A cable-stayed bridge damage identification method based on continuous Bayesian networks
By using a continuous Bayesian network-based method to infer cable damage using the rate of change of cable force, the problems of blind spots, high cost, and large errors in existing technologies are solved, and efficient and accurate cable damage identification is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- FUZHOU UNIV
- Filing Date
- 2022-11-13
- Publication Date
- 2026-06-30
Smart Images
Figure QLYQS_1 
Figure QLYQS_2 
Figure BDA0003940251020000021
Abstract
Description
Technical Field
[0001] This invention relates to the field of bridge health monitoring technology, and in particular to a method for identifying cable-stayed bridge damage based on a continuous Bayesian network. Background Technology
[0002] As a crucial component of cable-stayed bridges, stay cables connect the main girder and the bridge towers, and their stress performance has a significant impact on the overall lifespan of the bridge. The natural environment and load conditions in which bridges operate are often complex and variable, making stay cables susceptible to performance degradation due to corrosion and fatigue, leading to damage such as cracks and broken wires, thus affecting the bridge's operational safety. Therefore, detecting and assessing damage to stay cables through monitoring and inspection technologies is of significant theoretical and practical importance for developing subsequent maintenance plans.
[0003] Damage identification in cable-stayed bridges can employ both manual inspection and theoretical analysis. Manual methods utilize visual inspection and testing instruments (such as acoustic emission, magnetic leakage detection, X-ray inspection, and cable-climbing robots) to identify cable defects and damage. Theoretical analysis, based on cable vibration data such as acceleration time history, modal frequencies, and modal strain energy, combines neural networks and wavelet analysis to indirectly assess cable damage. However, both methods currently have limitations. For instance, manual inspection has blind spots and insufficient accuracy in detecting anchorage areas prone to problems. Furthermore, the method of checking each area individually is time-consuming, labor-intensive, costly, and lacks real-time capability. Theoretical methods can rely on the continuous data provided by bridge monitoring systems for real-time damage assessment; however, limitations in the number of sensors deployed, as well as the influence of test noise, environmental factors, and coupled vibrations, can lead to significant errors in the assessment results. Incomplete test data and uncertainties remain ongoing challenges that need to be addressed. Summary of the Invention
[0004] In view of this, the purpose of this invention is to provide a cable damage identification method based on a continuous Bayesian network, which can be used in practice when cable monitoring data is incomplete, and to infer the cable damage situation through the cable force change rate.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: a cable-stayed bridge damage identification method based on continuous Bayesian networks, comprising the following steps:
[0006] Step 1: Establish a cable damage sample library; use finite element simulation combined with random sampling to establish a damage sample library for cable-stayed bridge cables;
[0007] Step 2: Network topology and parameter learning; The CBN topology consists of network nodes and directed arcs between nodes. The directed arcs describe the logical causal relationships between cables, and the strength of the relationship is represented by the conditional probability density function; The classic K2 algorithm is used to learn the cable topology, and the maximum number of parent nodes required by the algorithm is determined according to the laws of cable damage mechanics.
[0008] Step 3: Network reasoning and damage identification; using the rate of change of cable tension of a certain cable or part of the cable as observational evidence, inputting it into CBN, and reasoning the posterior probability density distribution of other cables under the damage condition at time t.
[0009] In a preferred embodiment, step 1 specifically includes:
[0010] Step 11: Select the uncertainty parameters of the cable, and obtain samples with different combinations of parameters through random sampling. Each sample represents a damage condition of the cable.
[0011] Step 12: Using the cable tension change rate as the damage index, input the damage condition into the finite element model of the cable-stayed bridge and calculate the cable tension change rate before and after the damage:
[0012]
[0013] In the formula C_f i t Let f be the rate of change of cable force on cable i at time t. i t f i 0 These represent the cable force values at time t and in the initial state, respectively, for cable i.
[0014] Step 13: Divide the damage state range of the cable according to the rate of change of cable force.
[0015] In a preferred embodiment, step 2 specifically includes:
[0016] Step 21: Define the maximum number of parent nodes for the stay cables and the order of node variables; combine the mechanical relationship between the cables and the cable damage simulation to determine the maximum number of parent nodes for the network node variables, that is, the number of adjacent cables exceeding the threshold is taken as the maximum number of parent nodes;
[0017] Step 22: Define network nodes as continuous variables; discretize the continuous variables using the K-means clustering method to process them as discrete variables: first determine the number of clusters, then train the dataset and determine the data centers and sort them, then set the moving average with the maximum and minimum values as boundaries, and finally sort the dataset again according to the training results, merging the records of the same cluster into the same group.
[0018] Step 23: Based on the set maximum number of parent nodes and node variable order, and combined with the previously obtained damage sample library, calculate the score of each node using the K2 algorithm, define parent and child nodes according to the score size, and form a CBN topology; during this process, use a parameter learning algorithm to learn from the samples in the damage sample library to obtain the conditional probability density between nodes; the CBN for cable damage inference is now established.
[0019] In a preferred embodiment, step 3 specifically includes:
[0020] Step 31: Calculate the mean and variance of the rate of change of initial cable force for any cable i in the damage sample library;
[0021] Step 32: Input a portion of the cable tension change rate as evidence into the CBN, infer the posterior probability density distribution of the cable tension change rate, and obtain the mean cable tension change rate at time t;
[0022] Step 33: Define the relative error between the mean rate of change of cable force at time t and the mean rate of change of cable force at the initial time as an index of cable damage degree.
[0023] Compared with the prior art, the present invention has the following advantages: The present invention proposes a cable-stayed bridge damage identification method based on a continuous Bayesian network, which can be used in practice when cable-stayed bridge monitoring data is incomplete, and the cable damage status can be judged by inferring the cable force change rate. The advantages of the method are: (1) The continuous Bayesian network topology learning process considers the mechanical characteristics of the cable-stayed bridge, simplifies the determination of the maximum number of parent nodes, and greatly reduces the complexity of the network topology. (2) Only the test data of a single cable or part of the cable is needed to infer the damage status of the remaining cable, avoiding the problem of incomplete monitoring and the inability to evaluate due to missing sensor data in engineering practice. Detailed Implementation
[0024] The present invention will be further described below with reference to the embodiments.
[0025] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.
[0026] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application; as used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise; furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.
[0027] The cable-stayed bridge damage identification method based on continuous Bayesian networks includes three steps: establishing a cable damage sample library, learning network topology and parameters, and network inference and damage identification.
[0028] Step 1: Establish a cable damage sample library
[0029] The establishment of a CBN (Cable-Stayed Bridge Network) relies on samples, which can be obtained from numerical simulations or measured data in actual engineering projects. Considering that it is impossible to simulate the failure of a cable-stayed bridge cable in practice, and that numerical simulation can obtain complete damage conditions, a finite element simulation combined with random sampling method can be used to establish a damage sample library for cable-stayed bridge cables.
[0030] Step 11: Select cable uncertainty parameters, such as cable geometry and cable material properties, and obtain samples with different combinations of parameters through random sampling. Each sample represents a damage condition of the cable.
[0031] Step 12: Using the cable force change rate as the damage index, input the damage condition into the finite element model of the cable-stayed bridge and calculate the cable force change rate before and after the damage.
[0032]
[0033] In the formula C_f i t Let f be the rate of change of cable force on cable i at time t. i t f i 0 These represent the cable force values at time t and in the initial state, respectively, for cable i.
[0034] Step 13: Divide the damage state range of the cable according to the rate of change of cable force.
[0035] Step 2: Network Topology and Parameter Learning
[0036] The CBN topology consists of network nodes and directed arcs between them. These directed arcs describe the logical causal relationships between cables, and the strength of these relationships is represented by a conditional probability density function. The classic K2 algorithm is used to learn the cable-stayed bridge topology, while simultaneously determining the maximum number of parent nodes required by the algorithm based on the mechanical properties of cable damage.
[0037] Step 21: Define the maximum number of parent nodes and the order of node variables for the stay cables. Considering that the parent node in the topology is equivalent to the cause of a damage event, and the child nodes are equivalent to the result of the event, for example, finite element simulations may show that when the damage level of a stay cable reaches a certain threshold, the rate of change of cable force in several adjacent cables may also exceed the threshold. This threshold can be determined based on the average rate of change of cable force in other cables at the corresponding damage level. Therefore, the maximum number of parent nodes for network node variables can be determined by combining the mechanical relationships between cables and cable damage simulations; that is, the number of adjacent cables exceeding the threshold is taken as the maximum number of parent nodes.
[0038] Step 22: Define network nodes as continuous variables, but to simplify the solution process, discretize the continuous variables using the K-means clustering method and process them as discrete variables: First, determine the number of clusters, then train the dataset and determine the data centers and sort them, then set the moving average with the maximum and minimum values as boundaries, and finally sort the dataset again according to the training results, merging the records of the same cluster into the same group.
[0039] Step 23: Based on the set maximum number of parent nodes and node variable order, and combined with the previously obtained damage sample database, calculate the score of each node using the K2 algorithm. Define parent and child nodes according to the score, forming a CBN topology. During this process, a parameter learning algorithm (such as maximum likelihood estimation) is used to learn from the samples in the damage sample database to obtain the conditional probability density between nodes. At this point, the CBN for cable damage inference is complete.
[0040] Step 3: Network Reasoning and Damage Recognition
[0041] Using the rate of change of tension in one or more cables as observational evidence, and inputting it into the CBN (Contractual Risk Network), the posterior probability density distribution of other cables under the damage condition at time t is inferred. Since the conditional variance obtained from the CBN inference is usually independent of the parent state, the mean of the posterior probability density distribution can be analyzed.
[0042] Step 31: Calculate the mean and variance of the rate of change of initial cable force for any cable i in the damage sample library.
[0043] Step 32: Input a portion of the cable tension change rate as evidence into the CBN, infer the posterior probability density distribution of the cable tension change rate, and obtain the mean cable tension change rate at time t.
[0044] Step 33: Define the relative error between the mean rate of change of cable force at time t and the mean rate of change of cable force at the initial time as an index of cable damage degree.
[0045] The usage process of this invention is as follows:
[0046] Step 1: Select the cable uncertainty parameters, obtain the cable damage condition samples by random sampling, and input them one by one into the finite element model of the cable-stayed bridge to calculate the cable force change rate before and after the damage. Then, divide the cable damage state interval according to the cable force change rate.
[0047] Step 2: Combine the mechanical relationship between cables and the cable damage simulation to determine the maximum number of parent nodes and the order of node variables in the network. Then, the continuous variables of the nodes are discretized by clustering. Finally, the parent and child nodes of the network are defined by combining the damage sample library and the K2 algorithm, and the conditional probability density between nodes is learned. Finally, the CBN is established.
[0048] Finally, the conditional probability density function corresponding to each node is obtained through parameter learning.
[0049] Step 3: Using the rate of change of cable tension of a certain cable or part of the cable as observational evidence, input CBN to infer the posterior probability density distribution of other cables under the damage condition at time t, and use the relative error between the mean of the probability distribution and the initial value to evaluate the cable damage.
Claims
1. A method for cable damage detection based on continuous Bayesian networks, characterized in that, Includes the following steps: Step 1: Establish a cable damage sample library; use finite element simulation combined with random sampling to establish a damage sample library for cable-stayed bridge cables; Step 2: Network topology and parameter learning; The CBN topology consists of network nodes and directed arcs between nodes. The directed arcs describe the logical causal relationships between cables, and the strength of the relationship is represented by the conditional probability density function; The classic K2 algorithm is used to learn the cable topology, and the maximum number of parent nodes required by the algorithm is determined according to the laws of cable damage mechanics. Step 3: Network reasoning and damage identification; using the rate of change of cable tension of a certain cable or part of the cable as observational evidence, inputting it into CBN, and reasoning the posterior probability density distribution of other cables under the damage condition at time t; Step 1 specifically includes: Step 11: Select the uncertainty parameters of the cable, and obtain samples with different combinations of parameters through random sampling. Each sample represents a damage condition of the cable. Step 12: Using the cable tension change rate as the damage index, input the damage condition into the finite element model of the cable-stayed bridge and calculate the cable tension change rate before and after the damage: （1） In the formula is the cable force variation rate of the i-th cable at time t, , is the cable force value of the i-th cable at time t and initial state, respectively. Step 13: Divide the cable damage state range according to the cable tension change rate; Step 2 specifically includes: Step 21: Define the maximum number of parent nodes for the stay cables and the order of node variables; combine the mechanical relationship between the cables and the cable damage simulation to determine the maximum number of parent nodes for the network node variables, that is, the number of adjacent cables exceeding the threshold is taken as the maximum number of parent nodes; Step 22: Define network nodes as continuous variables; discretize the continuous variables using the K-means clustering method to process them as discrete variables: first determine the number of clusters, then train the dataset and determine the data centers and sort them, then set the moving average with the maximum and minimum values as boundaries, and finally sort the dataset again according to the training results, merging the records of the same cluster into the same group. Step 23: Based on the set maximum number of parent nodes and node variable order, and combined with the previously obtained damage sample library, calculate the score of each node using the K2 algorithm, define parent and child nodes according to the score size, and form a CBN topology; during this process, use a parameter learning algorithm to learn from the samples in the damage sample library to obtain the conditional probability density between nodes; the CBN for cable damage inference is now established.
2. The cable-stayed bridge damage identification method based on continuous Bayesian networks according to claim 1, characterized in that, Step 3 specifically includes: Step 31: Calculate the mean and variance of the rate of change of initial cable force for any cable i in the damage sample library; Step 32: Input a portion of the cable tension change rate as evidence into the CBN, infer the posterior probability density distribution of the cable tension change rate, and obtain the mean cable tension change rate at time t; Step 33: Define the relative error between the mean rate of change of cable force at time t and the mean rate of change of cable force at the initial time as an index of cable damage degree.