Bridge vortex vibration identification method and system based on convolutional neural network

Abstract

The invention provides a bridge vortex vibration identification method and system based on a convolutional neural network, and belongs to the technical field of bridge monitoring. The method comprises the following steps: acquiring bridge dynamic response data; carrying out normalization processing on the bridge dynamic response data, changing a Grubrum angle field, and converting a one-dimensional time sequence signal into a two-dimensional picture; establishing a convolutional neural network model, and training the neural network model as a bridge vortex vibration recognition model; a short-time sliding window is utilized to divide monitoring data of acceleration response into a short-time domain data sequence and convert the short-time domain data sequence into a two-dimensional picture, and the system carries out bridge vortex vibration recognition and gives an alarm through a preset convolutional neural network model. According to the invention, bridge monitoring data and a deep learning technology are combined, a vortex vibration identification and alarm system is constructed, manual intervention is not needed, the efficiency and accuracy of vortex vibration identification are improved, and the technical problems of low accuracy, poor universality and high false alarm rate of a traditional bridge vortex vibration identification algorithm can be solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bridge structure monitoring, and in particular to a bridge vortex vibration identification method and system based on a convolutional neural network. BACKGROUND

[0002] Large-span bridges are sensitive to wind loads due to their light structure and large flexibility, and vortex-induced vibration (Vortex-Induced Vibration, VIV for short) occurs. Vortex-induced vibration is a periodic vibration phenomenon caused by vortex shedding in wind load, although its amplitude is usually small and does not directly cause bridge collapse, but long-term vortex-induced vibration may cause fatigue damage to the bridge structure, and thus cause serious safety problems. In recent years, vortex-induced vibration events have occurred in many large-span bridges at home and abroad, for example, in 2020, a large-span suspension bridge caused a severe vibration of the bridge deck due to vortex-induced vibration, which attracted widespread social attention.

[0003] At present, bridge health monitoring systems have been widely used in large-span bridges, which can obtain real-time information of bridge structural response and environmental load, etc., providing important data support for analyzing the vibration state of the bridge. However, the existing methods mostly require human intervention (such as threshold setting, spectral feature screening), and for different bridges, the traditional method needs to be modeled repeatedly for each bridge, which has low processing efficiency, strong subjectivity and low universality. With the exponential growth of target bridges and monitoring data, the real-time performance and automation level of traditional methods cannot meet the actual demand.

[0004] Therefore, how to effectively and quickly identify vortex-induced vibration and issue an alarm to improve the universality of the bridge vortex-induced vibration identification system is a problem to be solved in the field of bridge structure monitoring. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a bridge vortex vibration identification method and system based on a convolutional neural network to solve the technical problems of low processing efficiency and poor universality of the bridge vortex vibration identification and alarm system in the prior art.

[0006] The technical solution adopted by the present application is to convert the bridge acceleration data into a Gram angle field image, and use a convolutional neural network for training and classification to achieve accurate identification of vortex-induced vibration events.

[0007] In a first implementation manner, a bridge vortex vibration identification method based on a convolutional neural network includes: Collecting acceleration data of a bridge.

[0008] Dividing the acceleration data into a plurality of short-time domain data sequences through a short-time sliding window.

[0009] Each of the time-domain data sequences is normalized to generate a normalized data sequence.

[0010] The normalized data sequence is converted into polar coordinates.

[0011] The correlation matrix between the polar coordinates over time is calculated using the Gram angle field method, and the correlation matrix is ​​then converted into a two-dimensional image.

[0012] The two-dimensional image is input into a pre-trained convolutional neural network model for feature extraction and classification, generating a result indicating whether vortex-induced vibration has occurred.

[0013] An alarm is triggered when the identification results indicate that a vortex-induced vibration event has occurred.

[0014] Furthermore, each of the time-domain data sequences is normalized to generate a normalized data sequence, including: Each of the time-domain data sequences is normalized according to the normalization formula: in This represents the normalized value of the i-th acceleration data in the time-domain data sequence. Represents the time-domain data sequence The i-th acceleration data in the data. This represents the minimum value in the time-domain data sequence. This represents the minimum value in the time-domain data sequence.

[0015] Furthermore, the normalized data sequence is converted into polar coordinate representation, including: According to the polar coordinate transformation formula, the normalized data sequence is converted into a polar coordinate representation: in, The time for the i-th sample point (if using actual time). The value in X. For corresponding The timestamp, where D is the total length of the timestamp. Polar angle, It is the polar radius.

[0016] Using the normalized numerical code as the angle cosine value and the timestamp as the polar radius, the polar coordinates are represented as ( ).

[0017] Furthermore, the correlation matrix is ​​calculated according to the Gram angle field calculation formula described below: Wherein, G represents the correlation matrix, d represents the length of the time series, represents the dth polar angle.

[0018] Preferably, the convolutional neural network model comprises: An input layer, a two-dimensional image with an input size of 100x100x1.

[0019] A first convolutional layer using 8 convolutional kernels with a size of 7x7 and a step size of 1, outputting a feature image with a size of 94x94x8.

[0020] A first max-pooling layer using a 3x3 pooling window with a step size of 2, outputting a feature image with a size of 46x46x8.

[0021] A second convolutional layer using 16 convolutional kernels with a size of 6x6 and a step size of 1, outputting a feature image with a size of 41x41x16.

[0022] A second max-pooling layer using a 3x3 pooling window with a step size of 2, outputting a feature image with a size of 20x20x16.

[0023] A fully connected layer for combining convolutional features into a global feature vector.

[0024] An output layer using a classifier with a size of 2x1 for outputting a binary classification result of the vortex event.

[0025] Further, the two-dimensional vector in the classifier is used to represent the binary classification result, including classification probability values for representing two categories of 'vortex occurrence' and 'no vortex occurrence'.

[0026] The output layer determines the vortex state corresponding to the data segment according to the size of the classification probability value.

[0027] Preferably, after obtaining the binary classification result, further comprising: Comparing the recognition result output by the convolutional neural network with the labeled true label to generate precision and recall to evaluate the recognition performance of the convolutional neural network.

[0028] The accuracy of the classification result is measured according to the precision calculation formula and the recall calculation formula as follows: Wherein, P represents the precision, R represents the recall, TP is the true positive, representing the number of samples correctly classified for each type of data; FP is the false positive, representing the number of samples classified as this type of image by the model from other types of images; FN is the false negative, representing the number of samples classified from this type of image to other types of images.

[0029] In combination with the first implementation, in a second implementation, a bridge vortex vibration identification system based on a convolutional neural network comprises: A data acquisition module configured to acquire acceleration data of the bridge.

[0030] A data processing module configured to divide the acceleration data into a plurality of shorter time-domain data sequences through a short sliding window, perform normalization processing on each of the time-domain data sequences to generate normalized data sequences, convert the normalized data sequences into polar coordinate representations, calculate a correlation matrix between the polar coordinates according to a Gram angle field method, and convert the correlation matrix into a two-dimensional image.

[0031] A convolutional neural network identification module configured to input the two-dimensional image into a pre-trained convolutional neural network model to perform feature extraction and classification, and generate an identification result of whether vortex vibration occurs.

[0032] An alarm module configured to trigger an alarm when the identification result indicates that a vortex vibration event occurs.

[0033] Preferably, the data acquisition module comprises acceleration sensors and / or deflection sensors arranged at the midspan, 1 / 4 span and 3 / 4 span of the bridge.

[0034] In combination with the first implementation, in a third implementation, a computer device comprises a memory and a processor, the memory stores computer instructions capable of running on the processor, and the processor runs the computer instructions to perform a bridge vortex vibration identification method based on a convolutional neural network.

[0035] From the above technical solutions, the present application has the following beneficial technical effects: 1. The time series signal is converted into a two-dimensional image by the Gram angle field method, the complete time domain information is retained, and a pre-trained convolutional neural network is used for feature extraction and classification, thereby improving the accuracy and stability of vortex vibration identification.

[0036] 2. The convolutional neural network can learn the cross-structure invariance of vortex vibration features, and the pre-trained model only needs a small amount of fine-tuning when deployed, thereby overcoming the limitation of traditional methods that require repeated modeling for each bridge, and significantly enhancing the engineering applicability.

[0037] 3. The model identification performance is quantitatively evaluated by precision and recall indicators, which facilitates subsequent monitoring and optimization of model effectiveness. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the specific embodiments of the present application or the prior art, the drawings required to be used in the specific embodiments or prior art description will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual proportions.

[0039] Figure 1 Flow chart of the method of embodiment 1 of the present application; Figure 2 Schematic diagram of data processing of embodiment 1 of the present application; Figure 3 System structure diagram of embodiment 2 of the present application. DETAILED DESCRIPTION

[0040] The embodiments of the technical solutions of the present application will be described in detail below with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present application, and therefore only serve as examples, and cannot limit the protection scope of the present application.

[0041] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in the present application should be understood as the usual meanings understood by the skilled person in the field to which the present application belongs.

[0042] Embodiment 1 The present embodiment provides a bridge vortex vibration identification method based on convolutional neural network, and the working principle of embodiment 1 is described in detail as follows: The flow chart of the method of the present embodiment is shown in Figure 1 , which includes: Collecting dynamic response acceleration data of the bridge.

[0043] Dividing the acceleration data into a plurality of shorter time domain data sequences through a short sliding window.

[0044] Normalizing each of the time domain data sequences to map the data values to the interval [-1, 1] to generate a normalized data sequence.

[0045] Converting the normalized data sequence into polar coordinate representation.

[0046] Calculating the correlation matrix between the polar coordinates according to the Gram angle field method, and converting the correlation matrix into a two-dimensional image.

[0047] Inputting the two-dimensional image into a pre-trained convolutional neural network model for feature extraction and classification to generate an identification result of whether vortex vibration occurs.

[0048] Triggering an alarm when the identification result indicates that a vortex vibration event occurs.

[0049] By automating the entire process from acceleration data acquisition to convolutional neural network classification, bridge vortex-induced vibration events can be identified quickly and accurately, and timely alarms can be triggered when events occur, thereby improving bridge operational safety and emergency response efficiency.

[0050] In this embodiment, further, each of the time-domain data sequences is normalized to generate a normalized data sequence, including: normalizing each of the time-domain data sequences according to a normalization formula: in This represents the normalized value of the i-th acceleration data in the time-domain data sequence. Represents the time-domain data sequence The i-th acceleration data in the data. This represents the minimum value in the time-domain data sequence. This represents the minimum value in the time-domain data sequence. Normalization eliminates differences in signal amplitude at different measurement points and time periods, unifying the input data to the same numerical range, enhancing the robustness of subsequent feature extraction, and improving recognition accuracy.

[0051] In this embodiment, the normalized data sequence is further converted into polar coordinate representation, including: According to the polar coordinate transformation formula, the normalized data sequence is converted into a polar coordinate representation: in, For the time of the i-th sample point, The value in X. For corresponding The timestamp, where D is the total length of the timestamp. Polar angle, Polar radius; Using the normalized numerical code as the angle cosine value and the timestamp as the polar radius, the polar coordinates are expressed as: ( ).

[0052] In this embodiment, further, for any element in the correlation matrix... ,according to = as well as We can obtain: = Therefore, the correlation matrix is ​​calculated according to the Gram angle field calculation formula described below: wherein G represents a correlation matrix, each element represents a correlation between the ith time point and the jth time point, encoded by the sum of two polar angles, d represents the length of the time series, represents the dth polar angle. In this embodiment, the value range of the correlation matrix G is [−1, 1], and in order to generate a gray image, a linear mapping is performed on the correlation matrix G, and the value range is mapped from [−1, 1] to [0, 1]: In this embodiment, the normalization processing and Gram angle field change are performed, and the data processing schematic diagram for converting the one-dimensional time sequence signal into a two-dimensional picture is as shown in Figure 2 In this embodiment, the length of the short-time sliding window is set to 5 seconds, and the sampling points in the window are uniformly distributed 20 sampling points, that is, the length of the time series is 20, as shown in Figure 2 (a) shows the acceleration detection data after normalization processing.

[0053] After polar coordinate conversion, the polar coordinate mapping diagram shown in Figure 2 (b) is obtained, the circumferential direction represents the polar angle, which is calculated from the normalized , and is used to reflect the amplitude information at this moment; the polar radius is obtained by time normalization, and reflects the time sequence.

[0054] The correlation between the time points is calculated to obtain the matrix diagram shown in Figure 2 (c), a linear mapping is performed on the correlation matrix G, and the value range is mapped from [−1, 1] to [0, 1], to obtain the gray image shown in Figure 2 (d), so that the convolutional neural network can use the image feature extraction capability to learn the dynamic pattern of the time series. The Gram angle field method is used to generate the correlation matrix, realize the two-dimensional representation of the global relationship of the time series, effectively capture the pattern characteristics of the vortex vibration signal, and adapt to the input format of the convolutional neural network.

[0055] In this embodiment, further, the convolutional neural network model is a model pre-trained on a plurality of bridge monitoring data sets, and is fine-tuned on the target bridge to reduce the modeling workload for different bridges. The convolutional neural network model comprises: an input layer, which inputs a two-dimensional image with an input size of 100x100x1; a first convolutional layer, which uses 8 convolutional kernels with a size of 7x7 and a step size of 1, and outputs a feature image with a size of 94x94x8; a first maximum pooling layer, which uses a 3x3 pooling window and a step size of 2, and outputs a feature image with a size of 46x46x8; a second convolutional layer, which uses 16 convolutional kernels with a size of 6x6 and a step size of 1, and outputs a feature image with a size of 41x41x16; The second max-pooling layer uses a 3x3 pooling window with a step of 2, and outputs a feature map with a size of 20x20x16; The full connection layer is used to combine the convolutional features into a global feature vector; The output layer uses a classifier with a size of 2x1 to output a binary classification result of the vortex event.

[0056] In the embodiment, the two-dimensional vector in the classifier is used to represent the binary classification result, including classification probability values representing two categories of 'vortex occurrence' and 'no vortex occurrence'. The output layer determines the vortex state corresponding to the data segment according to the size of the classification probability value, and takes the category with a larger probability value as the result of the vortex event. For example, in the embodiment, the category with a larger probability is taken as the recognition result, and when the 'vortex occurrence' probability value is greater than the 'no vortex occurrence' probability value, the recognition result of vortex occurrence is output. Feature extraction and classification are performed by the convolutional neural network with a specific structure, which can automatically learn the discriminative features of the vortex event, avoid manual setting of features, and improve the accuracy and generalization ability of recognition. The pre-trained model only needs a small amount of fine-tuning when deployed in other places, overcoming the limitation of traditional methods that require repeated modeling for each bridge, and significantly enhancing the engineering applicability.

[0057] In the embodiment, further, after obtaining the binary classification result, the method further includes: comparing the recognition result output by the convolutional neural network with a labeled true label to generate a precision and a recall to evaluate the recognition performance of the convolutional neural network. The true label is a sample from historical monitoring data labeled by manual labeling, indicating whether it is a vortex event. The generated precision and recall are used to verify the recognition accuracy and engineering usability of the model, and the performance indicators of the model are output together with the bridge vortex judgment result for measurement for the user to refer to.

[0058] In the embodiment, the accuracy of the classification result is measured according to the following precision calculation formula and recall calculation formula: wherein P represents the precision, R represents the recall, TP is the true positive, representing the number of samples correctly classified for each type of data; FP is the false positive, representing the number of samples classified as the current type of image by the model; FN is the false negative, representing the number of samples classified from the current type of image to other types of image.

[0059] Embodiment 2 In combination with Embodiment 1, the system structure diagram of the embodiment is as follows: Figure 3As shown, a bridge vortex vibration identification system based on a convolutional neural network comprises: a data acquisition module for acquiring acceleration data of a bridge. A data processing module is configured to divide the acceleration data into a plurality of shorter time domain data sequences by a short sliding window, normalize each time domain data sequence to generate a normalized data sequence, convert the normalized data sequence into a polar coordinate representation, calculate a correlation matrix between the polar coordinates according to the Gram angle field method, and convert the correlation matrix into a two-dimensional image.

[0060] A convolutional neural network identification module is configured to input the two-dimensional image into a pre-trained convolutional neural network model for feature extraction and classification, and generate an identification result of whether vortex vibration occurs. An alarm module is configured to trigger an alarm when the identification result indicates that a vortex vibration event occurs. When the identification result is "vortex vibration occurs", the alarm module triggers an alarm, sends an audible and visual alarm signal, and sends a remote alarm message to related equipment.

[0061] In this embodiment, further, the data acquisition module comprises acceleration sensors and / or deflection sensors arranged at the midspan, 1 / 4 span and 3 / 4 span of the bridge, and the data processing module is configured to set the short sliding window length to between 2 seconds and 10 seconds to adapt to the dynamic characteristics of different bridges.

[0062] Embodiment 3 In combination with Embodiment 1, this embodiment comprises a computer device comprising a memory and a processor, the memory storing computer instructions capable of running on the processor, and the processor executing the computer instructions to perform the vortex vibration identification method based on a convolutional neural network in Embodiment 1.

[0063] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the description of the present application.

Claims

1. A bridge vortex-induced vibration identification method based on convolutional neural networks, characterized in that, include: Collect acceleration data of the bridge; The acceleration data is divided into multiple short-time domain data sequences using a short-time sliding window. Each of the time-domain data sequences is normalized to generate a normalized data sequence; The normalized data sequence is converted into polar coordinate representation, and the Gram angle field method is used to determine the correlation matrix between time and polar coordinates in each data sequence. The correlation matrix is ​​then converted into a two-dimensional image. Each two-dimensional image is input into a pre-trained convolutional neural network model for feature extraction and classification to obtain the probability of vortex-induced vibration events in each two-dimensional image.

2. The bridge vortex-induced vibration identification method based on convolutional neural networks according to claim 1, characterized in that, Each of the time-domain data sequences is normalized to generate a normalized data sequence, including: Each of the time-domain data sequences is normalized according to the normalization formula: in This represents the normalized value of the i-th acceleration data in the time-domain data sequence. Represents the time-domain data sequence The i-th acceleration data in the data. This represents the minimum value in the time-domain data sequence. This represents the minimum value in the time-domain data sequence.

3. The bridge vortex-induced vibration identification method based on convolutional neural networks according to claim 1, characterized in that, Convert the normalized data sequence into polar coordinates, including: According to the polar coordinate transformation formula, the normalized data sequence is converted into a polar coordinate representation: in, For the time of the i-th sample point, The value in X. For corresponding timestamp, Polar angle; Using the normalized numerical code as the angle cosine value and the timestamp as the polar radius, the polar coordinates are expressed as: ( ).

4. The bridge vortex-induced vibration identification method based on convolutional neural networks according to claim 1, characterized in that, Calculate the correlation matrix according to the following Gram angle field calculation formula: Where G represents the correlation matrix (each element represents the correlation between the i-th and j-th time points, encoded by the sum of two polar angles), and d represents the length of the time series. This represents the d-th polar angle.

5. The bridge vortex-induced vibration identification method based on convolutional neural networks according to claim 1, characterized in that, The convolutional neural network model includes: Input layer, take a two-dimensional image with a size of 100×100×1; The first convolutional layer uses 8 convolutional kernels of size 7×7 with a stride of 1, and outputs a feature image of size 94×94×8. The first max pooling layer uses a 3×3 pooling window with a stride of 2, and the output feature image size is 46×46×8. The second convolutional layer uses 16 convolutional kernels of size 6×6 with a stride of 1, and outputs a feature image of size 41×41×16. The second max pooling layer uses a 3×3 pooling window with a stride of 2, and the output feature image size is 20×20×16. Fully connected layers are used to combine convolutional features into a global feature vector; The output layer uses a 2×1 classifier to output the binary classification results of vortex-induced vibration events.

6. The bridge vortex-induced vibration identification method based on convolutional neural networks according to claim 5, characterized in that, The two-dimensional vector in the classifier is used to represent the binary classification result, including classification probability values ​​for the two categories 'vortex oscillation occurred' and 'vortex oscillation did not occur'. The output layer determines the vortex-induced vibration state corresponding to the data segment based on the magnitude of the classification probability value.

7. The bridge vortex-induced vibration identification method based on convolutional neural networks according to claim 5, characterized in that, Training a convolutional neural network model includes: The recognition results output by the convolutional neural network are compared with the labeled real labels to generate precision and recall, thereby evaluating the recognition performance of the convolutional neural network.

8. A bridge vortex-induced vibration identification system based on a convolutional neural network, characterized in that, include: The data acquisition module is used to collect the acceleration data of the bridge. The data processing module is used to divide the acceleration data into multiple shorter time-domain data sequences through a short-time sliding window, normalize each time-domain data sequence to generate a normalized data sequence, convert the normalized data sequence into polar coordinate representation, calculate the correlation matrix between the polar coordinates over time according to the Gram angle field method, and convert the correlation matrix into a two-dimensional image. The convolutional neural network recognition module is used to input the two-dimensional image into a pre-trained convolutional neural network model for feature extraction and classification, and to generate a recognition result of whether vortex oscillation has occurred. The alarm module is used to trigger an alarm when the identification results indicate that a vortex-induced vibration event has occurred.

9. A bridge vortex-induced vibration identification system based on a convolutional neural network according to claim 7, characterized in that, The data acquisition module includes acceleration sensors and / or deflection sensors deployed at the mid-span, 1 / 4-span, and 3 / 4-span locations of the bridge.

10. A computer device comprising a memory and a processor, wherein the memory stores computer instructions executable on the processor, characterized in that, When the processor executes the computer instructions, it performs a bridge vortex vibration identification method based on any one of claims 1 to 6.

Images