monitoring accuracy and reliability, and provides effective technical support for building safety protection.

[0006] The technical scheme for solving the technical problem of the present application is a building structure health monitoring method based on machine learning, comprising the following steps:

[0007] S1, a three-dimensional numerical model of the target building is constructed by using finite element simulation technology, the dynamic response under different load conditions is simulated through the three-dimensional numerical model, the simulation signals corresponding to the real acquisition situation are generated, and the building health monitoring data is constructed;

[0008] S2, inject Gaussian white noise into the simulation signal to simulate simulation noise, and adjust the three-dimensional numerical model parameters to generate different types of simulation data sets, and label the health category for each sample to form a data set;

[0009] S3, data in the data set is preprocessed by using adaptive wavelet denoising combined with dynamic normalization in two stages, and enhanced data is generated;

[0010] S4, a two-stage method of adaptive window short-time Fourier transform combined with adaptive frequency band enhancement is used to extract high-resolution time-frequency features from the enhanced data, and the feature components related to structural damage are strengthened to generate an enhanced time-frequency feature matrix;

[0011] S5, by fusing the neural network model of structural physical prior guidance and multi-scale space-time interaction, a building structure health classification model is constructed, the enhanced time-frequency feature matrix is subjected to structural physical prior guided feature modulation, then the modulation result is subjected to multi-source feature fusion and external factor injection to generate a mixed feature vector, then multi-scale space-time interaction coding is performed to obtain a refined feature vector, and then the damage mode classification outputs the probability distribution of the damage state;

[0012] S6, based on the regularization term constraint of feature consistency and prediction smoothness, the building structure health classification model learning process is calculated, and the total loss function of the model is calculated, and then the model is trained based on the loss function;

[0013] S7, the trained building structure health classification model is deployed to the actual building structure health monitoring, and the building structure health state is evaluated and safety warning is given.

[0014] S1 is as follows:

[0015] A three-dimensional numerical model is constructed according to the actual geometric size, material properties, component connection mode and boundary conditions of the target building structure;

[0016] Different load cases include environmental excitation, simulated wind load and simulated traffic load;

[0017] The simulation signal is a vibration acceleration time history signal corresponding to the real acquisition situation, and the real acquisition situation is to deploy sensors on the target building structure and collect real vibration acceleration time history signals through the sensors.

[0018] S2 is as follows:

[0019] The method of adjusting the parameters of the three-dimensional numerical model includes locally reducing the material stiffness, introducing micro-crack elements and changing the properties of the connecting parts, and by adjusting the parameters, the structural damage state at different positions and different severity is simulated to generate simulation data sets of different health categories, including health, slight damage, severe damage and serious damage.

[0020] The simulation signal is taken as a sample in a fixed length data point, and each sample is labeled with a health category label according to the preset damage state during simulation.

[0021] S3 is specifically as follows:

[0022] S3.1, the building structure health monitoring data in the data set is subjected to stationary wavelet transform, and then adaptive threshold weights of each data point on each scale are dynamically calculated according to the difference between the local characteristics of the signal and the statistical characteristics of each scale, and the denoising monitoring data is obtained after weighting and reconstruction based on the adaptive threshold weights;

[0023] S3.2, the denoising monitoring data is subjected to dynamic range normalization processing to obtain enhanced data;

[0024] The scaling factor based on local statistical characteristics is used to dynamically adjust the global normalization result to obtain normalized monitoring data.

[0025] S4 is specifically as follows:

[0026] S4.1, the enhanced data is mapped from the time domain to the time-frequency domain through adaptive window short-time Fourier transform, the adaptive window short-time Fourier transform uses a window function with adaptive bandwidth to segment the processing on the time axis, and generates a high-resolution initial time-frequency feature matrix;

[0027] S4.2, the global average energy of each frequency bin of the initial time-frequency feature matrix is calculated, and the global frequency band weight is generated according to the global average energy to strengthen the high-energy frequency band, and the local time-frequency attention weight is calculated to focus on the time-frequency point with energy mutation, and then the two weights are respectively applied to the initial time-frequency feature matrix, and the enhanced time-frequency feature matrix is obtained after addition.

[0028] S5 is specifically as follows:

[0029] S5.1, the enhanced time-frequency feature matrix of all health state samples in the data set is estimated to represent the baseline mode atlas of the typical distribution of time-frequency energy under the health state, and the enhanced time-frequency feature matrix is subjected to physical guided attention modulation and deviation enhancement combined with the modulation intensity coefficient, to generate a physical perception feature matrix fused with health baseline prior knowledge;

[0030] S5.2, the physical perception feature matrix is flattened into a vector, and a group of time domain statistical and envelope spectrum features are extracted from the denoising time domain signal to form a statistical feature vector, and an external environmental factor feature vector is introduced, and the three parts of information are adaptively weighted and fused through a feature fusion gate unit to obtain a mixed feature vector;

[0031] S5.3, adopt a multi-scale space-time interaction block, extract different time receptive field features through parallel multi-dilation rate space-time hole convolution, and interactively modulate the relationship between the features channels through adaptive feature channel interaction modeling, and output refined feature vectors after stacking multiple modules;

[0032] wherein the input of the first multi-scale space-time interaction block is the mixed feature vector, and the output of the previous multi-scale space-time interaction block is used as the input of the next multi-scale space-time interaction block;

[0033] S5.4, map the refined feature vector to the non-normalized score corresponding to different structural health states through a fully connected layer, and obtain the final classification probability distribution by using a Softmax function that fuses adaptive temperature scaling and class prior weight.

[0034] The calculation process of the total loss function is as follows:

[0035] (1) Use a loss term combining adaptive margin cross-entropy and class center constraint to simultaneously reduce the distance between sample features and their own class center and increase the distance between sample features and other class centers, and adaptively adjust the margin value according to the difficulty of class distinction;

[0036] (2) Based on the prior knowledge that damage evolution is continuous and should be consistent in time-frequency response, a regularization loss term containing time-frequency feature consistency constraint and prediction distribution smoothness constraint is used to encourage similar features of samples of the same class and smooth changes in prediction of continuous or adjacent states, which is expressed as:

[0037] (3) The total loss is the weighted sum of the adaptive margin contrast center loss and the regularization loss.

[0038] The training process of the building structure health classification model is as follows:

[0039] The data set is divided into a training data set and a validation data set, and the data in the training data set is processed through S3 and S4, and then the model is trained end-to-end;

[0040] In each training iteration, a batch of sample data is sequentially subjected to forward propagation of the model to obtain the predicted probability distribution, and then the loss value of the current batch is calculated according to the total loss function;

[0041] An adaptive matrix estimator optimizer is used to calculate the gradient of all trainable parameters in the model according to the total loss function, and these parameters are updated through a backpropagation algorithm to minimize the total loss function;

[0042] During the training process, an independent validation data set is used to periodically evaluate the performance of the model;

[0043] The training iteration of the model will continue until the performance indicator on the validation set no longer improves for a plurality of consecutive training cycles, or a preset maximum number of iteration cycles is reached, at which time the training stops, the model parameters with the best validation performance are saved, and the training of the building structure health classification model is completed.

[0044] S7 is specifically as follows:

[0045] After the building structure health classification model is trained, it is deployed in actual building structure health monitoring, and raw vibration signals are collected in real time or periodically by vibration sensors installed at key parts of the building structure to form vibration data to be monitored; at the same time, environmental temperature and humidity data are collected synchronously to form an external environmental factor feature vector;

[0046] Then, the collected data is processed and input into the trained building structure health classification model, the model outputs the probability of the building structure belonging to each health category in the monitoring period; the category corresponding to the highest probability is determined as the health status of the current structure, and the monitoring result, probability distribution and key features are visualized; and different levels of early warning are automatically triggered according to the health status of the current structure.

[0047] The application also provides a building structure health monitoring system based on machine learning, which executes a building structure health monitoring method based on machine learning, and the system structure is as follows:

[0048] A three-dimensional numerical model and simulation data set generation module is used to construct a three-dimensional numerical model of the target building by using finite element simulation technology, simulate dynamic responses under different load conditions to generate simulation signals, inject Gaussian white noise and adjust model parameters to generate a multi-category simulation data set, and label the sample health category;

[0049] A data preprocessing and enhancement module is used to process the simulation data set to generate enhanced data by using an adaptive wavelet denoising combined with a dynamic normalization two-stage method;

[0050] A time-frequency feature extraction and enhancement module is used to extract high-resolution time-frequency features from the enhanced data by using an adaptive window short-time Fourier transform combined with adaptive frequency band enhancement, and to strengthen damage-related feature components to generate an enhanced time-frequency feature matrix;

[0051] A health classification model construction and training module is used to construct a neural network model that integrates structure physical prior guidance and multi-scale space-time interaction, to perform feature modulation, multi-source fusion and external factor injection on the enhanced time-frequency feature matrix, to obtain refined feature vectors through multi-scale space-time interaction coding, to construct a total loss function based on feature consistency and prediction smoothness regularization terms, and to train the model;

[0052] Health monitoring and early warning module: deploy the trained model to evaluate the health state of the building structure, output the damage probability distribution and realize safety warning.

[0053] The effects provided in the summary are only the effects of the embodiments, not all the full effects of the invention, and the above technical solutions have the following advantages or beneficial effects:

[0054] The application discloses a building structure health monitoring system and method based on machine learning, adopts a two-stage preprocessing method of adaptive wavelet denoising and dynamic normalization, dynamically adjusts processing parameters according to local features of signals, and retains local fluctuation features related to damage while suppressing noise; adopts a two-stage time-frequency feature extraction strategy of adaptive window short-time Fourier transform and adaptive frequency band enhancement, generates high-resolution and high-discrimination time-frequency representation through a time-varying window function and a double weight mechanism; adopts a neural network model that fuses structure physical prior guidance and multi-scale space-time interaction, realizes fine perception of damage patterns through health benchmark modal atlas modulation, multi-source feature fusion and learnable space-time relationship modeling; and adopts a composite loss function combining adaptive marginal contrast center loss and physical law regularization, guides the model to learn feature representation with high intra-class consistency, inter-class separability and damage evolution continuity. BRIEF DESCRIPTION OF DRAWINGS

[0055] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the specification, illustrate the application together with the embodiments thereof, and explain the application without limiting the application.

[0056] Figure 1 The present application is a method flowchart.

[0057] Figure 2 The present application is a simulation modeling example using fine finite element simulation technology.

[0058] Figure 3 The present application is a vibration acceleration time history signal example.

[0059] Figure 4 The present application is a structure schematic diagram of step S4.

[0060] Figure 5 The present application is a classification accuracy comparison diagram of different classification methods under four damage states. DETAILED DESCRIPTION

[0061] In order to clearly illustrate the technical features of the present application, the present application will be described in detail below with reference to the specific embodiments and the accompanying drawings.

[0062] Embodiment 1

[0063] As Figure 1As shown, a machine learning-based building structure health monitoring method comprises the following steps:

[0064] S1, a three-dimensional numerical model of the target building is constructed by using finite element simulation technology, the dynamic response under different load conditions is simulated through the three-dimensional numerical model, the simulation signals corresponding to the real collection are generated, and the building health monitoring data is constructed;

[0065] S2, Gaussian white noise is injected into the simulation signal to simulate simulation noise, and the parameters of the three-dimensional numerical model are adjusted to generate different types of simulation data sets, and the health category is labeled for each sample to form a data set;

[0066] S3, the data in the data set is preprocessed by using the adaptive wavelet denoising combined with dynamic normalization two-stage preprocessing method for data enhancement, and the enhanced data is generated;

[0067] S4, a two-stage method of adaptive window short-time Fourier transform combined with adaptive frequency band enhancement is used to extract high-resolution time-frequency features from the enhanced data, and the feature components related to structural damage are strengthened to generate an enhanced time-frequency feature matrix;

[0068] S5, a neural network model is constructed by fusing the structure physical prior guidance and the multi-scale space-time interaction, a building structure health classification model is constructed, the enhanced time-frequency feature matrix is subjected to feature modulation guided by the structure physical prior, and then the modulation result is subjected to multi-source feature fusion and external factor injection to generate a mixed feature vector, then subjected to multi-scale space-time interaction coding to obtain a refined feature vector, and then subjected to damage mode classification to output the probability distribution of the damage state;

[0069] S6, the regularization term based on feature consistency and prediction smoothness is used to constrain the building structure health classification model learning process, the total loss function of the model is calculated, and then the model is trained based on the loss function;

[0070] S7, the trained building structure health classification model is deployed to the actual building structure health monitoring to evaluate and safety warning of the building structure health state.

[0071] In the specific implementation, S1 is as follows:

[0072] Due to the implementation of the system and method of the present application, a large amount of vibration monitoring data with clear health state label is relied on, in order to obtain such data, a refined finite element simulation technology is used to construct an accurate three-dimensional numerical model of the target building structure in the computer, and the model is strictly established according to the actual structure of the geometric size, material properties, component connection mode and boundary conditions; wherein, the material properties include the elastic modulus, density, Poisson's ratio and the like of the concrete;

[0073] As Figure 2As shown, it is an example diagram of simulation modeling by using fine finite element simulation simulation technology.

[0074] By simulating the dynamic response of the structure under various load conditions, vibration acceleration time history signals corresponding to real sensor acquisition are generated, i.e. building structure health monitoring data.

[0075] As shown in Figure 3 , it is an example diagram of the vibration acceleration time history signal of the sample.

[0076] In the specific implementation, S2 is specifically as follows:

[0077] In order to be close to the actual monitoring environment, Gaussian white noise conforming to specific statistical characteristics is actively injected in the simulation signal to simulate simulation noise. By systematically changing the key parameters in the finite element model to simulate the structure damage state at different positions and different severity, simulation data sets covering health, slight damage, moderate damage and severe damage are generated.

[0078] The length of the vibration signal of each sample is fixed as 8192 data points, and is directly labeled according to the preset damage condition during simulation. The label category strictly corresponds to the above four structure health states.

[0079] Based on this, large-scale training data sets and verification data sets with accurate labels are generated, providing a basis for subsequent model construction and training.

[0080] In the specific implementation, S3 is specifically as follows:

[0081] The building structure health monitoring data is obtained by fine finite element simulation simulation, has the characteristics of high dimension, contains simulation noise and large amplitude scale difference, and the conventional minimum-maximum normalization or fixed threshold wavelet denoising preprocessing method usually cannot adaptively remove the simulation noise and ignore the local fluctuation characteristics, which is easy to cause distortion of subsequent feature extraction.

[0082] The present application adopts an adaptive wavelet denoising combined with a dynamic normalization two-stage preprocessing procedure to enhance data robustness and retain key local features, and the specific steps are as follows:

[0083] S3.1, the building structure health monitoring data is subjected to stationary wavelet transform, then according to the difference between the signal local feature and the statistical characteristics of each scale, the adaptive threshold weight of each data point on each scale is dynamically calculated, and the denoising monitoring data is obtained after weighted reconstruction based on the adaptive threshold weight, expressed as:

[0084]

[0085] In the formula, indicates the building structure health monitoring data, and the dimension is , is the vibration signal obtained by fine finite element simulation, containing simulation noise; denotes the time point index, identifying the sampling position of the vibration signal in the time domain, with a value range of ; denotes the detail coefficient value of the time point on the scale , obtained by the stationary wavelet transform algorithm, specifically using the Daubechies wavelet basis, obtaining the detail coefficient matrix of each scale through the stationary wavelet transform algorithm, with a dimension of ; denotes the approximation coefficient value of the time point on the scale , obtained by the stationary wavelet transform algorithm, with a dimension of , using the coarsest scale to obtain the low-frequency profile of the signal, representing the global trend and low-frequency components of the signal; denotes the vibration amplitude of the building structure health monitoring data at the th time point, representing the sampling value of the signal in the time domain; denotes the vibration amplitude of the noise reduction monitoring data at the th time point, which is the clean signal obtained through adaptive wavelet denoising processing; denotes the total number of wavelet decomposition scales, used to control the depth of multi-scale analysis, its value is determined by the length of the vibration signal, calculated by as , where is the floor operation; denotes the scale index, which is a positive integer, with a value range of ; denotes the adaptive threshold weight of the th scale and the th data point, used to dynamically adjust the contribution of each scale detail coefficient to the reconstruction result according to the local signal characteristics, allowing important fluctuation characteristics to be preserved while denoising, with a calculation method represented as ; denotes the sensitivity coefficient of the th scale, used to control the sensitivity of the scale weight to signal differences, thereby adjusting the denoising intensity of the scale, according to the scale noise level setting, for example, defining , so that with smaller scales (more high-frequency noise) is larger, with smaller denoising weights, thereby enhancing the denoising effect; denotes the absolute value mean of the th scale detail coefficient, representing the central tendency of the signal of that scale, with a calculation method represented as ; This represents the natural exponential function.

[0086] It should be noted that, through stationary wavelet transform, building structural health monitoring data can be... Decomposing it into approximation coefficients and detail coefficients allows us to obtain the scale. Up to the time point Detail coefficient value and scale Up to the time point Approximate coefficient values .

[0087] S3.2 To address the issue of large amplitude scale differences, the normalization interval for the noise reduction monitoring data is not fixed at [0,1]. Instead, a scaling factor based on local statistical characteristics is used to dynamically adjust the global normalization result to enhance the retention of local fluctuations, as expressed below:

[0088]

[0089] In the formula, Indicates normalized monitoring data In the The vibration amplitude at each time point, with dimension [missing information]. It represents the vibration signal after noise removal and normalization, where the global trend is standardized and local anomalies are enhanced; This represents the function that takes the minimum value. The item is used to calculate noise reduction monitoring data. The minimum value of all elements in the range is used to determine the lower limit of the global magnitude; This represents the function that takes the maximum value. The item is used to calculate noise reduction monitoring data. The maximum value of all elements in the range is used to determine the global amplitude limit; Indicates the first The scaling factor at each time point is used to dynamically adjust the normalization amplitude based on local data fluctuations, in order to preserve local features while addressing large differences in amplitude scale. The calculation method is expressed as follows: In the actual implementation, symmetrical expansion filling is used at the boundary; This represents the local sensitivity parameter, used to control the degree to which local fluctuations affect the scaling factor, thereby adjusting the intensity of local enhancement. An example value is 0.05. This represents the local window size, defining the neighborhood range for calculating local statistics. Examples of possible values ​​are provided. ; This represents the position index within a local window, with a value range of 1. ; This indicates that the noise reduction monitoring data is in the first... The vibration amplitude at each time point.

[0090] It should be noted that, Item calculation with the first After a certain point in time The local mean calculated from each data point represents the average behavior of the signal within a local window. It is used to quantify the deviation between the current data point and the local average behavior, thereby dynamically adjusting the scaling factor during the normalization process to enhance the preservation of local fluctuation characteristics.

[0091] In specific implementation methods, such as Figure 4 As shown, S4 is as follows:

[0092] Vibration data contains frequency and time domain features of structural health status. Conventional short-time Fourier transform or fixed-scale wavelet transform has inherent limitations in terms of insufficient time and frequency resolution and cannot adaptively focus on key frequency bands related to structural damage, resulting in low feature discrimination and affecting the performance of subsequent classification models.

[0093] This invention employs a two-stage method combining adaptive window short-time Fourier transform and adaptive frequency band enhancement to extract high-resolution time-frequency features from preprocessed vibration data and enhance feature components related to structural damage. The specific steps are as follows:

[0094] S4.1. The normalized monitoring data is mapped from the time domain to the time-frequency domain using the adaptive window short-time Fourier transform. The adaptive window short-time Fourier transform uses a window function with adaptive bandwidth to perform piecewise processing on the time axis to generate a high-resolution initial time-frequency feature matrix, expressed as:

[0095]

[0096] In the formula, Represents the initial time-frequency characteristic matrix The Middle The first time frame, the first The value of each frequency bin, with dimension . It characterizes the energy distribution of the signal in the time-frequency domain, reflecting the frequency band and time-varying modes of damage-related vibrations in structures; Represents the time frame index, with a value range of 100. ; The total number of time frames is represented by the window length. and step length Calculation, expressed as For example, if , ,but ; This represents the frequency bin index, with a value range of [value range missing]. ; represents the total number of frequency bins, used to control the frequency resolution of the time-frequency representation, with an example value of 128; represents the th time frame, the th frequency bin, where the center position and bandwidth vary with time and frequency to achieve high-resolution time-frequency analysis, represented as ; represents the imaginary unit, satisfying ; represents the th frequency bin, used to identify the frequency domain position, ranging between , calculated as ; represents the th time frame, used to position the center of the window function, determined by the time segmentation parameter; represents the th frequency bin, used to control the bandwidth of the window function, with higher frequency having narrower bandwidth to adapt to the characteristics of the signal changing rapidly in the high-frequency region, calculated as ; represents a constant to prevent the denominator from being zero, which is a very small constant, with an example value of .

[0097] In the specific implementation process, is determined by the time segmentation method. Assuming that the time sequence is divided into segments, each with a window length of , then can be defined as the time index of the center point of the th segment. For example, if the segments are non-overlapping and uniform, then is defined, where the window length , if the segments have overlap, the adjustment needs to be made according to the overlap rate, such as the window length , and the overlap rate , then .

[0098] It should be noted that is the Fourier kernel function, used to convert the windowed time-domain signal to the frequency domain. In order to capture both time and frequency information in time-frequency analysis, it allows the analysis of the frequency domain characteristics of the signal within each time frame, and further generates a high-resolution time-frequency representation.

[0099] S4.2 Calculate the global average energy of each frequency compartment in the initial time-frequency feature matrix, and generate global frequency band weights to enhance high-energy frequency bands. Simultaneously, calculate local time-frequency attention weights to focus on time-frequency points with energy abrupt changes. Then, apply both weights to the initial time-frequency feature matrix and sum them to obtain the enhanced time-frequency feature matrix, expressed as:

[0100]

[0101] In the formula, Represents the enhanced time-frequency feature matrix The Middle The first time frame, the first The value of each frequency bin, with dimension [missing information]. This characterizes the adaptively enhanced time-frequency energy distribution; Indicates the first The global frequency band weight of each frequency compartment is used to adaptively enhance key frequency bands that may be related to structural damage based on the average energy of that frequency over the entire time period. The calculation method is expressed as follows: ; Indicates the first The first time frame, the first The local time-frequency attention weights for each frequency bin are used to highlight higher-energy frequency components within the same time frame, improving the local discriminative power of features. The calculation method is expressed as follows: . Indicates the first The global average energy of each frequency compartment is derived from the initial time-frequency characteristic matrix. In the The average of the absolute values ​​of all time frames across a frequency cell is used to represent the average intensity of that frequency component over the entire time range. The calculation method is expressed as follows: ; The slope parameter of the Sigmoid function is used to control the steepness of the weight changes, thereby adjusting the smoothness of the weight transition. The example value is 5. This represents the energy threshold parameter, used to determine whether a frequency band is significant, thereby filtering important frequency bands. An example value is 0.1. This represents the attention temperature parameter, which controls the concentration of the weight distribution, thereby adjusting the intensity of attention. An example value is 0.5.

[0102] In a specific implementation, S5 is as follows:

[0103] The enhanced time-frequency feature matrix contains key time-frequency information related to structural damage. The damage patterns of building structures are closely related to their physical properties, and damage evolution exhibits spatiotemporal correlation. Conventional convolutional neural networks or recurrent neural networks struggle to incorporate prior knowledge of structural physics and have limited ability to model cross-scale spatiotemporal feature interactions. This invention constructs a neural network model that integrates prior knowledge of structural physics with multi-scale spatiotemporal interactions to achieve high-precision and robust classification of the health status of building structures. The specific steps are as follows:

[0104] S5.1. Using the enhanced time-frequency feature matrix of all healthy state samples in the training set, estimate the baseline mode map representing the typical distribution of time-frequency energy under healthy states. Combine this with the modulation intensity coefficient to perform physically guided attention modulation and bias enhancement on the enhanced time-frequency feature matrix, generating a physical perception feature matrix that integrates prior knowledge of the healthy baseline, expressed as:

[0105]

[0106] In the formula, Represents the physical perception feature matrix The Middle The first time frame, the first The value of each frequency bin, with dimension [missing information]. It integrates original time-frequency energy with prior physical knowledge to enhance the damage-related frequency band while suppressing the health baseline signal; The modulation intensity coefficient is a trainable scalar parameter used to control the strength of the physical prior's guidance on the attention weights. Indicates the first The first time frame, the first The modulation coefficients of each frequency cell are used to quantify the deviation of the current feature point from the health baseline, and are calculated as follows: ; The bias amplification factor is a trainable scalar parameter used to amplify significant deviations from a healthy baseline. This represents the hyperbolic tangent activation function, which maps the input to the interval [0, 1]. ; In the reference modal spectrum, the first The first time frame, the first The values ​​of each frequency bin are obtained by representing the initial time-frequency feature matrix of all healthy samples in the training set. The average along the upper sample dimension is used to characterize the typical distribution of time-frequency energy in a healthy state; The reference modal spectrum is shown in the first... The standard deviation of the frequency dimension of each time frame is used to normalize the bias, and is calculated as follows: ; express The mean; represents a vector consisting of values of the reference modal map on the i-th frequency bin and all time frames (from t 1 to t 2 ) ; represents a vector consisting of values of the reference modal map on the i-th time frame and all frequency bins (from f 1 to f 2 ) ; represents a vector consisting of values of the reference modal map on the i-th time frame and all frequency bins (from f 1 to f 2 ) ; represents a variance operation; represents a minimum constant to prevent the denominator from being zero, with a value example of .

[0107] S5.2, for comprehensive characterization of the structural state, the physical perception feature matrix is flattened into a vector, and a set of time domain statistics and envelope spectrum features are extracted from the denoised time domain signal to form a statistical feature vector, and an external environment factor feature vector is introduced, and the three parts of information are adaptively weighted and fused through a feature fusion gate unit to obtain a mixed feature vector, represented as:

[0108] In the formula, represents a mixed feature vector, with a dimension of , adaptively fusing time-frequency, time-domain statistics and external environment multi-source information; represents a fusion weight matrix, with a dimension of , which is a trainable parameter matrix, and information compression and fusion are realized by projection to represents the dimension of the mixed feature vector; represents the dimension of the statistical feature vector; represents the dimension of the external environment factor feature vector; represents a vector splicing operation; represents a matrix flattening operation, flattens into a vector with a length of ; represents a statistical feature vector, with a dimension of , composed of time domain indicators extracted from denoised monitoring data and envelope spectrum features based on normalized monitoring data , wherein the time domain indicators include mean, standard deviation, skewness, kurtosis, and root mean square; represents an external environment factor feature vector, with a dimension of , with a value example of , containing, for example, environmental temperature and environmental humidity information, which is obtained by simulating the way of collecting environmental factors by sensors; ​​​​​​​represents an element-wise multiplication operation; represents a Sigmoid activation function, whose output value range is , which is used to generate adaptive gating weights; represents a gating weight matrix, whose dimension is , which is a trainable parameter matrix, used to construct a trainable gating mechanism; represents a gating bias vector, whose dimension is , which is a trainable parameter vector.

[0109] In the specific implementation process, the external environmental factor feature vector includes two element attributes of environmental temperature and environmental humidity . Temperature change will cause the change of the elastic modulus of the material, resulting in the change of the overall stiffness of the structure, and then affecting its natural frequency. Humidity may affect the mass of the concrete structure (water absorption) or the stiffness of the wood structure, thereby changing its dynamic response.

[0110] The time domain indicators extracted from the denoised monitoring data include mean (the average amplitude of the signal, reflecting the static offset or direct current component of the vibration), standard deviation (the intensity of the signal fluctuation around the mean, reflecting the vibration energy), skewness (the asymmetry of the signal amplitude distribution, positive skewness indicating more large amplitude pulses, which may be caused by impact damage), kurtosis (the sharpness of the signal amplitude distribution, high kurtosis indicating the existence of significant impact components in the signal, which is strongly related to bearing failure and crack development), and root mean square (the effective amplitude of the signal). The envelope spectrum feature is obtained by performing Hilbert transform on the normalized monitoring data to obtain the envelope signal, and then performing frequency spectrum analysis on it. The envelope spectrum can highlight the periodic impact characteristics of the signal (such as bearing fault characteristic frequency, gear meshing frequency sideband, etc.).

[0111] It should be noted that the physical perception feature matrix mainly captures the frequency and time-varying patterns of the signal, but may lose or weaken some intuitive time domain statistical characteristics, while the statistical feature vector extracts classical statistical features and envelope spectrum features from the denoised monitoring data and the normalized monitoring data , which are effective supplements to the time-frequency features, providing different perspectives and more physically interpretable information for the model, and improving the robustness of the model.

[0112] It should also be noted that the dynamic characteristics (such as frequency and damping) of the building structure are significantly affected by environmental factors (temperature and humidity), and ignoring these factors may cause the model to misjudge the normal performance fluctuations caused by environmental changes as damage. The external environmental factor feature vector The aim is to enable the model to perceive and "exclude" environmental disturbances, focusing on changes caused by real damage, thus improving the reliability of the monitoring system.

[0113] S5.3, to capture the complex multi-scale dependence between time and feature channels that may exist for damage features, a multi-scale space-time interaction block is used to extract different time receptive field features through parallel multi-dilation rate space-time hole convolution, and interactively modulate the inter-channel relationship modeled by adaptive feature channels, and output refined feature vectors after stacking multiple modules, wherein the interaction process of a single module is represented as:

[0114]

[0115] In the formula, represents the feature vector input to the th multi-scale space-time interaction block, for the first multi-scale space-time interaction block, define ; represents the feature vector input to the th multi-scale space-time interaction block, that is, the feature vector output by the th multi-scale space-time interaction block, which deeply fuses multi-scale time information and inter-channel correlation information, to obtain high-level abstract features that can better reflect the nature of damage space-time evolution; represents the layer normalization operation; represents the dilation rate of the hole convolution, which belongs to a predefined dilation rate set , for example , used to control the size of the receptive field of the convolution kernel in the time dimension; represents the rectified linear unit activation function, used to introduce nonlinearity, so that the model can learn and represent more complex function relationships; represents the time convolution kernel parameter corresponding to the dilation rate , with a dimension of , which is a trainable parameter; represents the convolution kernel size, with a value example of ; represents a one-dimensional hole convolution operation with a dilation rate of , allowing the network to efficiently capture long-range temporal dependencies; represents the element-wise multiplication operation; represents the activation function used for spatial attention weights, specifically using the Softplus function; represents the graph adjacency matrix, with a dimension of , used to model the spatial relationship between feature channels, which is a trainable parameter; It represents the interaction operation of feature channels. It models the dependencies between feature channels through a learnable adjacency matrix, so that the feature integrates information from other channels on each channel, thereby simulating the propagation and coupling effect of damage among multiple physical quantities.

[0116] In the specific implementation process, the definition The item is simplified as The number of time steps is For dimension tensor First, it is reshaped into dimensions. A two-dimensional matrix, then with a trainable graph adjacency matrix. Performing matrix multiplication yields the dimension. The result is that it is eventually reshaped back to the original dimension. , represented as ,in, Reshaping into dimensions is tensor, Reshaping into dimensions is The matrix.

[0117] Stacking After multiple multi-scale spatiotemporal interaction blocks, a refined feature vector is obtained. ,in This represents the total number of layers in the multi-scale spatiotemporal interaction block stack; an example value is 4.

[0118] It should be noted that conventional temporal networks either handle time dependencies or feature channel relationships separately. This invention modulates multi-scale temporal convolutions with adaptive feature channel interactions, using dilated convolutions with different dilation rates in parallel to capture different scale patterns of the damage response that may exist in time (such as instantaneous impact and slowly varying drift), and utilizes a learnable graph adjacency matrix. This study models the correlation strength between different feature channels (potentially representing different frequency bands or physical quantities), simulates the propagation and coupling effects of damage among multiple physical quantities, and utilizes the "spatial relationship" information after feature channel interaction processing to weight multi-scale temporal features. This guides the extraction process of temporal features with the semantic relationships between features, simulating the cognitive process of "analyzing time-varying patterns with emphasis based on the correlation between features." (Graph adjacency matrix) As a trainable matrix, its first... Line number Column elements It can be understood as the first The first feature channel and the first The model automatically learns the correlation strength between feature channels through training data. For example, it may learn that there is a strong correlation between "energy anomaly in a certain high-frequency band" and "phase change in a certain low-frequency band", which may correspond to a certain damage pattern, thereby strengthening this cross-channel collaborative perception during feature extraction.

[0119] S5.4. The refined feature vector is mapped to unnormalized scores corresponding to different structural health states through a fully connected layer, and the final classification probability distribution is obtained by using a Softmax function that integrates adaptive temperature scaling and class prior weights, as follows:

[0120]

[0121] In the formula, This indicates that the model predicts the sample belongs to the first... The probability of a damage state, with a range of The sum of the probabilities of all categories is ; This represents the category index, which is a positive integer with a value range of 1 to 2. ; Indicates the corresponding to the first The class output weight vector has dimensions of , are trainable parameters used to linearly map refined feature vectors to scalar scores; Indicates the corresponding to the first The output bias scalar of the class is a trainable parameter; Indicates the first The adaptive temperature parameter of a class is a trainable positive scalar parameter used to adjust the smoothness of the output probability distribution of that class. Indicates the first Prior weights of classes are used to mitigate class imbalance problems, and are calculated as follows: ; This represents a smoothing factor used to smooth prior weights. The calculation, when hour, The penalty is directly proportional to the inverse of the category frequency, and it dominates when... At that time, all No category prior, take It can strike a balance between mitigating class imbalance and preventing excessive distortion of the prediction distribution, with an example value of 0.8; The total number of categories representing the structural health status, specifically including: healthy, minor injury, moderate injury, and severe injury, for a total of four categories; This represents the total number of samples in the training set. Indicates the first training set The number of class samples.

[0122] In the detailed implementation, S6 is specifically as follows:

[0123] S6.1, the total loss function is calculated as follows:

[0124] To effectively train the neural network model and guide it to learn a robust feature representation conforming to the structural mechanics principle, in the loss function calculation, a regularization term based on feature consistency and prediction smoothness is adopted to constrain the model learning process, to realize the consistency of feature representation and the smoothness of prediction while encouraging accurate classification, and the specific steps are as follows:

[0125] (1) To enhance the intra-class compactness and inter-class discriminability of the learned features of the model, a loss term combining adaptive margin cross-entropy and class center constraint is adopted, which simultaneously reduces the distance between the sample features and the class center to which they belong and increases the distance between the sample features and other class centers, and the margin value is adaptively adjusted according to the difficulty of class distinction, and is expressed as:

[0126] ,

[0127] In the formula, represents the adaptive margin contrast center loss, which combines the angular interval classification loss and the center loss, and introduces a learnable adaptive margin; represents the sample index, which is a positive integer; represents the size of the training batch; represents the logarithmic function, and the default base is the natural constant; represents the index of the true class of the th sample; represents the natural logarithmic function; represents the natural exponential function; represents the cosine function; represents the feature scaling factor, which is a fixed scalar parameter for amplifying the difference in the cosine value of the angle, and an example of the value is ; represents the angle between the refined feature vector of the th sample and the class center vector corresponding to the class ; represents the adaptive margin of the class , which is a trainable non-negative scalar parameter, as a trainable parameter, it means that the model can automatically learn a larger margin for difficult-to-distinguish classes and a smaller margin for easy-to-distinguish classes, and the optimization process is more refined; represents the index of the non-real class, which is used to traverse all classes except the class ; represents the refined feature vector of the With category Corresponding class center vector The angle between them; This represents the weighting coefficient of the center loss term, used to balance the cross-entropy loss and the center loss. An example of its value is shown below. ; Indicates the first Refined feature vectors of each sample; Indicate category The class center vector is a trainable parameter vector with dimension equal to the first... Refined feature vectors of each sample same; This represents the L2 norm.

[0128] It should be noted that, through The requirement is that not only is the cosine value of the true category greater than that of other categories, but it is also required to be significantly larger by a margin. This forces clearer boundaries between classes, and, through Weighted central loss term This brings the characteristics of similar samples closer together, making the intra-class distribution more compact.

[0129] (2) Based on the prior knowledge of the continuity of damage evolution and its consistency in time-frequency response, a regularization loss term containing time-frequency feature consistency constraints and prediction distribution smoothness constraints is adopted to encourage similar features of samples of the same type and smooth changes in prediction of continuous or neighboring states, expressed as:

[0130]

[0131] In the formula, This represents the regularization loss, which serves as the overall objective function to be minimized during model training, taking into account the requirements of classification accuracy and the physical rationality and smoothness of model predictions / features. The weight coefficients for feature consistency regularization are shown in the example below. This is used to control the relative weight of the feature consistency regularization term in the regularization loss; The weight coefficients represent the prediction smoothness regularization, with examples of possible values. This is used to control the relative weight of the prediction smoothness regularization term in the regularization loss; This represents the set of all true class labels in a batch of training. For example, if a batch has two classes, "healthy" and "minor injury", then... ; Represents a set The number of unique categories appearing in the data; Indicates the category label index; Indicates that the current batch belongs to the category The number of samples; represents another sample belonging to the same class as the represents the total number of steps / total number of points in a continuous time sequence or a working condition gradual change sequence; represents the time step or sequence point index in the sequence; represents the class probability distribution vector predicted by the model at the represents the Kullback-Leibler divergence used to measure the difference between two probability distributions; represents the Frobenius norm.

[0132] (3) The total loss of model training is the weighted sum of adaptive marginal contrastive center loss and regularization loss, and the model learns robust features with high discriminability and in line with physical laws by optimizing the total loss, denoted as:

[0133]

[0134] In the formula, represents the total loss function of model training, which is the overall optimization objective of model training; represents the weight coefficient of regularization loss, which is used to control the importance of the regularization term in the total loss, and an example of the value is 0.1.

[0135] S6.2, the model training is as follows:

[0136] After the construction of the building structure health classification model is completed, the training data set generated by the fine finite element simulation in step S1 and preprocessed and feature extracted in steps S2 and S3 is used to train the model end to end.

[0137] ​​​​​​​​​​​The training process starts with the calculation of the baseline modal atlas using the enhanced time-frequency feature matrix of all healthy state samples in the training set, which is solidified as physical prior knowledge and used for feature modulation in the S401 step of all subsequent training batches. In each training iteration, a batch of sample data is sequentially subjected to forward propagation of the model to obtain the predicted probability distribution, and then the loss value of the current batch is calculated according to the total loss function defined in the S402 step, which includes an adaptive marginal contrast center loss and a regularization term.

[0138] An adaptive matrix estimator optimizer is used to calculate the gradient of all trainable parameters in the model according to the loss function, and update these parameters through the backpropagation algorithm to minimize the total loss function.

[0139] During the training process, an independent validation set is used to periodically evaluate the performance of the model and monitor its classification accuracy and other indicators on unseen data. The training iteration of the model will continue until the performance indicators on the validation set no longer improve for consecutive multiple training periods, or the maximum number of preset iteration periods is reached, at which point the training stops and the model parameters with the best validation performance are saved, completing the training of the building structure health classification model.

[0140] In the specific implementation, S7 is specifically as follows:

[0141] After the building structure health classification model is trained, it can be deployed and applied to actual building structure health monitoring. During monitoring, the system collects real-time or periodic raw vibration signals through vibration sensors installed at key parts of the building structure to form vibration data to be monitored. These data first flow into the preprocessing and feature extraction pipeline consistent with the training stage, and are sequentially subjected to adaptive wavelet denoising and dynamic range normalization processing in the S2 step, and adaptive window short-time Fourier transform and adaptive time-frequency feature enhancement in the S3 step, to be converted into an enhanced time-frequency feature matrix. At the same time, the system synchronously collects environmental temperature and humidity data to form an external environmental factor feature vector, and extracts a statistical feature vector from the preprocessed time-domain signal.

[0142] Then, the processed features are input into the trained building structure health classification model. The model performs forward inference, and after a series of internal calculations such as structure physical prior guided feature modulation, multi-source feature fusion, and multi-scale space-time interaction coding, finally outputs the probability of the building structure belonging to each of the categories of "healthy", "slight damage", "moderate damage", and "severe damage" in the monitoring period.

[0143] The system determines the health state of the current structure according to the category corresponding to the highest probability, and visualizes the monitoring results, probability distribution and key features to the user. Once the system determines that there is a damage state (minor, moderate or severe), different levels of early warning will be triggered automatically, thereby realizing automated and intelligent building structure health state evaluation and safety warning.

[0144] Embodiment 2

[0145] A machine learning-based building structure health monitoring system, which executes a machine learning-based building structure health monitoring method, has the following system structure:

[0146] A three-dimensional numerical model and simulation data set generation module is configured to construct a three-dimensional numerical model of a target building using finite element simulation technology, simulate dynamic responses under different load conditions to generate simulation signals, inject Gaussian white noise and adjust model parameters to generate a multi-category simulation data set, and label sample health categories.

[0147] A data preprocessing and enhancement module is configured to process the simulation data set to generate enhanced data using an adaptive wavelet denoising combined with a dynamic normalization two-stage method.

[0148] A time-frequency feature extraction and enhancement module is configured to extract high-resolution time-frequency features from the enhanced data by adaptive window short-time Fourier transform combined with adaptive frequency band enhancement, and to strengthen damage-related feature components to generate an enhanced time-frequency feature matrix.

[0149] A health classification model construction and training module is configured to construct a neural network model that integrates structural physical prior guidance and multi-scale space-time interaction, to perform feature modulation, multi-source fusion and external factor injection on the enhanced time-frequency feature matrix, to obtain refined feature vectors through multi-scale space-time interaction encoding, to construct a total loss function based on feature consistency and prediction smoothness regularization terms, and to train the model.

[0150] A health monitoring and early warning module is configured to deploy the trained model to evaluate the health state of the building structure, output damage probability distribution, and realize safety warning.

[0151] Embodiment 3

[0152] As shown in Figure 5 , the classification accuracy of different classification methods under four damage states is compared to evaluate the classification performance of the neural network model constructed by the present application, which integrates structural physical prior guidance and multi-scale space-time interaction. The conventional technologies compared include traditional support vector machines based on handcrafted features, convolutional neural networks using only time-frequency images, and long short-term memory networks processing sequential data. The present technology implements a complete process from physical prior guidance modulation to multi-scale space-time interaction encoding. Figure 5The middle horizontal coordinate represents four structural health states, and the vertical coordinate represents the classification accuracy in percentage. The experimental results show that the accuracy of the method of the application is significantly higher than that of the other three comparison methods under each health state. It is particularly noteworthy that the accuracy of the method of the application is particularly obvious in the two states of slight damage and moderate damage, which are difficult to distinguish and easy to misjudge, while the accuracy of the traditional support vector machine is relatively low in these two categories, proving that the model of the application can capture the subtle differences between different damage states more finely by integrating health benchmark modal prior knowledge, multi-source feature fusion and modeling feature channel relationship, thereby realizing high-precision health state recognition.

[0153] Although the specific embodiments of the application are described above with reference to the drawings, the description is not a limitation on the scope of protection of the application. Various modifications or variations made by those skilled in the art without creative labor on the basis of the technical solutions of the application are still within the scope of protection of the application.

Claims

1. A method for monitoring the health of a building structure based on machine learning, characterized by, The method comprises the following steps: S1, a three-dimensional numerical model of the target building is constructed by using finite element simulation technology, the dynamic response under different load conditions is simulated through the three-dimensional numerical model, the simulation signals corresponding to the real acquisition situation are generated, and the building health monitoring data is constructed; S2, Gaussian white noise is injected into the simulation signal to simulate simulation noise, and the parameters of the three-dimensional numerical model are adjusted to generate different types of simulation data sets, and each sample is labeled with a health category to form a data set; S3, the data in the data set is preprocessed by using a two-stage preprocessing method of adaptive wavelet denoising combined with dynamic normalization for data enhancement, and enhanced data is generated; S4, a two-stage method of adaptive window short-time Fourier transform combined with adaptive frequency band enhancement is used to extract high-resolution time-frequency features from the enhanced data, and the feature components related to structural damage are strengthened to generate an enhanced time-frequency feature matrix; S5, a neural network model is constructed by fusing structure physical prior guidance and multi-scale space-time interaction to build a building structure health classification model, the enhanced time-frequency feature matrix is subjected to feature modulation guided by structure physical prior, and then the modulation result is subjected to multi-source feature fusion and external factor injection to generate a mixed feature vector, and then multi-scale space-time interaction coding is performed to obtain a refined feature vector, and then the refined feature vector is subjected to damage mode classification to output the probability distribution of the damage state; S5 is as follows: S5.1, the enhanced time-frequency feature matrix of all health state samples in the data set is estimated to represent the baseline mode atlas of the typical distribution of time-frequency energy under the health state, and the enhanced time-frequency feature matrix is subjected to attention modulation and bias enhancement guided by physical guidance to generate a physical perception feature matrix fused with health baseline prior knowledge; S5.2, the physical perception feature matrix is flattened into a vector, a group of time domain statistical and envelope spectrum features are extracted from the denoising time domain signal to form a statistical feature vector, and an external environmental factor feature vector is introduced, and the three parts of information are adaptively weighted and fused by a feature fusion gate unit to obtain a mixed feature vector; S5.3, a multi-scale space-time interaction block is used to extract different time receptive field features through parallel multi-dilation rate time-space hole convolution, and the features are interactively modulated with the inter-feature channel relationship modeled by adaptive feature channel interaction, and a refined feature vector is output after stacking multiple modules; Wherein, the input of the first multi-scale space-time interaction block is the mixed feature vector, and the output of the last multi-scale space-time interaction block is the input of the next multi-scale space-time interaction block; S5.4, the refined feature vector is mapped to the corresponding non-normalized score of different structure health states through a fully connected layer, and a Softmax function combining adaptive temperature scaling and class prior weight is used to obtain the final classification probability distribution; S6, the building structure health classification model learning process is constrained based on the regularization terms of feature consistency and prediction smoothness, the total loss function of the model is calculated, and the model is trained based on the loss function; S7, the trained building structure health classification model is deployed to the actual building structure health monitoring to evaluate and warn the building structure health state.

2. The method of claim 1, wherein the method further comprises: S1 is as follows: According to the actual geometric size, material properties, component connection mode and boundary conditions of the target building structure, a three-dimensional numerical model is constructed; Different load cases include environmental excitation, simulated wind load and simulated traffic load; The simulation signal is a vibration acceleration time history signal corresponding to the actual acquisition situation, and the actual acquisition situation is to deploy sensors on the target building structure and collect real vibration acceleration time history signals through the sensors.

3. The method of claim 1, wherein the method further comprises: S2 is specifically as follows: The adjustment method of the three-dimensional numerical model parameters includes locally reducing the material stiffness, introducing micro-crack elements and changing the properties of the connectors. By adjusting the parameters, the structural damage states at different positions and different severities are simulated, and the simulation data sets of different health categories are generated, including healthy, slight damage, severe damage and serious damage; Taking the data points of the simulation signal within a fixed length as a sample, the health category label is labeled for each sample according to the preset damage state during simulation.

4. The method of claim 1, wherein the method further comprises: S3 is specifically as follows: S3.1, stationary wavelet transform is performed on the building structure health monitoring data in the data set, and then according to the difference between the local characteristics of the signal and the statistical characteristics of each scale, the adaptive threshold weight of each data point on each scale is dynamically calculated, and the denoising monitoring data is obtained based on the adaptive threshold weight; S3.2, the enhanced data is subjected to dynamic range normalization processing to obtain enhanced data; Specifically, the scaling factor based on local statistical characteristics is used to dynamically adjust the global normalization result to obtain normalized monitoring data.

5. The method of claim 1, wherein the method further comprises: S4 is specifically as follows: S4.1, the enhanced data is mapped from time domain to time-frequency domain by adaptive window short-time Fourier transform, and the adaptive window short-time Fourier transform uses a window function with adaptive bandwidth to perform segmentation processing on the time axis to generate a high-resolution initial time-frequency feature matrix; S4.2, the global average energy of each frequency bin of the initial time-frequency feature matrix is calculated, and the global frequency band weight is generated accordingly to strengthen the high-energy frequency band, and the local time-frequency attention weight is calculated to focus on the time-frequency points with energy mutation, and then the two weights are respectively applied to the initial time-frequency feature matrix, and the enhanced time-frequency feature matrix is obtained after addition.

6. The method of claim 1, wherein the method further comprises: The calculation process of the total loss function is specifically as follows: (1) the loss term combining adaptive marginal cross-entropy and class center constraint is used to simultaneously shorten the distance between sample features and their own class center and to increase the distance between other class centers, and the marginal value is adaptively adjusted according to the difficulty of class distinction; (2) based on the prior knowledge that damage evolution is continuous and should have consistency in time-frequency response, the regularization loss term containing time-frequency feature consistency constraint and prediction distribution smoothness constraint is used to encourage similar features of samples of the same class and smooth changes for continuous or adjacent states, which is expressed as: (3) the total loss is the weighted sum of the adaptive marginal contrast center loss and the regularization loss.

7. The method of claim 1, wherein the building is a building. The training process of the structural health classification model is specifically as follows: The data set is divided into a training data set and a validation data set, and the data in the training data set is processed by S3 and S4, and the model is trained end-to-end; In each training iteration, a batch of sample data is passed through the forward propagation of the model to obtain the predicted probability distribution. Then, the loss value of the current batch is calculated based on the total loss function. An adaptive moment estimation optimizer is used to calculate the gradient of all trainable parameters in the model based on the total loss function, and these parameters are updated through the backpropagation algorithm to minimize the total loss function. During training, model performance is evaluated periodically using a separate validation dataset; The training iterations of the model will continue until the performance metrics on the validation set no longer improve over multiple consecutive training cycles, or until the preset maximum number of iteration cycles is reached. At this point, training stops, the model parameters with the best validation performance are saved, and the training of the building structure health classification model is completed.

8. The method of claim 1, wherein the method further comprises: S7 is detailed below: After the building structure health classification model is trained, it is deployed in actual building structure health monitoring. The original vibration signals are collected in real time or periodically by vibration sensors installed on key parts of the building structure to form the vibration data to be monitored. At the same time, ambient temperature and humidity data are collected to form the feature vector of external environmental factors. Then, after the collected data is processed, it is input into the pre-trained building structure health classification model. The model outputs the probability that the building structure belongs to each health category during the monitoring period. The category corresponding to the highest probability is determined as the health status of the current structure, and the monitoring results, probability distribution, and key features are visualized. Then, based on the current health status of the structure, different levels of warnings will be automatically triggered.

9. A machine learning based building structure health monitoring system, which performs a machine learning based building structure health monitoring method according to any one of claims 1 to 8, characterized in that, The system structure is as follows: 3D Numerical Model and Simulation Dataset Generation Module: Used to construct a 3D numerical model of the target building using finite element simulation technology, simulate the dynamic response under different load conditions to generate simulation signals, inject Gaussian white noise and adjust model parameters to generate multi-class simulation datasets, and label the health categories of samples. Data preprocessing and enhancement module: A two-stage method combining adaptive wavelet denoising and dynamic normalization is used to process the simulation dataset and generate enhanced data; Time-frequency feature extraction and enhancement module: Through adaptive window short-time Fourier transform combined with adaptive frequency band enhancement, high-resolution time-frequency features are extracted from the enhanced data, and damage-related feature components are enhanced to generate an enhanced time-frequency feature matrix; Health classification model construction and training module: Construct a neural network model that integrates structured physical prior guidance and multi-scale spatiotemporal interaction, perform feature modulation, multi-source fusion and external factor injection on the enhanced time-frequency feature matrix, obtain refined feature vectors through multi-scale spatiotemporal interaction encoding, construct the total loss function based on feature consistency and prediction smoothness regularization term and train the model; Health monitoring and early warning module: Deploy a pre-trained model to assess the health status of building structures, output damage probability distribution, and provide safety early warning.

Images