A method and apparatus for operational modal analysis of a building structure based on rbmamba2

Abstract

A method and apparatus for operational modal analysis of building structures based on RBMamba2 are disclosed, establishing a mapping relationship between structural response and modal parameters to achieve rapid, accurate, and stable identification of structural modal parameters. The specific implementation process is as follows: A. Collect structural health monitoring data and construct training and validation sets based on the SSI-COV method; B. Construct a Mamba2 model incorporating a time-weighted loss function to establish a mapping relationship between multi-sensor monitoring data and modal parameters; C. Quantify the cognitive and data uncertainties of the model prediction results; D. Use a recursive Bayesian posterior fusion method to recursively update the modal parameters, obtaining smooth posterior estimates and their confidence intervals; E. Input the data collected by the sensors to identify the structural modal parameters. The method disclosed in this invention can achieve efficient identification of building structural modal parameters, providing support for structural condition assessment.

Description

Technical Field

[0001] This invention relates to a method and apparatus for operational modal analysis of building structures, belonging to the field of structural health monitoring technology. Background Technology

[0002] During long-term service, the structural performance of high-rise buildings inevitably declines due to corrosion, material aging, and component fatigue. Structural health monitoring (SHM) technology provides real-time structural information and plays a crucial role in ensuring the safe operation of structures. Modal parameters (such as natural frequencies, damping ratios, and mode shapes) identified through monitoring data reflect the dynamic characteristics of the structure and are important indicators for evaluating the performance of high-rise building structures. Therefore, developing an effective modal parameter identification method has significant engineering application value.

[0003] Current modal parameter identification methods mainly include operational modal analysis methods such as random subspace identification and Bayesian fast Fourier transform, which utilize structural response data to identify structural modal parameters. However, these methods do not establish a mapping relationship between multi-source monitoring data and modal parameters, nor do they provide smoothed posterior estimates and confidence intervals for modal parameters, and they still have limitations in terms of computational efficiency and automation. Therefore, it is necessary to develop a fast modal parameter identification method that can effectively capture the mapping relationship between measured data and modal parameters and consider uncertainties. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a building structure operation modal analysis method based on RBMamba2 to improve the efficiency and stability of building structure modal parameter identification.

[0005] This invention constructs training and validation sets based on the SSI-COV method. By combining Mamba2 and time-weighted loss functions, uncertainty quantification, and recursive Bayesian posterior fusion methods, it efficiently and stably identifies building structure modal parameters.

[0006] A method for operational modal analysis of building structures based on RBMamba2 establishes a mapping relationship between structural response and modal parameters, enabling rapid, accurate, and stable identification of structural modal parameters. The method includes the following steps: A. Collect structural health monitoring data and construct training and validation sets based on the SSI-COV method; B. Construct a Mamba2 model incorporating a time-weighted loss function to establish a mapping relationship between multi-sensor monitoring data and modal parameters; C. Quantify the cognitive uncertainty and data uncertainty of the model prediction results; D. The modal parameters are updated recursively using the recursive Bayesian posterior fusion method to obtain smooth posterior estimates of the modal parameters and their confidence intervals; E. Input data collected by the sensor to identify structural modal parameters.

[0007] Furthermore, step A specifically includes: A1. Collect building structural health monitoring data and divide it into N samples with a duration of T; A2. For each sample, according to the time step Constructing cumulative subsequences yields a total of The cumulative subsequence, the nth... The cumulative subsequence contains numbers from 0 to... Cumulative monitoring data; A3. The SSI-COV method was used to identify the modal parameters corresponding to each cumulative subsequence, and training and validation sets were constructed by combining the monitoring data.

[0008] Furthermore, step B specifically includes: B1. Establish a Mamba2 network, use linear projection to map the monitoring data to a unified feature dimension, and input the projected features into a convolutional module to extract local temporal features; B2. Inputting local features into the structured state space module, modeling long-range dependencies through recursive state updates, and combining this with the sigmoid-gated output, yields the hidden states and outputs at each time step, which can be represented as: (1) (2) In the formula: In time step The hidden state; , and For the system matrix; For input; For output; B3. Construct a time-weighted loss function for Mamba2 model training to reduce the impact of early SSI-COV instability identification results on model training, which can be expressed as: (3) (4) In the formula: Time weighting; This is the current time step; Total time; It is a constant.

[0009] Furthermore, step C specifically includes: C1. Using the Monte Carlo Dropout method, multiple random forward propagations are performed on the same input during the inference phase to obtain multiple prediction results. These prediction results are used to evaluate the cognitive uncertainty of the model, which can be expressed as: (5) (6) In the formula: For the first The cognitive uncertainty at each time step; For the first Sub-model predictions; The next number is the number of Mond Carlo samplings; Predict the mean for the model; C2. Using the exponentially weighted moving average method, the variance of data uncertainty over time can be estimated as follows: (7) (8) (9) In the formula: For the first The random uncertainty of each time step; This is an estimate of the first-order residual moment; These are estimates of the second-order residual moments; It is a constant; These are time-weighted coefficients; C3. Adding the uncertainty of understanding to the uncertainty of data, we obtain the total uncertainty of the model prediction, which can be expressed as: (10) In the formula: For the first The total uncertainty at each time step.

[0010] Furthermore, step D specifically includes: D1. Since the modal parameters drift slowly over time, a random walk model is used to describe the state evolution, which can be expressed as: (11) (12) In the formula: To fuse modal parameters; For process noise variance; This is the mean value predicted by the model; To observe the noise variance; D2. The prior distribution of the observed values ​​is updated and corrected using Bayesian methods, which can be expressed as: (13) (14) In the formula: For fusion gain; For time step The prior variance; For time step The posterior variance; D3. Further updates to the posterior mean and variance can be expressed as: (15) (16) In the formula: The posterior mean represents the fused estimate of the modal parameters; The prior mean; D4. The updated modal parameter posterior distribution and confidence interval can be expressed as: (17) (18) In the formula: To correspond to the confidence level The standard normal quantile.

[0011] A second aspect of the present invention relates to a building structure operation modal analysis device based on RBMamba2, comprising a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement a building structure operation modal analysis method based on RBMamba2 of the present invention.

[0012] The advantages of this invention are: (1) This invention combines Mamba2 with a time-weighted loss function, extracts local temporal features through a convolution module, and establishes a mapping relationship between monitoring data and modal parameters using a state-space model, which effectively improves the efficiency of modal parameter identification.

[0013] (2) This invention combines Monte Carlo Dropout with the exponentially weighted moving average method, taking into account both cognitive uncertainty and data uncertainty, and effectively quantifies the uncertainty of the model prediction results.

[0014] (3) The present invention uses a recursive Bayesian posterior fusion method to recursively update the modal parameters, and obtains smooth posterior estimation results and confidence intervals, which effectively improves the stability and reliability of modal parameter identification results. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a diagram showing the arrangement of dynamic strain measurement points in the South Taihu CBD Building according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the training sample construction of the present invention; Figure 4 This is a schematic diagram of the RBMamba2 framework of the present invention; Figure 5 This is a comparison chart of the frequency results identified by the present invention and other different methods; Figure 6 This is a comparison chart of the damping ratio results identified by the present invention and other different methods; Figure 7 This is a comparison chart of the vibration mode results identified by the present invention and other different methods; Figure 8 This is a comparison chart of the computational efficiency of the present invention with other different methods. Detailed Implementation

[0016] The present invention will now be described in further detail with reference to the accompanying drawings and an embodiment.

[0017] Example 1

[0018] like Figure 1 This embodiment relates to a building structure operation modal analysis method based on RBMamba2, which specifically includes the following steps: A. Collect structural health monitoring data and construct training and validation sets based on the SSI-COV method, specifically including: A1. The South Taihu CBD Building is equipped with a structural health monitoring system. Thirty-two dynamic strain sensors based on fiber Bragg gratings are deployed in the main core tube, with a sampling frequency of 10Hz. Data was continuously collected for 48 hours, yielding a total of 1,728,000 data points. The raw data was divided into 40 samples, each 4000 seconds long. The measurement points are arranged as follows: Figure 2 As shown; A2. For each sample, according to the time step Constructing cumulative subsequences yields a total of The cumulative subsequence, the nth... The cumulative subsequence contains numbers from 0 to... Cumulative monitoring data; A3. The SSI-COV method is used to identify the modal parameters corresponding to each cumulative subsequence, such as... Figure 3 As shown, the 40 samples are divided into 32 sets, which are used as the training set for the model and 8 sets are used as the test set for the model.

[0019] B. Construct a Mamba2 model incorporating a time-weighted loss function to establish a mapping relationship between multi-sensor monitoring data and modal parameters, specifically including: B1. Establish a Mamba2 network, use linear projection to map the monitoring data to a unified feature dimension, and input the projected features into a convolutional module to extract local temporal features; B2. Input local features into the structured state space module, model long-range dependencies through recursive state updates, and fuse them with the sigmoid-gated output to obtain the hidden states at each time step. and output ; B3. Construct a time-weighted loss function for training the Mamba2 model. This is to reduce the impact of early SSI-COV instability identification results on model training.

[0020] C. Quantify the cognitive uncertainty and data uncertainty of the model prediction results, specifically including: C1. The Monte Carlo Dropout method is used to perform multiple random forward propagations on the same input during the inference phase, obtaining multiple prediction results. These prediction results are then used to evaluate the cognitive uncertainty of the model. ; C2. Use the exponentially weighted moving average method to estimate the variance of data uncertainty over time. ; C3. Add the uncertainty of understanding to the uncertainty of data to obtain the total uncertainty of the model prediction. .

[0021] D. A recursive Bayesian posterior fusion method is used to recursively update the modal parameters, obtaining smooth posterior estimates of the modal parameters and their confidence intervals: D1. Since the modal parameters drift slowly over time, a random walk model is used to describe the state evolution process of the modal parameters, and a state model is established. and observation model ; D2. Based on prior variance and observation noise variance To obtain the Bayesian update gain ; D3. Using observations Correct the prior estimates of the modal parameters to obtain the posterior mean. and posterior variance ; D4. Obtain the posterior distribution of modal parameters and confidence interval .

[0022] E. Input the data collected by the sensor into... Figure 4 The RBMamba2 model shown identifies structural modal parameters and is compared with other methods. Frequency identification results are as follows: Figure 5 As shown, the damping ratio identification results are as follows: Figure 6 As shown, the mode shape identification results are as follows: Figure 7 As shown, the computational efficiency comparison results are as follows: Figure 8 As shown.

[0023] Example 2

[0024] This embodiment relates to a building structure operation modal analysis device based on RBMamba2, including a memory and one or more processors. The memory stores executable code, and when the one or more processors execute the executable code, they are used to implement a building structure operation modal analysis method based on RBMamba2 according to Embodiment 1.

[0025] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms stated in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for operational modal analysis of building structures based on RBMamba2, comprising the following steps: A. Collect structural health monitoring data and construct training and validation sets based on the SSI-COV method; B. Construct a Mamba2 model incorporating a time-weighted loss function to establish a mapping relationship between multi-sensor monitoring data and modal parameters; C. Quantify the cognitive uncertainty and data uncertainty of the model prediction results; D. The modal parameters are updated recursively using the recursive Bayesian posterior fusion method to obtain smooth posterior estimates of the modal parameters and their confidence intervals; E. Input data collected by the sensor to identify structural modal parameters.

2. The method for operational modal analysis of building structures based on RBMamba2 as described in claim 1, characterized in that, Step A specifically includes: A1. Collect building structural health monitoring data and divide it into N samples with a duration of T; A2. For each sample, according to the time step Constructing cumulative subsequences yields a total of The cumulative subsequence, the nth... The cumulative subsequence contains numbers from 0 to... Cumulative monitoring data; A3. The SSI-COV method was used to identify the modal parameters corresponding to each cumulative subsequence, and training and validation sets were constructed by combining the monitoring data.

3. The method for operational modal analysis of building structures based on RBMamba2 as described in claim 1, characterized in that, Step B specifically includes: B1. Establish a Mamba2 network, use linear projection to map the monitoring data to a unified feature dimension, and input the projected features into a convolutional module to extract local temporal features; B2. Input local features into the structured state space module, model long-range dependencies through recursive state updates, and combine this with the sigmoid-gated output to obtain the hidden states and outputs at each time step, represented as: (1) (2) In the formula: In time step The hidden state; , and For the system matrix; For input; For output; B3. Construct a time-weighted loss function for Mamba2 model training to reduce the impact of early SSI-COV instability identification results on model training, expressed as: (3) (4) In the formula: Time weighting; This is the current time step; Total time; It is a constant.

4. The method for operational modal analysis of building structures based on RBMamba2 as described in claim 1, characterized in that, Step C specifically includes: C1. Using the Monte Carlo Dropout method, multiple random forward propagations are performed on the same input during the inference phase to obtain multiple prediction results. These prediction results are used to evaluate the cognitive uncertainty of the model, expressed as: (5) (6) In the formula: For the first The cognitive uncertainty at each time step; For the first Sub-model predictions; The next number is the number of Mond Carlo samplings; Predict the mean for the model; C2. Using the exponentially weighted moving average method, the variance of the data uncertainty over time is estimated, expressed as: (7) (8) (9) In the formula: For the first The random uncertainty of each time step; This is an estimate of the first-order residual moment; These are estimates of the second-order residual moments; It is a constant; These are time-weighted coefficients; C3. Adding the uncertainty of understanding to the uncertainty of data yields the total uncertainty of the model prediction, expressed as: (10) In the formula: For the first The total uncertainty at each time step.

5. The method for operational modal analysis of building structures based on RBMamba2 as described in claim 1, characterized in that, Step D specifically includes: D1. Since the modal parameters drift slowly over time, a random walk model is used to describe the state evolution, expressed as: (11) (12) In the formula: To fuse modal parameters; For process noise variance; This is the mean value predicted by the model; To observe the noise variance; D2. The prior distribution of the observed values ​​is updated and corrected using Bayesian methods, as follows: (13) (14) In the formula: For fusion gain; For time step The prior variance; For time step The posterior variance; D3. Further update the posterior mean and variance, expressed as: (15) (16) In the formula: The posterior mean represents the fused estimate of the modal parameters; The prior mean; D4. The updated modal parameter posterior distribution and confidence intervals are expressed as follows: (17) (18) In the formula: To correspond to the confidence level The standard normal quantile.

6. A building structure operation modal analysis device based on RBMamba2, characterized in that, The system includes a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the building structure operation modal analysis method based on RBMamba2 according to any one of claims 1-5.

Images