Bridge early warning method and system based on support state

Abstract

The invention relates to a bridge early warning method and system based on a support state. The method comprises the following steps: S1, establishing a detection signal database; s2, training the enhanced YOLO-v5s model based on the detection signal database to establish a support state detection model; s3, collecting a detection signal of the target bridge support through an active sensing experiment, preprocessing the detection signal of the target bridge support, and converting the detection signal into a wavelet spectrum through continuous wavelet transform; s4, inputting the wavelet spectrum corresponding to the target bridge support detection signal into the support state detection model to obtain the state of the target bridge support; and S5, obtaining a shear stiffness change rate, a compression stiffness change rate and a support axial pressure index according to the state of the target bridge support, judging an early warning level of the target bridge, performing early warning, and outputting damage information of the support and the bridge. According to the invention, high-precision detection of the support state and early warning of the bridge state can be realized.

Description

Technical Field

[0001] The present invention relates to the fields of rubber bearing health detection and bridge detection technology, and in particular to a bridge early warning method and system based on bearing status. Background Art

[0002] Laminated rubber bearings are widely used in bridge engineering due to their excellent seismic isolation performance, simple structure, low production cost, and excellent durability. Small and medium-span bridges are the predominant structural form in my country, with simply supported beam bridges representing a significant proportion. Factors such as foundation settlement, load fluctuations, and temperature changes can cause bearing conditions to fluctuate, reducing their seismic isolation performance and further compromising structural safety. Therefore, timely and accurate bearing condition monitoring and bridge status warnings are crucial to ensuring structural safety.

[0003] Active sensing methods for detecting bearing conditions such as axial compression, void damage, debonding damage, and aging damage have been proven feasible, offering advantages such as ease of operation and high accuracy. However, to ensure the maximum safety of bridge structures, new detection methods are needed to further improve the accuracy of detection models. Furthermore, changes in bearing condition often reflect the stress state of the superstructure, playing a crucial role in providing early warning of bridge conditions. However, existing bridge early warning methods fail to effectively integrate bearing condition information.

[0004] Therefore, the present invention proposes a bridge early warning method and system based on support status with higher accuracy. Summary of the Invention

[0005] The purpose of the present invention is to provide a bridge early warning method and system based on support status, which can achieve high-precision detection of support status and realize early warning of bridge status.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A bridge early warning method based on support status, comprising:

[0008] S1. Establish a detection signal database from laminated rubber bearings with different damage types and multiple damage levels through active sensing experiments;

[0009] S2. Training the enhanced YOLO-v5s model based on the detection signal database to establish a bearing status detection model, where the bearing status includes damage type, damage degree, compression stiffness, shear stiffness, and axial pressure;

[0010] S3. collecting detection signals of the target bridge support through the active sensing experiment, preprocessing the detection signals of the target bridge support, and converting them into wavelet spectra through continuous wavelet transform;

[0011] S4. Inputting the wavelet spectrum corresponding to the target bridge bearing detection signal into the bearing state detection model to obtain the state of the target bridge bearing;

[0012] S5. According to the state of the target bridge support, the shear stiffness change rate, the compression stiffness change rate and the support axial pressure index are obtained, the warning level of the target bridge is determined and a warning is issued, and damage information of the support and the bridge is outputted at the same time.

[0013] Optionally, in S3, the detection signal of the target bridge bearing is collected through the active sensing experiment, the detection signal of the target bridge bearing is preprocessed and converted into a wavelet spectrum through continuous wavelet transform, including: collecting a one-dimensional time domain detection signal of the target bridge bearing through the active sensing experiment, filtering the one-dimensional time domain detection signal, and converting it into a two-dimensional time-frequency image, i.e., a wavelet spectrum, through continuous wavelet transform, wherein normalization is performed during the wavelet spectrum generation process.

[0014] Optionally, the support status detection model described in S2 includes: a damage type identification model, a damage degree identification model, a compression stiffness prediction model, a shear stiffness prediction model and an axial compression prediction model.

[0015] Optionally, establishing the support state detection model includes:

[0016] The wavelet spectrum in the detection signal database is used as input, and the damage type, damage degree, compression stiffness, shear stiffness and axial pressure are used as output. The enhanced YOLO-v5s model is trained to construct a bearing status detection model.

[0017] Optionally, the enhanced YOLO-v5s model is obtained by sequentially adding a Squeeze-and-Excitation module and a multi-head attention module after the neck network of the traditional pre-trained YOLO-v5s model, and replacing the detection head with a classification head or a regression head.

[0018] Optionally, the loss function of the enhanced YOLO-v5s model includes:

[0019]

[0020] Among them, L total represents the total loss function, λ MMD is a hyperparameter used to balance the weight of MMD loss relative to cross entropy loss, L MMD is the maximum mean difference loss, is the loss function on the source domain, is the loss function on the target domain.

[0021] Optionally, in S5, the shear stiffness change rate, the compression stiffness change rate, and the bearing axial pressure index are obtained according to the state of the target bridge bearing, the warning level of the target bridge is determined and a warning is issued, and damage information of the bearing and the bridge is outputted at the same time, including:

[0022] According to the shear stiffness, the compression stiffness and the axial pressure, obtaining a shear stiffness change rate, a compression stiffness change rate and a support axial pressure index, wherein the support axial pressure index includes a support axial pressure deviation rate, a lateral imbalance coefficient and a support axial pressure variation coefficient;

[0023] When the shear stiffness change rate and the compression stiffness change rate exceed the warning threshold, the warning level of the target bridge is determined and a warning is issued, and the damage type and damage degree output by the bearing state detection model are outputted;

[0024] When the support axial pressure deviation rate, the lateral imbalance coefficient and the support axial pressure variation coefficient exceed the warning threshold, the damage type of the support and the bridge is judged according to the support axial pressure index, the warning level of the target bridge is judged and a warning is issued, and the damage type of the support and the bridge is output at the same time.

[0025] The present invention also provides a bridge early warning system based on support status, including a data acquisition module, a preprocessing module, a model prediction module, and an early warning module:

[0026] A data acquisition module, configured to acquire detection signals of target bridge supports through the active sensing experiment;

[0027] A preprocessing module, used for preprocessing the detection signal of the target bridge support and converting it into a wavelet spectrum through continuous wavelet transform;

[0028] A model prediction module is used to input the wavelet spectrum corresponding to the target bridge bearing detection signal into a bearing state detection model to obtain the state of the target bridge bearing;

[0029] The early warning module is used to obtain the shear stiffness change rate, compression stiffness change rate and bearing axial pressure index according to the status of the target bridge bearing, determine the early warning level of the target bridge and issue an early warning, and output damage information of the bearing and the bridge at the same time.

[0030] The beneficial effects of the present invention are as follows: the existing technology often uses the wavelet packet energy spectrum obtained by wavelet packet decomposition of the one-dimensional time domain detection signal as the input parameter, and adopts the conventional machine learning algorithm to establish the detection model of the support status. Compared with the existing technology, the present invention combines the continuous wavelet transform with the pre-trained YOLO-v5s model, and the prediction accuracy of the detection model established on this basis can be significantly improved. The original YOLO-v5s model architecture is fine-tuned and enhanced from four aspects, including model input, attention mechanism, loss function and model output, so that it can adapt to the support damage detection task and improve its generalization ability. The bridge status early warning method based on the support status proposed in the present invention monitors the changes in the support status, timely and accurately evaluates the bridge status and issues early warnings, which is crucial to ensuring the safety of the bridge structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 A flowchart of a support state detection model established in an embodiment of the present invention;

[0033] Figure 2 This is a flow chart of a bridge early warning method based on support status according to an embodiment of the present invention;

[0034] Figure 3 This is a structural block diagram of a bridge early warning system based on support status according to an embodiment of the present invention;

[0035] Figure 4 This is an architecture diagram of the enhanced YOLO-v5s model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] Example 1:

[0039] This embodiment provides a bridge early warning method based on support status, including:

[0040] S1. Establish a detection signal database from laminated rubber bearings with different damage types and multiple damage levels through active sensing experiments;

[0041] S2. Training the enhanced YOLO-v5s model based on the detection signal database to establish a bearing status detection model, where the bearing status includes damage type, damage degree, compression stiffness, shear stiffness, and axial compression;

[0042] S3. collecting detection signals of the target bridge support through the active sensing experiment, preprocessing the detection signals of the target bridge support, and converting them into wavelet spectra through continuous wavelet transform;

[0043] S4. Inputting the wavelet spectrum corresponding to the target bridge bearing detection signal into the bearing state detection model to obtain the state of the target bridge bearing;

[0044] S5. According to the state of the target bridge support, the shear stiffness change rate, the compression stiffness change rate and the support axial pressure index are obtained, the warning level of the target bridge is determined and a warning is issued, and damage information of the support and the bridge is outputted at the same time.

[0045] Furthermore, in S3, the detection signal of the target bridge bearing is collected through the active sensing experiment, the detection signal of the target bridge bearing is preprocessed and converted into a wavelet spectrum through continuous wavelet transform, including: collecting a one-dimensional time domain detection signal of the target bridge bearing through the active sensing experiment, filtering the one-dimensional time domain detection signal, and converting it into a two-dimensional time-frequency image, i.e., a wavelet spectrum, through continuous wavelet transform, wherein normalization is performed during the wavelet spectrum generation process.

[0046] Furthermore, the support status detection model in S2 includes: damage type identification model, damage degree identification model, compression stiffness prediction model, shear stiffness prediction model and axial compression prediction model.

[0047] Specifically, the damage type identification model, damage degree identification model, compression stiffness prediction model, shear stiffness prediction model and axial pressure prediction model are established with wavelet spectrum as input parameters and damage type, damage degree label, compression stiffness, shear stiffness and axial pressure as output parameters respectively.

[0048] Furthermore, the establishment of a bearing status detection model includes:

[0049] The wavelet spectrum in the detection signal database is used as input, and the damage type, damage degree, compression stiffness, shear stiffness and axial pressure are used as output. The enhanced YOLO-v5s model is trained to construct a bearing status detection model.

[0050] Furthermore, the enhanced YOLO-v5s model is obtained by sequentially adding a Squeeze-and-Excitation module and a multi-head attention module after the neck network of the traditional pre-trained YOLO-v5s model, and replacing the detection head with a classification head or a regression head.

[0051] Specifically, the Squeeze-and-Excitation module and the multi-head attention module are used for feature channel weighting and spatial dependency capture, respectively, to enhance the model's ability to capture key features. The detection head in the YOLO-v5s model is removed and replaced with a classification head or regression head. For classification tasks, a fully connected layer maps global features to the dimension of the number of damage categories, outputting the probability distribution of each category. For regression tasks, a fully connected layer maps global features to the corresponding regression output dimension to predict the mechanical properties of the support (i.e., compression and shear stiffness).

[0052] Furthermore, the loss function of the YOLO-v5s model includes:

[0053]

[0054] Among them, L total represents the total loss function, λ MMD is a hyperparameter used to balance the weight of MMD loss relative to cross entropy loss, L MMD is the maximum mean difference loss, is the loss function on the source domain, is the loss function on the target domain.

[0055] Furthermore, in S5, the shear stiffness change rate, the compression stiffness change rate, and the bearing axial pressure index are obtained according to the state of the target bridge bearing, the warning level of the target bridge is determined, and a warning is issued. At the same time, damage information of the bearing and the bridge is output, including:

[0056] According to the shear stiffness, compression stiffness and axial pressure, the shear stiffness change rate, compression stiffness change rate and support axial pressure index are obtained, wherein the support axial pressure index includes the support axial pressure deviation rate, the lateral imbalance coefficient and the support axial pressure variation coefficient;

[0057] When the shear stiffness change rate and the compression stiffness change rate exceed the warning threshold, the warning level of the target bridge is determined and a warning is issued. At the same time, the damage type and damage degree output by the bearing status detection model are output;

[0058] When the bearing axial pressure deviation rate, lateral imbalance coefficient and bearing axial pressure variation coefficient exceed the warning threshold, the damage type of the bearing and the bridge is obtained according to the bearing axial pressure index, the warning level of the target bridge is judged and a warning is issued, and the damage type of the bearing and the bridge judged according to the bearing axial pressure index is output at the same time.

[0059] Specifically, the bearing damage degree (DShear) is divided into multiple levels based on the shear stiffness change rate (e.g., 10 levels, with each level decreasing by 1%), and the bearing damage degree (DCompressive) is divided into multiple levels based on the compression stiffness change rate (e.g., 10 levels, with each level increasing by 3%). When either the bearing damage degree (DShear) or DCompressive exceeds the warning threshold (e.g., level 2), an alert is issued according to the warning level specified in Table 2, and information on the bearing damage type and severity is output.

[0060] Specifically, if any of the three indicators exceeds the warning threshold, such as ALDR exceeding 10%, LIC exceeding 5%, or CVBL exceeding 20%, the warning level is determined according to the warning level judgment criteria described in Table 2, and an alert is issued. In addition, the damage type is preliminarily determined and output according to Table 1.

[0061] The method of this embodiment is further described below through two parts: detection model establishment and application:

[0062] 1. Such as Figure 1 As shown, the detection model establishment includes:

[0063] S11. Bearing Preparation. Through on-site collection or laboratory fabrication, a batch of laminated rubber bearings with various types of damage, including aging, peeling, and debonding, are obtained, and the detection device is installed. Common self-testing bearing structures include protective steel plates, piezoelectric ceramic transducers, and bolt covers. The piezoelectric ceramic transducers are secured to the bearing body via bolts, and the bolt covers transfer uniform loads and secure the transducer position.

[0064] S12. Active Sensing Experiment Preparation. Check that the equipment required for the active sensing experiment, including the acquisition instrument (excitation module: swept frequency signal, frequency 1 kHz-100 kHz, duration 1.0 s; acquisition module: sampling frequency 1 MHz, lead time 0.2 s, duration 1.5 s), piezoelectric ceramic transducer (longitudinal wave, resonant frequency 58 kHz, height 30 mm, diameter 38 mm), and charge amplifier (sensitivity 2.4, gain multiplier 30, allows signals with frequencies below 100 kHz to pass) are functioning properly.

[0065] S13. Data acquisition. Through active sensing experiments, collect one-dimensional time-domain detection signals of multiple bearings under different damage types, damage degrees, working axial pressures, installation errors, and other conditions.

[0066] S14. Data preprocessing: Filter the one-dimensional time-domain detection signal obtained in step 1.3 to filter out signals with frequencies below 1 kHz and prevent the detection signal from being affected by factors such as instrument vibration.

[0067] S15. Wavelet spectrum generation. A continuous wavelet transform (real-valued Morlet wavelet with scale parameters [1, 63]) is used to convert the one-dimensional detection data into a two-dimensional time-frequency image, i.e., a wavelet spectrum. During image generation, the minimum and maximum values ​​in each coefficient matrix are set to the color bar range, i.e., the coefficient matrix is ​​normalized.

[0068] The calculation process of wavelet coefficients is as follows:

[0069]

[0070] Where ψ(t) represents the mother wavelet function (real-valued Morlet wavelet), π -1 / 4 represents the normalization factor, w0 represents the dimensionless center frequency parameter, ψ s,g (t) represents the sub-wavelet function obtained by scale transformation and translation transformation, s represents the scale parameter, g represents the translation parameter, x(nΔt) represents the detection signal, Δt represents the sampling interval (Δt = 1×10 -6 s), n represents the index of the sampling point, and N represents the total number of sampling points (N = 1 × 10 6 ).

[0071] S16, Pre-trained model YOLO-v5s enhancement: The pre-trained model YOLO-v5s is used as the basic model, and the original model architecture is fine-tuned and enhanced from four aspects: model input, attention mechanism, loss function and model output. Figure 4 shown.

[0072] (1) Model input: Remove the color bars, scales, and other information from the wavelet spectrum, leaving only the pure image. During the image reading process, resize all wavelet spectra to 128×128 pixels to meet the model input requirements. Divide the database into a source domain, a target domain, and a test set using a fixed partitioning method and a ratio of approximately 8:1:1. Both the target domain and the test set are data that have not been seen in the training set (i.e., data on new supports).

[0073] (2) Attention mechanism: After the neck network, the Squeeze-and-Excitation module and the multi-head attention module are added in sequence, which are used for feature channel weighting and spatial dependency capture respectively to improve the model's ability to capture key features.

[0074] (3) Loss function: Based on the original loss function (cross entropy loss or mean square error), the maximum mean difference (MMD) loss is introduced.

[0075] D i,j =||T i -T j || 2

[0076]

[0077] L MMD =0.5(K XX +K YY -K XY -K YX )

[0078]

[0079] Where, T i and T j All represent samples, σ represents the initial bandwidth of the Gaussian kernel, n represents the total number of samples after splicing, k m Represents the kernel width multiplication factor, k n Indicates the number of Gaussian kernels, k indicates the index variable, k (k) (x,y) represents each bandwidth σ k The corresponding Gaussian kernel matrix, K XX represents the kernel matrix between the source domain and the source domain, K YY The kernel matrix between the target domain and the target domain is K XY and K YX Represents the kernel matrix between the source domain and the target domain, L total represents the total loss function, λ MMD is a hyperparameter (λ) used to balance the weight of MMD loss relative to cross entropy loss. MMD 5) is recommended.

[0080] (4) Model output: The detection head in the original architecture of the YOLO-v5s model is removed and replaced with a classification head or regression head. For classification tasks, a fully connected layer is used to map the global features to the dimension of the number of damage categories, and the probability distribution of each category is output; for regression tasks, a fully connected layer is used to map the global features to the corresponding regression output dimension to predict the mechanical properties of the support (i.e., compression stiffness and shear stiffness).

[0081] S17. Model Establishment: Establish a bearing condition detection model. Detection results include damage type, damage degree, shear stiffness, compressive stiffness, and axial pressure. The learning rate can be 0.0001, the batch size can be 64, and the number of training epochs can be 200. During training, the original weight parameters of the pre-trained model are used as the initial values, while the initial weight parameters of the newly added modules are determined by random generation. The model with the highest validation set accuracy and no obvious over- or underfitting among all epochs is selected as the final model.

[0082] Phase 1: Damage type identification: A multi-class classification model is built based on the enhanced YOLO-v5s model to identify bearing damage types such as healthy, void, debonding, and aging.

[0083] Phase 2: Sub-model refinement. For the specific type (or type combination) determined in Phase 1, a special sub-classification / regression model is constructed to output more fine-grained information (i.e., damage degree, mechanical properties, and axial pressure). Phase 2 further establishes four sub-models, including damage degree identification (classification), compression stiffness prediction (regression), shear stiffness prediction (regression), and axial pressure prediction (regression), so as to further output more fine-grained information based on the identification results of Phase 1. When modeling, the wavelet spectrum is used as the input parameter, and the damage degree label, compression stiffness, shear stiffness, and axial pressure are used as output parameters to establish the above four sub-models.

[0084] 2. Such as Figure 2 Applications of the model shown include:

[0085] S21. Obtaining the wavelet spectrum of the bearing to be detected. According to steps S11 and S13, a one-dimensional time-domain detection signal of the target bridge bearing is obtained through an active sensing experiment.

[0086] S22 , filtering the one-dimensional time domain detection signal according to S14 and S15 , and converting it into a wavelet spectrum through continuous wavelet transform.

[0087] S23. Model prediction results. The wavelet spectrum is input into the established multiple detection models to obtain the damage type of the target bridge bearing, as well as further prediction results such as the damage degree, mechanical properties (compression stiffness, shear stiffness), and axial pressure.

[0088] S24. According to the shear stiffness, compression stiffness and axial pressure, obtain the shear stiffness change rate, compression stiffness change rate, support axial pressure deviation rate, lateral imbalance coefficient and support axial pressure variation coefficient.

[0089] Calculation of support axial compression index:

[0090] Since debonding damage does not affect the compressive stiffness of bearings, the degree of aging damage to almost all bearings on the same bridge is consistent. Therefore, the axial load of the bearing is not closely related to debonding and aging damage. Therefore, three indicators are used: the bearing axial load deviation rate (ALDR), the lateral imbalance coefficient (LIC), and the bearing axial load variation coefficient (CVBL). The calculation formula for these indicators is as follows:

[0091] (1) Axial Load Deviation Ratio (ALDR):

[0092] ALDR refers to the percentage difference between the actual axial load and the design axial load of a single support. The calculation formula is as follows:

[0093]

[0094] Where P represents the actual axial pressure of the support, P design Indicates the design axial pressure of the support.

[0095] (2) Lateral Imbalance Coefficient (LIC):

[0096] LIC refers to the difference in axial pressure between the left and right supports of the same cross section. The calculation formula is as follows:

[0097]

[0098] Where, P left Indicates the actual axial pressure of the left support, P right Indicates the actual axial pressure on the right support.

[0099] (3) Coefficient of Variation for Bearing Loads (CVBL):

[0100] CVBL refers to the dispersion of the axial loads of all supports (the ratio of the standard deviation to the mean), and is calculated as follows:

[0101]

[0102] Where σ p Indicates the standard deviation of the actual axial load of all supports, μ p It represents the average of the actual axial pressure of all supports.

[0103] S25. When the shear stiffness change rate and the compression stiffness change rate exceed the warning threshold, the warning level of the target bridge is determined and a warning is issued, and the bearing damage type and damage degree are output at the same time.

[0104] The damage degree of the support (the first damage level D Shear ) is divided into multiple levels (such as ten levels, each level decreases by 1%), and the bearing damage degree (the second damage level D Compressive ) is divided into multiple levels (such as ten levels, each level increases by 3%). Shear or D Compressive When the warning threshold is exceeded (e.g., the threshold is level 2), a warning is issued according to the warning level specified in Table 2, and the damage type and degree information of the bearing is output.

[0105] S26. When the bearing axial pressure deviation rate, lateral imbalance coefficient and bearing axial pressure variation coefficient exceed the warning threshold, the damage type of the bearing and the bridge is obtained according to the bearing axial pressure index, the warning level of the target bridge is determined and a warning is issued, and the bearing and bridge damage type is output at the same time.

[0106] If any of the three indicators exceeds the warning threshold, such as ALDR exceeding 10%, LIC exceeding 5%, or CVBL exceeding 20%, the warning level is determined according to the warning level determination criteria described in Table 2, and an alert is issued. Furthermore, based on the axial pressure indicator response in Table 1, the damage type is preliminarily determined and output. Void (center) and Void (edge) are considered bearing damage types, while overload, lateral eccentric load, and foundation settlement are considered bridge damage types.

[0107] Table 1

[0108] Injury type Indicator response Empty (medium) ALDR of the hollow support = -100%, ALDR of the support on the same side of the same span increases, LIC = 0, CVBL increases Empty (edge) Empty support ALDR = -100%, LIC increases, CVBL increases overload Increased ALDR Lateral eccentric load Increased LIC and CVBL foundation settlement CVBL enlargement

[0109] Table 2

[0110]

[0111] Example 2:

[0112] like Figure 3 As shown, this embodiment proposes a bridge early warning system based on support status, including an equipment inspection module, a data acquisition module, a preprocessing module, a model prediction module, and an early warning module.

[0113] The device inspection module is used to initially collect test data and determine whether the data collector, vibrator, sensor, charge amplifier, and other equipment are functioning properly, and whether any transducer connections are loose. The data collected by this module is automatically deleted after the inspection is completed and does not enter the subsequent processing flow.

[0114] The data acquisition module collects the detection signal of the target bridge support through the active sensing experiment. The process in the data acquisition module is consistent with steps S11 and S13 of the first embodiment, and the purpose is to obtain the one-dimensional time domain detection signal of the support to be detected.

[0115] The preprocessing module is configured to filter the detection signal of the bridge bearing and convert the preprocessed detection signal into a wavelet spectrum via continuous wavelet transform. The model prediction module is configured to input the wavelet spectrum corresponding to the target bridge bearing detection signal into a bearing state detection model to obtain the state of the target bridge bearing, wherein the bearing state detection model is established in step S17 of the first embodiment.

[0116] The early warning module is used to obtain the shear stiffness change rate, compression stiffness change rate and bearing axial pressure index according to the status of the target bridge bearing, determine the early warning level of the target bridge and issue an early warning, and at the same time output damage information of the bearing and bridge.

[0117] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A bridge early warning method based on support status, characterized in that: include: S1. Establish a detection signal database from laminated rubber bearings with different damage types and multiple damage levels through active sensing experiments; S2. Training the enhanced YOLO-v5s model based on the detection signal database to establish a bearing status detection model, where the bearing status includes damage type, damage degree, compression stiffness, shear stiffness, and axial pressure; S3. collecting detection signals of the target bridge support through the active sensing experiment, preprocessing the detection signals of the target bridge support, and converting them into wavelet spectra through continuous wavelet transform; S4. Inputting the wavelet spectrum corresponding to the target bridge bearing detection signal into the bearing state detection model to obtain the state of the target bridge bearing; S5. According to the state of the target bridge support, the shear stiffness change rate, the compression stiffness change rate and the support axial pressure index are obtained, the warning level of the target bridge is determined and a warning is issued, and damage information of the support and the bridge is outputted at the same time.

2. The bridge early warning method based on support status according to claim 1 is characterized in that: In S3, the detection signal of the target bridge bearing is collected through the active sensing experiment, and the detection signal of the target bridge bearing is preprocessed and converted into a wavelet spectrum through continuous wavelet transform, including: collecting a one-dimensional time domain detection signal of the target bridge bearing through the active sensing experiment, filtering the one-dimensional time domain detection signal, and converting it into a two-dimensional time-frequency image, i.e., a wavelet spectrum, through continuous wavelet transform, wherein normalization is performed during the generation process of the wavelet spectrum.

3. The bridge early warning method based on support status according to claim 1 is characterized in that: The bearing status detection model described in S2 includes: a damage type identification model, a damage degree identification model, a compression stiffness prediction model, a shear stiffness prediction model and an axial compression prediction model.

4. The bridge early warning method based on support status according to claim 1 is characterized in that: Establishing the support state detection model includes: The wavelet spectrum in the detection signal database is used as input, and the damage type, damage degree, compression stiffness, shear stiffness and axial pressure are used as output. The enhanced YOLO-v5s model is trained to construct a bearing status detection model.

5. The bridge early warning method based on support status according to claim 1 is characterized in that: The enhanced YOLO-v5s model is obtained by sequentially adding a Squeeze-and-Excitation module and a multi-head attention module after the neck network of the traditional pre-trained YOLO-v5s model, and replacing the detection head with a classification head or a regression head.

6. The bridge early warning method based on support status according to claim 1 is characterized in that: The loss function of the enhanced YOLO-v5s model includes: Among them, L total represents the total loss function, λ MMD is a hyperparameter used to balance the weight of MMD loss relative to cross entropy loss, L MMD is the maximum mean difference loss, is the loss function on the source domain, is the loss function on the target domain.

7. The bridge early warning method based on support status according to claim 1 is characterized in that: In S5, the shear stiffness change rate, the compression stiffness change rate, and the bearing axial pressure index are obtained according to the state of the target bridge bearing, the warning level of the target bridge is determined and a warning is issued, and damage information of the bearing and the bridge is output at the same time, including: According to the shear stiffness, the compression stiffness and the axial pressure, obtaining a shear stiffness change rate, a compression stiffness change rate and a support axial pressure index, wherein the support axial pressure index includes a support axial pressure deviation rate, a lateral imbalance coefficient and a support axial pressure variation coefficient; When the shear stiffness change rate and the compression stiffness change rate exceed the warning threshold, the warning level of the target bridge is determined and a warning is issued, and the damage type and damage degree output by the bearing state detection model are outputted; When the support axial pressure deviation rate, the lateral imbalance coefficient and the support axial pressure variation coefficient exceed the warning threshold, the damage type of the support and the bridge is judged according to the support axial pressure index, the warning level of the target bridge is judged and a warning is issued, and the damage type of the support and the bridge is output at the same time.

8. A bridge early warning system based on support status, characterized in that: The system applies the bridge early warning method based on support status according to any one of claims 1 to 7, and the system includes a data acquisition module, a preprocessing module, a model prediction module, and an early warning module: A data acquisition module, configured to acquire detection signals of target bridge supports through the active sensing experiment; A preprocessing module, used for preprocessing the detection signal of the target bridge support and converting it into a wavelet spectrum through continuous wavelet transform; A model prediction module is used to input the wavelet spectrum corresponding to the target bridge bearing detection signal into a bearing state detection model to obtain the state of the target bridge bearing; The early warning module is used to obtain the shear stiffness change rate, compression stiffness change rate and bearing axial pressure index according to the status of the target bridge bearing, determine the early warning level of the target bridge and issue an early warning, and output damage information of the bearing and the bridge at the same time.

Images