A bridge sliding support wear identification method based on longitudinal temperature-induced displacement amplitude monitoring of a main beam

Abstract

The present application belongs to the technical field of bridge structure infrastructure equipment component performance monitoring and evaluation, and discloses a bridge sliding support wear identification method based on longitudinal temperature-induced displacement amplitude monitoring of a main beam, which comprises the following steps: (1) sliding support wear identification index extraction based on monitoring data; (2) bridge sliding support wear identification method; and (3) bridge sliding support wear online identification. The present application establishes a multivariate temperature-displacement correlation model based on PSO-LSTM, realizes accurate prediction of support temperature-induced displacement, and eliminates the influence of temperature effect. According to the change characteristics of support temperature-induced displacement after sliding support wear, a sliding support wear identification index is proposed, and the identification of sliding support wear is realized. Finally, the feasibility and practicality of the present application are verified through the measured data of an in-service arch bridge. The present application has high engineering application value in the field of bridge sliding support wear identification based on monitoring data.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of bridge structure infrastructure equipment component performance monitoring and evaluation, and particularly relates to a bridge sliding support wear identification method based on longitudinal temperature-induced displacement amplitude monitoring of a main beam. BACKGROUND

[0002] The bridge support plays an important transmission role in the bridge system and is a key component connecting the upper structure and the lower structure of the bridge, and its health condition directly affects the use performance of the bridge system. Under the influence of vehicle load and environmental change, the characteristics of the support will change, even be damaged, resulting in the relative movement of the support being limited. The upper structure and the lower structure of the bridge will bear greater additional force, and then cause the cracking and damage of the beam body, shorten the service life of the bridge. Therefore, it is necessary to further study the support damage problem of the bridge under the action of external load based on monitoring data, so as to maintain and replace the support in time, and ensure the safe operation of the bridge.

[0003] During bridge operation, structural health monitoring systems can be used to assess bridge conditions in order to analyze and identify actual structural damage. Meanwhile, the assessment of bearing conditions has received increasing attention in recent years. Current research on bridge bearings based on monitoring data includes: Wu Guangming et al., who established a time-varying temperature-displacement nonlinear model (Early Warning Method for Bearing Displacement of Long-Span Bridges Using a Proposed Time-Varying Temperature-Displacement Model) based on the traditional linear temperature-displacement model. They then used the model's prediction error to provide early warning of abnormal bearing displacement. Huang Haibin et al. proposed a sparse Bayesian temperature-displacement relationship model (Sparse Bayesian Identification of Temperature-Displacement Model for Performance Assessment and Early Warning of Bridge Bearings) to predict bearing displacement caused by temperature effects probabilistically. However, these methods do not consider the time lag effect between temperature and bearing displacement. Wang Gaoxin et al. studied the cumulative displacement of bearings under dynamic loads and proposed a reliability index (Safety Evaluation of the Wear Life of High-Speed ​​Railway Bridge Bearings by Monitoring Train-Induced Dynamic Displacements) to assess the failure probability of the wear limit within the service life of the bearings. However, due to the low sampling frequency of the displacement sensor, it cannot accurately monitor the relatively high-frequency displacement components of vehicle loads and wind loads. Therefore, the obtained cumulative displacement of the bearings is too small, resulting in an inability to accurately assess the wear of the bearings. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying wear of bridge sliding bearings based on longitudinal temperature-induced displacement amplitude monitoring of the main beam. The specific technical solution is as follows:

[0005] Step 1. Extraction of sliding bearing wear identification indicators based on monitoring data

[0006] (1.1) Extraction of temperature-induced displacement data of bridge bearings. Wavelet packet decomposition was used to extract the temperature-induced displacement data of the bearings. Wavelet packet decomposition divides the temperature-induced displacement data of the bearings into different frequency bands in the frequency domain. Since the daily temperature variation frequency is 1.16 × 10⁻⁶... -5The frequency is Hz, which is much lower than the natural frequency of the bridge. By selecting the db3 wavelet basis function and setting the decomposition level to 5, the high-frequency displacement components caused by vehicle load and wind load can be completely eliminated.

[0007] (1.2) Extraction of multi-dimensional temperatures in bridge structures. This invention uses the effective temperature ET of bridge components and the temperature difference TD between components as structural temperature variables. The effective temperature is the average temperature distributed across the cross-section of the bridge component, defined as follows:

[0008]

[0009] In the formula, ET represents the effective temperature of the bridge component; A represents the cross-sectional area of ​​the bridge component; and T(x,y) represents the temperature distribution on the cross-section.

[0010] To simplify the calculation, the cross-section is divided into multiple sub-regions based on the sensor arrangement. Assuming that the temperature value is the same in each sub-region, the effective temperature can be simplified and calculated using the temperature values ​​of all sub-regions as follows:

[0011]

[0012] In the formula, A h Let h be the area of ​​the h-th region; q represents the measured temperature of the h-th region; q represents the total number of regions divided on the cross-section.

[0013] The temperature difference between components is the difference in the effective temperatures of different components. Finally, the structural temperature input variable T is obtained. * It is represented by the following formula:

[0014] T * =[ET1,…,ET] m ,TD1,…,TD m (3)

[0015] In the formula, TD represents the effective temperature difference between components; m represents the number of components.

[0016] Then, the structural temperature data is subjected to feature standardization to eliminate the influence of the dimensions of the sample data. The calculation formula is shown below:

[0017] T = (T * -μ) / s (4)

[0018] In the formula, T represents the normalized temperature data; T * is the original temperature data; μ is the mean of the temperature data; s is the standard deviation of the temperature data.

[0019] (1.3) Training of the multivariate temperature-displacement correlation model. Establishing a multivariate temperature-displacement correlation model involves establishing the input-output relationship between temperature and support temperature-induced displacement over time. This invention utilizes a Long Short-Term Memory (LSTM) neural network, which can explain the time dependence between variables and express highly nonlinearity, for model training.

[0020] LSTM neural networks are an improved type of recurrent neural network, comprising an input layer, LSTM hidden layers, fully connected layers, dropout layers, and an output layer. The LSTM neural network architecture in this invention consists of an input layer, two LSTM hidden layers, two dropout layers, a fully connected layer, and an output layer.

[0021] The parameters of an LSTM neural network include the number of hidden layer units, the dropout rate of the dropout layer, and the number of fully connected layer units. Changes in these parameters affect the model's prediction accuracy. To obtain the optimal parameters for an LSTM neural network, the Particle Swarm Optimization (PSO) algorithm is introduced. PSO is a biomimetic algorithm that seeks the optimal solution in the solution space using existing data. Its fundamental difference from other optimization algorithms is that it only requires the objective function and does not rely on the objective gradient or any differential form. Specifically, in a certain objective search space, there exists a particle swarm of size M, whose state at a given time is controlled by two vectors: position and velocity, as expressed by the following equation:

[0022]

[0023]

[0024]

[0025] In the formula, N i This represents the i-th individual in the particle swarm; This indicates the position of the particle at time t; The velocity of the particle at time t is represented by ; n represents the dimension of the space, i.e., the number of independent variables. This represents the position in the n-dimensional dimension at time t; This represents the velocity in the n-dimensional dimension at time t.

[0026] The position update of a particle is determined by three indices: inertial direction, individual optimal direction, and swarm optimal direction. The particle swarm position update formula is:

[0027]

[0028]

[0029] In the formula, Let be the velocity of the particle at time t+1; The position of the particle at time t+1; w is the inertia weight coefficient; r1 and r2 are random numbers between [0,1]; c1 is the self-learning factor; c2 is the group learning factor; p in For the individual's historical best position; p gn This is the best historical position for the population.

[0030] PSO-optimized LSTM neural networks continuously update the particle velocity and position, and use the root mean square error of the multivariate temperature-displacement correlation model as the fitness function of PSO to find the optimal position of the particles and the particle swarm. The optimal parameters of the LSTM neural network are shown in the table below.

[0031] Table 1. Optimal parameters of LSTM neural network

[0032]

[0033] Finally, the normalized temperature data T is input into the trained multivariate temperature-displacement correlation model to predict the temperature-induced displacement of the support.

[0034] (1.4) Establishing a wear identification index for sliding bearings. When a bearing wears, the increased sliding friction coefficient leads to a decrease in the amplitude of the temperature-induced displacement and an increase in the friction-overcoming stage. Using the measured amplitude of the temperature-induced displacement as the identification index is ineffective due to seasonal and weather variations. Therefore, this paper proposes using the difference between the amplitude predicted by the multivariate temperature-displacement correlation model and the measured amplitude as the bearing wear identification index, thus eliminating the influence of temperature effects. When the bearing wears, this index accurately reflects the change in the amplitude of the temperature-induced displacement. The calculation formula is as follows:

[0035] TDAE = D pa -D ma (10)

[0036] In the formula, TDAE represents the prediction error of the temperature-induced displacement amplitude of the support; D pa D represents the predicted amplitude of the multivariate temperature-displacement correlation model; ma This represents the measured amplitude of the temperature-induced displacement of the support.

[0037] Step 2. Method for Identifying Wear of Bridge Sliding Bearings

[0038] (2.1) Extract the wear identification index TDAE of the bridge sliding bearing under normal conditions according to step 1.

[0039] (2.2) When the support is in normal condition, the TDAE fluctuates within a small range due to the prediction error of the multivariate temperature-displacement correlation model, and has a stable mean and variance. When the support is worn, the amplitude of the measured temperature-induced displacement data of the support decreases, leading to an increase in TDAE. Therefore, statistical analysis of TDAE can determine whether the support is worn.

[0040] In the field of structural health monitoring, Shewhart control charts are widely used for structural damage monitoring. For bearings, when slight wear occurs, the TDAE (Total Scale Effect) will show a small shift. This small shift may occur below the warning line on the Shewhart control chart, making it ineffective in identifying bearing wear. Therefore, a Cumulative Sum (CUSUM) control chart is used to identify the abnormal TDAE, thereby enabling bridge bearing wear monitoring. Based on the direction of data shift, the CUSUM control chart can define positive and negative cumulative processes. As the bearing gradually wears, the TDAE will only gradually increase, so a one-sided CUSUM control chart reflecting positive shift is sufficient to accurately and effectively monitor bearing changes. When the bearing is in a normal state, the TDAE follows a normal distribution: TDAE ~ N(μ0, σ). 2 The CUSUM control chart with a unilateral positive offset is represented as follows:

[0041]

[0042] In the formula, S L is the cumulative sum of TDAE; k is the tolerance value for TDAE fluctuation; l is the number of indicators.

[0043] When the TDAE offset is less than k, it will show a decreasing trend, but will not be less than 0. When the TDAE offset is greater than k, it will show an increasing trend. When an alarm occurs, it indicates that the TDAE is gradually increasing and the sliding support is wearing. The tolerance value k and the alarm line AL are important parameters of the CUSUM control chart. Usually, the tolerance value k and the alarm line AL of the CUSUM control chart are set to 0.5σ and 4σ respectively to achieve anomaly identification when data offset occurs. σ is the standard deviation of the data collected when the support is intact.

[0044] Step 3. Online Identification of Wear on Bridge Sliding Bearings

[0045] (3.1) During the online monitoring phase, the monitoring data is extracted according to steps (1.1) and (1.2) in step 1, and input into the model trained in steps (1.3) and (1.4) to obtain the bearing wear identification index TDAE. Then, the TDAE is input into the CUSUM control chart. When the cumulative sum S L When the alarm line AL is exceeded, the wear of the support is identified.

[0046] The beneficial effects of this invention are:

[0047] 1. The bridge sliding bearing wear identification method based on temperature-induced displacement amplitude prediction error established in this invention can realize online evaluation of the bearing's real-time condition. Furthermore, this method has strong universality and can be applied to various types of sliding bearings.

[0048] 2. The bridge bearing wear identification index proposed in this invention can accurately reflect the temperature-induced displacement change characteristics when the bearing is worn, and can be combined with the CUSUM control chart to identify the wear of sliding bearings. Attached Figure Description

[0049] Figure 1 This is a flowchart of the method of the present invention;

[0050] Figure 2 This is a diagram of the temperature-induced displacement residuals of the supports.

[0051] Figure 3 A chart showing the indicators for bearing wear identification;

[0052] Figure 4 The following are images showing bearing wear identification: (a) 5% bearing degradation; (b) 10% bearing degradation; (c) 15% bearing degradation; (d) 20% bearing degradation. Detailed Implementation

[0053] The present invention will now be described in further detail with reference to the accompanying drawings and a calculation example.

[0054] The method for identifying wear of bridge sliding bearings according to the present invention consists of three steps: "extraction of wear identification indicators for sliding bearings based on monitoring data", "bridge sliding bearing wear identification method", and "online identification of bridge sliding bearing wear". The implementation process of the method is as follows: Figure 1 As shown.

[0055] In a specific numerical example, temperature and support displacement monitoring data from a Chinese in-service arch bridge from February to October 2013 were used for verification. First, the support displacement data was extracted using the temperature-induced displacement data of the supports according to step (1.1). Then, according to step (1.2), the structural temperature variable was calculated and extracted, and the first 80% of the support temperature-induced displacement and temperature datasets were used as the training set, and the last 20% as the test set. Finally, according to step (1.3), a multivariate temperature-displacement correlation model based on PSO-LSTM was trained using the training set data. Figure 2 The figure shows the residual between the predicted and measured values ​​of the temperature-induced displacement of the support. It can be seen from the figure that the predicted values ​​of the model are in good agreement with the measured values ​​and the prediction accuracy is high. Figure 3 The figure shows that the bearing wear identification index TDAE extracted according to step (1.4) does not have obvious discreteness or offset phenomenon, which further verifies the effectiveness of the model.

[0056] According to step 2, a bridge bearing wear identification model based on PSO-LSTM is established, where the obtained TDAE reflects that the bridge bearing is in a normal state, and the tolerance value k and alarm line AL of the CUSUM control chart are set according to step (2.2).

[0057] According to step 3, online identification of bridge sliding bearing wear is achieved. To verify the effectiveness of the bearing wear identification method proposed in this invention, the measured data of bearing temperature-induced displacement are reduced by 5%, 10%, 15%, and 20% respectively to simulate different working conditions when bearing wear occurs. The TDAE under bearing wear conditions is calculated according to step (1.4), and then the TDAE is input into the CUSUM control chart. When the obtained cumulative sum S... L When the alarm line AL is exceeded, it indicates that the support has worn out. Figure 4 This is a diagram for identifying bridge bearing wear. From the diagram, it can be concluded that when the bearing is in normal condition, the cumulative wear and tear (S)... L There will be no fluctuations, and the alarm line AL will not be exceeded. Support wear identification begins first... Figure 4 (b) Triggered. For a 10% reduction in bearing temperature-induced displacement, the alarm delay is 9 data points. This result shows that the identification method can detect a minimum bearing degradation of 10%. As the degree of bearing wear increases, the magnitude of the reduction in bearing temperature-induced displacement is greater, and the alarm delay is lower. For a 15% reduction in bearing temperature-induced displacement, all abnormal indicators after the 4th data point exceed the alarm line. For a 20% reduction in bearing temperature-induced displacement, all abnormal indicators after the 3rd data point exceed the alarm line. This phenomenon indicates that the identification method can perfectly identify degradation of more than 15% in bridge sliding bearings. In summary, the bridge sliding bearing wear identification method proposed in this invention can realize online real-time monitoring and damage identification of bearing wear based on monitoring data, verifying the feasibility and practicality of the method.

Claims

1. A method for identifying wear of bridge sliding bearings based on longitudinal temperature-induced displacement amplitude monitoring of the main girder, characterized in that, The steps are as follows: Step 1. Extraction of sliding bearing wear identification indicators based on monitoring data (1.1) Extraction of temperature-induced displacement data of bridge bearings: The temperature-induced displacement data of bearings is extracted by wavelet packet decomposition method; the wavelet packet decomposition method divides the temperature-induced displacement data of bearings into different frequency bands in the frequency domain, and the db3 wavelet basis function is selected, with the number of decomposition layers set to 5. (1.2) Extraction of multi-dimensional temperature of bridge structure: based on the effective temperature of bridge components and temperature difference between components As a structural temperature variable; where the effective temperature is the average temperature distributed across the cross-section of the bridge component, defined as follows: ； In the formula, Indicates the effective temperature of bridge components; Represents the cross-sectional area of ​​bridge components; This represents the temperature distribution across a cross-section; To simplify the calculation, the cross-section is divided into multiple sub-regions based on the sensor arrangement, and it is assumed that the temperature value in each sub-region is the same. Therefore, the effective temperature is simplified and calculated using the temperature values ​​of all sub-regions as follows: ； In the formula, For the first The area of ​​each sub-region; For the first Temperature measurement in each sub-region; This represents the total number of sub-regions divided on the cross-section; The temperature difference between components is the difference in the effective temperature of different components, which is used to obtain the structural temperature input variable. It is expressed by the following formula: ； In the formula, The temperature difference between components; Expressed as the number of components; Then, the structural temperature data is subjected to feature standardization to eliminate the influence of the dimensions of the sample data. The calculation formula is shown below: ； In the formula, The normalized structural temperature data; This is the original structural temperature data; This represents the average of the structural temperature data; The standard deviation of the structural temperature data; (1.3) Training of the multivariate temperature-displacement correlation model: Establishing a multivariate temperature-displacement correlation model means establishing the input-output relationship between temperature and support temperature-induced displacement over time; the model is trained using a long short-term memory (LSTM) neural network. The architecture of an LSTM neural network consists of an input layer, two LSTM hidden layers, two dropout layers, a fully connected layer, and an output layer. The parameters of an LSTM neural network include the number of hidden layer units, the dropout rate of the dropout layer, and the number of fully connected layer units. To obtain the optimal parameters of the LSTM neural network, the Particle Swarm Optimization (PSO) algorithm is introduced. Specifically, in a certain target search space set, there exists a particle swarm of size M, and the state of the particle at a certain moment is controlled by two vectors, position and velocity, as expressed by the following formula: ； ； ； In the formula, Represents the first particle in the swarm. Individual; Indicates that the particle is in The position at that moment; Indicates that the particle is in The speed of time; The dimension of the space represents the number of independent variables. express Always Position in dimensions; express Always Velocity in dimensions; The position update of a particle is determined by three indices: inertial direction, individual optimal direction, and swarm optimal direction; the particle swarm position update formula is: ； ； In the formula, For particles in The speed of time; For particles in The position at that moment; This is the inertia weighting coefficient; , for Random numbers between; As a self-learning factor; For group learning factors; For the individual's best historical position; This is the best historical position for the population; PSO optimizes the LSTM neural network by continuously updating the particle velocity and position, and using the root mean square error of the multivariate temperature-displacement correlation model as the fitness function of PSO to find the optimal position of the particle and the particle swarm. The optimal parameters of the LSTM neural network are shown in the table below. Table 1. Optimal parameters of LSTM neural network ； Finally, the normalized structural temperature data were analyzed. Inputting the data into a trained multivariate temperature-displacement correlation model enables the prediction of temperature-induced displacement of supports. (1.4) Establishment of wear identification index for sliding bearings: The difference between the amplitude predicted by the multivariate temperature-displacement correlation model and the amplitude of the measured data is proposed as the identification index for bearing wear, eliminating the influence of temperature effect; when the bearing wears, it can reflect the change in the amplitude of temperature-induced displacement of the bearing, and the calculation formula is as follows: ； In the formula, This indicates the error in predicting the amplitude of temperature-induced displacement of the support; This represents the predicted amplitude of the multivariate temperature-displacement correlation model; This indicates the measured amplitude of the temperature-induced displacement of the support. Step 2. Method for Identifying Wear of Bridge Sliding Bearings (2.1) Extract wear identification indicators of bridge sliding bearings under normal conditions according to step 1. ; (2.2) When the support is in normal condition, Because the prediction error of the multivariate temperature-displacement correlation model fluctuates within a small range and has a stable mean and variance; when wear occurs in the support, the amplitude of the measured temperature-induced displacement data of the support decreases, leading to... Increase; therefore, for Statistical analysis can determine whether the support is worn. Wear identification indicators are identified using cumulative and CUSUM control charts. Identification is performed to monitor the wear of bridge sliding bearings; based on the data offset direction, the CUSUM control chart defines positive and negative accumulation processes; as the bearing gradually wears... It will only gradually increase, so the changes in the sliding support can be accurately and effectively monitored simply by using a one-sided CUSUM control chart that reflects the positive offset; when the support is in a normal state, Follows a normal distribution ~ A CUSUM control chart with a unilateral positive offset is represented as follows: ； In the formula, for The cumulative sum; for Tolerance for fluctuations; The number of indicators; when The offset is less than When it is low, it will show a downward trend, but it will not be less than 0; when The offset is greater than When the time is right, it will show an upward trend; when an alarm is triggered, it indicates... As the tolerance value gradually increases, wear appears on the sliding support; With alarm line These are important parameters of the CUSUM control chart; the tolerance value of the CUSUM control chart. and alarm line Set as follows and This means enabling anomaly detection when data shifts; among which, The standard deviation of the data collected when the support is intact; Step 3. Online Identification of Wear on Bridge Sliding Bearings (3.1) During the online monitoring phase, the monitoring data is extracted according to steps (1.1) and (1.2) in step 1, and then input into the model trained in steps (1.3) and (1.4) to obtain the bearing wear identification index. Then When inputting into the CUSUM control chart, when the cumulative sum Exceeding the set alarm line This allows for the identification of wear on the sliding support.

Images