CN111967185B - Cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling - Google Patents
Cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling Download PDFInfo
- Publication number
- CN111967185B CN111967185B CN202010794104.XA CN202010794104A CN111967185B CN 111967185 B CN111967185 B CN 111967185B CN 202010794104 A CN202010794104 A CN 202010794104A CN 111967185 B CN111967185 B CN 111967185B
- Authority
- CN
- China
- Prior art keywords
- cable
- cable force
- distribution
- vertical displacement
- main beam
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Geometry (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Architecture (AREA)
- Civil Engineering (AREA)
- Structural Engineering (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling, which comprises the steps of carrying out trend item removing treatment on cable force and main beam displacement monitoring, then carrying out normalization, removing data points within a standard deviation and obtaining the probability distribution of the car induced effect; building an unsupervised image transformation model, and assembling into a distribution matrix as network input; selecting training set data to input into an unsupervised image transformation model for training, using a loss function as a comprehensive function constructed according to network characteristics, and using an optimization algorithm as an ADAM algorithm; and inputting the vertical displacement distribution of the main beam with concentrated test into a trained unsupervised image transformation model, calculating the EMD distance between the prediction and the real cable force distribution, and using the EMD distance as an evaluation index of the change of the bridge state to evaluate whether the bridge state is changed. The method can realize intelligent evaluation of the state of the cable-stayed bridge based on the distribution correlation modeling under the condition that neither a bridge structure system nor load space-time information is known.
Description
Technical Field
The invention relates to the field of bridge engineering, in particular to a cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling.
Background
With the rapid development of national economy, China is facing the climax of the construction of large-span bridges. The cable-stayed bridge is particularly favored by design and selection due to the large span and beautiful appearance. The main key stressed member of the cable-stayed bridge is a cable, and the cable is easy to vibrate back and forth under the action of wind, rain and alternating loads of vehicles due to small mass, small damping and large flexibility, so that fatigue damage is caused, and further breakage is possible. In addition, the cable-stayed bridge inevitably suffers from the coupling action of complex factors such as environmental erosion, material aging and the like in the service period of the whole life, and the gradual damage is initiated, developed and accumulated, so that the resistance of the structure is attenuated, and even catastrophic accidents occur under extreme conditions.
In order to ensure the safe service of the cable-stayed bridge, a bridge management department can invest a large amount of financial resources to install a structural health monitoring system on the cable-stayed bridge, monitor structural responses of various types of bridges for a long time and evaluate the state of the bridge according to monitoring data. The structural health monitoring system can monitor different bridge structural responses (such as stay cable force, girder vertical displacement and the like) and accumulate massive structural health monitoring big data in an operation period. At present, a plurality of bridge structure health monitoring and state evaluation methods based on data driving exist. However, some of these methods are directed to finite element simulation or test data under laboratory conditions, and cannot be really effectively applied to the actual large-span cable-stayed bridge because: on one hand, the actual large-span cable-stayed bridge has a very complex structure, and the numerical simulation or the laboratory model test is greatly simplified; on the other hand, the site environment of the large-span cable-stayed bridge is very complex, the structural response data is obviously influenced by the temperature action, and the magnitude of the actual vehicle, wind and other dynamic loads during operation is difficult to accurately obtain. Furthermore, the structural response relationship modeling method based on the traditional mechanical analysis needs a known bridge impulse response function or frequency response function and accurate load size and position information. Therefore, the traditional mechanical analysis method is difficult to be applied to the actual bridge structure. In recent years, researchers have proposed some bridge state evaluation methods based on single type of monitoring data, but the evaluation effect of such methods is highly dependent on the quality of the monitoring data. Considering the actual situation of the bridge structure health monitoring system, sensor faults and data anomalies are the situations which are necessarily faced in the actual monitoring data. Therefore, the bridge state evaluation method based on single type monitoring data is extremely easy to be restricted by data quality: when the sensor data of some channels are lost or abnormal, the evaluation result is often inaccurate. The above factors all bring great difficulty to the state evaluation of the cable-stayed bridge based on the health monitoring data of the measured structure.
The stay cable is used as a key component of the cable-stayed bridge, and the dead weight of the main beam, the vehicle load and the like are transmitted to the bridge tower, so that an extremely important connection effect is achieved; the vertical displacement of the main beam is an important index for directly measuring the state of the bridge during the construction and operation of the bridge. At present, bridge health monitoring data mining research performed by scholars at home and abroad is usually performed on a single variable, and most of the research is performed from the perspective of continuous time-course data analysis, and the research on relationship modeling between two or more variable clusters is very little, and particularly, the research performed from the perspective of data distribution is almost none: even in the case of partial data or abnormal sensor state, the overall distribution information of the partial time-course information of the monitoring data can still be kept basically unchanged although the partial time-course information is lost. Therefore, the distribution correlation of two important structural response monitoring variables, namely the cable tension and the vertical displacement of the main beam, is modeled, and the overall and local states of the cable-stayed bridge can be reflected better. If the distribution correlation of the two important structural response monitoring variables changes, the bridge state is possible to change, and the two important structural response monitoring variables correspond to different modes. Therefore, how to establish a cable-stayed bridge state evaluation method based on distribution correlation modeling by using monitoring data of cable force of a cable and vertical displacement of a main beam under the condition that neither a bridge structure system nor load space-time information is known, an evaluation index of state change of the cable-stayed bridge is provided, and an inherent incentive is excavated, so that a decision basis of bridge management and traffic control is provided for whether maintenance is needed or not for the large-span cable-stayed bridge due to the state change, wherein the decision basis includes but is not limited to abnormal events such as bridge health monitoring system upgrade, sensor abnormity, cable breakage, earthquake, strong wind over-limit vibration, traffic jam, overload, ship collision and the like.
Disclosure of Invention
Based on the defects, the invention aims to disclose a cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling, which is characterized in that under the condition that a bridge structure system and load space-time information are unknown, the cable-stayed bridge state evaluation method based on distribution correlation modeling is established by utilizing monitoring data of cable force of a cable and vertical displacement of a main beam, evaluation indexes of cable-stayed bridge state change are provided, internal inducement is mined, the maintenance and management decision problem of an actual large-span cable-stayed bridge under the condition of facing abnormal events is solved, a basis is provided for maintenance decision and traffic control of the actual large-span cable-stayed bridge due to state change, and the intelligent cable-stayed bridge state evaluation based on distribution correlation modeling can be realized under the condition that the bridge structure system and the load space-time information are unknown.
A cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling comprises the following steps:
the method comprises the following steps of firstly, preprocessing the cable force of a stay cable and the displacement of a main beam: firstly, trend removing item processing is carried out, temperature influence is removed according to data characteristics, and response of cable force of a vehicle-induced cable and vertical displacement of a main beam is obtained; then carrying out normalization processing on the data, removing data points within 0.3 times of standard deviation, and obtaining the probability distribution of the cable force of the vehicle-induced stay cable and the vertical displacement of the main beam;
step two, building an unsupervised image transformation model, selecting the distribution of the guy cable force and the main beam vertical displacement data of all available channels, assembling the guy cable force and the main beam vertical displacement distribution matrixes according to rows, and respectively using the guy cable force and the main beam vertical displacement distribution matrixes as two inputs of a network: a main beam vertical displacement distribution matrix and a inhaul cable force distribution matrix;
selecting a part of the data processed in the step one to form a training set, inputting the training set into an unsupervised image transformation model for training, wherein a loss function used in the training process is a comprehensive function constructed according to network characteristics, an optimization algorithm is an ADAM algorithm, and default hyper-parameter setting is used;
and step four, inputting the vertical displacement distribution of the main beam concentrated in the test into a trained unsupervised image transformation model to obtain the predicted guy cable force distribution as output, calculating the EMD (Earth Mover's Distance) between the predicted guy cable force distribution and the real guy cable force distribution, and evaluating whether the bridge state changes and a corresponding action mode by taking the EMD between the predicted guy cable force distribution and the real guy cable force distribution as an evaluation index of the bridge state change.
The invention also has the following features:
1. the first step specifically comprises:
step one, adopting the same preprocessing flow for the stay cable force and the vertical displacement time course data of the main beam, taking 30 minutes as a time window, linearly normalizing all data points in the time window to be in a range of [0,1], then drawing a frequency histogram of the section of data, taking the position of a frequency peak value as a center, respectively taking 0.03 from the left and the right, and selecting an initial data extreme point falling in the frequency range;
performing median filtering, linear interpolation and smoothing on extreme points selected one by one in the first step, taking 3600 in a smooth window to obtain a temperature trend term, and subtracting the temperature trend term from original data to obtain a preprocessed vehicle-induced vertical displacement response and a preprocessed stay cable force;
and step three, respectively carrying out normalization processing on the vehicle-induced vertical displacement response and the stay cable force data of each channel, keeping the numerical value of each channel in the range of [ -1,1], removing noise data points within 0.3 times of standard deviation, and obtaining the probability distribution of the vehicle-induced stay cable force and the vertical displacement of the main beam.
2. In the second step, when the model is transformed without supervision, all available 27 main beam vertical displacement and 139 cable force channels are selected, and data distribution of each channel is integrated into a matrix according to rows and then used as two inputs of the network: the vertical displacement distribution matrix of the girder and the cable force distribution matrix of the stay cables.
The method for building the unsupervised image transformation model mainly comprises two modules, and the second step specifically comprises the following steps:
step two, building a variational self-encoder module: for the vertical displacement of the main beam, taking a probability distribution matrix of the vertical displacement of the main beam as input, wherein an encoder consists of 4 convolutional layers, 4 residual connecting layers and 2 shared convolutional layers, the input becomes an implicit characteristic vector after passing through the encoder and then enters a decoder, and the decoder consists of 2 shared convolutional layers, 4 residual connecting layers and 4 convolutional layers; similarly, for the cable force, the cable force probability distribution matrix is used as input, the encoder is composed of 4 convolution layers, 4 residual error connecting layers and 2 shared convolution layers, the input is converted into an implicit characteristic vector through the encoder and then enters the decoder, and the decoder is composed of 2 shared convolution layers, 4 residual error connecting layers and 4 convolution layers. All the shared convolution layers and the implicit characteristic vectors in the variational self-encoder share parameters for the vertical displacement of the main beam and the cable force of the inhaul cable.
Step two, building a generation confrontation network module: for the vertical displacement of the main beam, a decoder of a variational self-encoder is used as a generator, and a discriminator consists of 8 convolution layers; similarly, for guy cable forces, with the decoder of the variational self-encoder as the generator, the discriminator consists of 8 convolutional layers. And as described in the second step I, the vertical displacement of the main beam and the generator of the stay cable force (namely a decoder in the variational self-encoder) have parameter sharing.
3. In the third step, the original monitoring period of the vertical displacement of the main beam and the cable force of the stay cable is 2006-2015, and the sampling frequency is 3 Hz. The data of 2006-2007 is selected as a training set, and the rest data is used as a test to verify the universality of the established model. The vertical displacement of the main beam and the cable force of the stay cable are monitored by data from 3 to 8 points in the morning every day, the time window is 3 hours, the sliding step length is 10min, and therefore 13 training samples can be generated by the monitoring data of 5 hours every day. Finally, the training and test sets included 2986 and 6236 samples, respectively.
In the third step, the error function adopted by the model training is specifically as follows:
in the formula, L is a loss function, subscript 1 represents the vertical displacement of the main beam, and subscript 2 represents the cable force of the inhaul cable; VAE represents a variation self-editor module (composed of an encoder E and a decoder G), GAN represents a generation countermeasure network module (composed of a generator G and a discriminator D); CC stands for cyclic constraint term, and due to the existence of parameter and implicit feature vector sharing, the following may exist: encoder E by vertical displacement of girder 1 Decoder G for cable force of inhaul cable 2 Encoder E for generating a cable force, and cable force generated by the cable 2 Decoder G with girder vertical displacement 1 Generating vertical displacement of the main beam; min and max represent the minimum and maximum operations, respectively.
The optimization algorithm is an ADAM algorithm, and the model updating formula specifically comprises the following steps:
in which t represents the t-th time step, ω t Representing the parameters of the network model and,
and
respectively representing the first and second order gradients of the error at the t-th time step, m t And v t Intermediate updating variables, beta, taking into account first and second moments of the gradient, respectively 1 And beta 2 Are respectively corresponding first-order and second-order attenuation factors, eta 0 And epsilon is a denominator non-zero control parameter for the initial learning rate. Hyper-parameter setting default value as beta 1 ＝0.9,β 2 ＝0.99,ε＝10 -6 ,η 0 ＝10 -3 。
4. In the fourth step, after the model training is completed, the probability distribution of the vehicle-induced cable force and the vertical displacement of the girder is used as input, the probability distribution of the predicted cable force can be obtained, and then the EMD Distance (Earth Mover's Distance) between the predicted cable force and the real cable force probability distribution is calculated, wherein the specific calculation mode is as follows:
in the formula, EMD represents EMD distance, P and Q represent guy cable force probability distribution obtained by model prediction and real monitoring data respectively, m and n represent the number of segmented intervals of P and Q respectively, and d ij And f ij Respectively representing a displacement change value and a probability value change value from the ith P distribution section to the jth Q distribution section, and min represents the minimum value operation.
The invention has the following advantages and beneficial effects:
(1) aiming at the difficult problem of modeling of mechanical correlation of the stay cable force and the vertical displacement of the main beam of the large-span cable-stayed bridge, the unsupervised image transformation model of the probability distribution of the vertical displacement of the main beam and the stay cable force is established under the condition that a bridge structure system and load space-time information are unknown;
(2) the method provides the EMD distance between the prediction and the real cable force probability distribution as an index of the change of the bridge state, and can reflect the change of the correlation of the structural response monitoring variable probability distribution, namely the change of the bridge state;
(3) the method carries out trend removing processing on the temperature trend item of the monitoring data, is insensitive to temperature change and is generally suitable for annual monitoring data of the vertical displacement of the main beam and the cable force of the stay cable of the long-span bridge structure;
(4) the whole relational modeling process is data-driven, and errors caused by unreasonable and inaccurate model assumptions in the traditional mechanical analysis are obviously reduced;
(5) the method meets the requirements of online prediction and real-time data processing of the cable force of the large-span cable-stayed bridge, namely after model training is finished, the training set is not required to be updated, and the collected vertical displacement of the main beam and the cable force monitoring data of the stay cable are directly used as input, so that the predicted cable force distribution can be obtained;
(6) the invention improves the automation and intelligence degree, accuracy and robustness of the modeling of the mechanical correlation of the stay cable force of the large-span cable-stayed bridge and the vertical displacement of the main beam, and provides technical support for the autonomous intelligent evaluation of the service state of the large-span cable-stayed bridge.
(7) The invention can diagnose the structural state change of the large-span cable-stayed bridge caused by abnormal events such as the upgrade of a bridge health monitoring system, the abnormity of a sensor, the wire breakage of a guy cable, the earthquake, the over-limit vibration of strong wind, the traffic jam, the overload, the ship collision and the like, thereby providing scientific basis for bridge traffic management and routing inspection maintenance decision and greatly improving the intelligent degree of bridge management.
Drawings
FIG. 1 is a flow chart of one embodiment of the present invention;
FIG. 2 is a schematic diagram of data preprocessing detrending items in step one of the present invention;
FIG. 3 is a diagram illustrating an unsupervised image transformation model in step two of the present invention;
FIG. 4 is a diagram of a partial cable force distribution prediction effect obtained based on an unsupervised image transformation model according to the present invention;
FIG. 5 is a partial cable force distribution EMD distance variation graph obtained based on an unsupervised image transformation model.
Detailed Description
Example 1
The present embodiment is a cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling, as shown in fig. 1, including:
the method comprises the following steps of firstly, preprocessing the cable force of a stay cable and the displacement of a main beam: firstly, trend removing item processing is carried out, temperature influence is removed according to data characteristics, the response of the cable force of the vehicle-induced stay cable and the vertical displacement of the main beam is obtained, and the trend removing item process is shown in figure 2; then carrying out normalization processing on the data, removing data points within 0.3 times of standard deviation, and obtaining the probability distribution of the cable force of the vehicle-induced stay cable and the vertical displacement of the main beam;
step two, building an unsupervised image transformation model, wherein a model schematic diagram is shown in fig. 3, selecting the distribution of the stay cable force and the main beam vertical displacement data of all available channels, assembling the stay cable force and the main beam vertical displacement distribution matrix into a stay cable force and main beam vertical displacement distribution matrix according to rows, and respectively using the stay cable force and the main beam vertical displacement distribution matrix as two inputs of a network (a main beam vertical displacement distribution matrix and a stay cable force distribution matrix);
selecting a part of the data processed in the step one to form a training set, inputting the training set into an unsupervised image transformation model for training, wherein a loss function used in the training process is a comprehensive function constructed according to network characteristics, an optimization algorithm is an ADAM algorithm, and default hyper-parameter setting is used;
inputting the vertical displacement distribution of the main beam concentrated in the test into a trained unsupervised image transformation model to obtain the predicted cable force distribution of the cable as output, wherein the prediction effect of partial cable force distribution is shown in figure 4, calculating the EMD (Earth Mover's Distance) between the prediction and the real cable force distribution, evaluating whether the bridge state changes and corresponding action modes by taking the EMD between the prediction and the real cable force distribution as an evaluation index of the bridge state change, and the EMD Distance change of the partial cable force distribution is shown in figure 5.
Example 2
The first difference between the present embodiment and the specific embodiment is:
the first step specifically comprises the following steps:
step one, adopting the same preprocessing flow for the stay cable force and the vertical displacement time course data of the main beam, taking 30 minutes as a time window, linearly normalizing all data points in the time window to be in a range of [0,1], then drawing a frequency histogram of the section of data, taking the position of a frequency peak value as a center, respectively taking 0.03 from the left and the right, and selecting an initial data extreme point falling in the frequency range;
performing median filtering, linear interpolation and smoothing treatment on extreme points selected one by one in the steps, taking 3600 in a smoothing window to obtain a temperature trend term, and subtracting the temperature trend term from original data to obtain a preprocessed vehicle-induced vertical displacement response and a preprocessed stay cable force;
and step three, respectively carrying out normalization processing on the vehicle-induced vertical displacement response and the stay cable force data of each channel, keeping the numerical value of each channel in the range of [ -1,1], removing noise data points within 0.3 times of standard deviation, and obtaining the probability distribution of the vehicle-induced stay cable force and the vertical displacement of the main beam.
Other steps are the same as in the first embodiment.
Example 3
The present embodiment differs from the first or second embodiment in that:
the method for building the unsupervised image transformation model mainly comprises two modules, and the second step specifically comprises the following steps:
step two, building a variational self-encoder module: for the vertical displacement of the main beam, taking a probability distribution matrix of the vertical displacement of the main beam as input, wherein an encoder consists of 4 convolutional layers, 4 residual connecting layers and 2 shared convolutional layers, the input becomes an implicit characteristic vector after passing through the encoder and then enters a decoder, and the decoder consists of 2 shared convolutional layers, 4 residual connecting layers and 4 convolutional layers; similarly, for the cable force, the cable force probability distribution matrix is used as input, the encoder is composed of 4 convolution layers, 4 residual error connecting layers and 2 shared convolution layers, the input is converted into an implicit characteristic vector through the encoder and then enters the decoder, and the decoder is composed of 2 shared convolution layers, 4 residual error connecting layers and 4 convolution layers. All shared convolution layers and implicit characteristic vectors in the variational self-encoder share parameters for the vertical displacement of the main beam and the cable force of the inhaul cable.
Step two, building a generation confrontation network module: for the vertical displacement of the main beam, a decoder of a variational self-encoder is used as a generator, and a discriminator consists of 8 convolution layers; similarly, for guy cable forces, with the decoder of the variational self-encoder as the generator, the discriminator consists of 8 convolutional layers. And as described in the second step I, the vertical displacement of the main beam and the generator of the stay cable force (namely a decoder in the variational self-encoder) have parameter sharing.
The other steps are the same as those in the first to second embodiments.
Example 4
The difference between this embodiment mode and one of the first to third embodiment modes is:
in the third step, the original monitoring period of the vertical displacement of the main beam and the cable force of the stay cable is 2006-2015, and the sampling frequency is 3 Hz. The data of 2006-2007 is selected as a training set, and the rest data is used as a test to verify the universality of the established model. The vertical displacement of the main beam and the cable force of the stay cable are monitored by data from 3 to 8 points in the morning every day, the time window is 3 hours, the sliding step length is 10min, and therefore 13 training samples can be generated by the monitoring data of 5 hours every day. Finally, the training and test sets included 2986 and 6236 samples, respectively.
In the third step, the error function adopted by the model training is specifically as follows:
in the formula, L is a loss function, subscript 1 represents the vertical displacement of the main beam, and subscript 2 represents the cable force of the inhaul cable; VAE represents variation self-editor module (composed by)Decoder E and decoder G), GAN stands for the generation countermeasure network module (consisting of generator G and discriminator D); CC stands for cyclic constraint term, and due to the existence of parameter and implicit feature vector sharing, the following may exist: encoder E by vertical displacement of girder 1 Decoder G for cable force of inhaul cable 2 Encoder E for generating a cable force, and cable force generated by the cable 2 Decoder G with girder vertical displacement 1 Generating vertical displacement of the main beam; min and max represent the minimum and maximum operations, respectively.
The optimization algorithm is an ADAM algorithm, and the model updating formula specifically comprises the following steps:
in which t represents the t-th time step, ω t Representing the parameters of the network model and,
and
respectively representing the first and second order gradients of the error at the t-th time step, m t And v t Intermediate updating variables, beta, taking into account first and second moments of the gradient, respectively 1 And beta 2 Are respectively corresponding first-order and second-order attenuation factors, eta 0 And epsilon is a denominator non-zero control parameter for the initial learning rate. Hyper-parameter setting default value as beta 1 ＝0.9,β 2 ＝0.99,ε＝10 -6 ,η 0 ＝10 -3 。
The other steps are the same as those in the first to third embodiments.
The fifth concrete implementation mode:
the difference between this embodiment and one of the first to fourth embodiments is:
in the fourth step, after the model training is finished, the probability distribution of the cable force of the vehicle-driven cable and the vertical displacement of the girder is used as input, the probability distribution of the cable force of the predicted cable can be obtained, then the EMD Distance (Earth Mover's Distance) between the probability distribution of the cable force of the predicted cable and the probability distribution of the cable force of the real cable is calculated, and the specific calculation mode is as follows:
in the formula, EMD represents EMD distance, P and Q represent guy cable force probability distribution obtained by model prediction and real monitoring data respectively, m and n represent the number of segmented intervals of P and Q respectively, and d ij And f ij Respectively representing a displacement change value and a probability value change value from the ith P distribution section to the jth Q distribution section, and min represents the minimum value operation.
The other steps are the same as those in the first to fourth embodiments.
Aiming at the maintenance management decision problem of the actual large-span cable-stayed bridge facing abnormal events, the invention builds a deep learning agent model of the mechanical correlation between the vertical displacement of the main beam and the cable force by carrying out data mining on the response of the key components of the cable-stayed bridge (namely the vertical displacement of the main beam and the cable force of the cable), and evaluates the state change of the large-span cable-stayed bridge by observing the change of the evaluation index of the agent model, thereby providing a basis for maintenance decision and traffic control of the actual large-span cable-stayed bridge due to the state change, wherein the basis includes but is not limited to abnormal events such as bridge health monitoring system upgrade, sensor abnormity, cable breakage, earthquake, strong wind overrun vibration, traffic jam, overload, ship collision and the like.
Claims (4)
1. A cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling is characterized by comprising the following steps:
the method comprises the following steps of firstly, preprocessing the cable force of a stay cable and the displacement of a main beam:
adopting the same preprocessing flow for the stay cable force and the vertical displacement time course data of the girder, taking 30 minutes as a time window, linearly normalizing all data points in the time window to a range of [0,1], then drawing a frequency histogram of all data points in the time window, taking the position of a frequency peak value as a center, respectively taking 0.03 from the left and the right, and selecting an initial data extreme point falling in the frequency range;
performing median filtering, linear interpolation and smoothing on the selected extreme points of the original data, taking 3600 in a smooth window to obtain a temperature trend term, and subtracting the temperature trend term from the original data to be used as preprocessed vehicle-induced vertical displacement response and cable force of a guy cable;
respectively carrying out normalization processing on the vehicle-induced vertical displacement response and the stay cable force data of each channel to keep the numerical value of each channel within the range of [ -1,1], and removing noise data points within 0.3 times of standard deviation to obtain the probability distribution of the vehicle-induced stay cable force and the vertical displacement of the main beam;
step two, constructing a variational self-encoder module: for the vertical displacement of the main beam, taking a probability distribution matrix of the vertical displacement of the main beam as input, wherein an encoder consists of 4 convolutional layers, 4 residual connecting layers and 2 shared convolutional layers, the input becomes an implicit characteristic vector after passing through the encoder and then enters a decoder, and the decoder consists of 2 shared convolutional layers, 4 residual connecting layers and 4 convolutional layers; similarly, for the guy cable force, a guy cable force probability distribution matrix is used as input, an encoder consists of 4 convolution layers, 4 residual error connecting layers and 2 shared convolution layers, the input is converted into an implicit characteristic vector through the encoder and then enters a decoder, the decoder consists of 2 shared convolution layers, 4 residual error connecting layers and 4 convolution layers, the variation is obtained from all the shared convolution layers and the implicit characteristic vector in the encoder, and parameters are shared for the vertical displacement of a main beam and the guy cable force; building and generating a confrontation network module: for the vertical displacement of the main beam, a decoder of a variational self-encoder is used as a generator, and a discriminator consists of 8 convolution layers; for the stay cable force, a decoder of a variational self-encoder is used as a generator, a discriminator consists of 8 convolution layers, and the vertical displacement of a main beam and the generator parameter of the stay cable force are shared;
selecting a part from the data processed in the step one to form a training set, inputting the training set into an unsupervised image transformation model for training, wherein a loss function used in the training process is a comprehensive function constructed according to network characteristics, an optimization algorithm is an ADAM algorithm, and default hyper-parameter setting is used, wherein the loss function adopted by the model training is specifically as follows:
in the formula, L is a loss function, subscript 1 represents the vertical displacement of the main beam, and subscript 2 represents the cable force of the inhaul cable;
VAE represents a variational self-encoder module, which consists of an encoder E and a decoder G;
the GAN represents a generation countermeasure network module and consists of a generator G and a discriminator D;
CC stands for cyclic constraint term, and due to the existence of parameter and implicit feature vector sharing, the following may exist: encoder E by vertical displacement of girder 1 Decoder G for cable force of inhaul cable 2 Encoder E for generating a cable force, and cable force generated by the cable 2 Decoder G with girder vertical displacement 1 Generating vertical displacement of the main beam; min and max respectively represent the minimum value and the maximum value;
inputting the vertical displacement distribution of the main girder concentrated in the test into a trained unsupervised image transformation model to obtain the predicted cable force distribution of the inhaul cable as output, calculating the EMD distance between the predicted cable force distribution and the real cable force distribution, and evaluating whether the bridge state changes and a corresponding action mode by taking the EMD distance between the predicted cable force distribution and the real cable force distribution as an evaluation index of the bridge state change;
the EMD distance is calculated in a specific mode as follows:
in the formula (I), the compound is shown in the specification,
the EMD represents the distance of the EMD,
p and Q represent the probability distribution of the guy cable force obtained by model prediction and real monitoring data respectively,
m and n represent the number of segment intervals of the distributions P and Q respectively,
d ij and f ij Respectively representing a displacement change value and a probability value change value from the ith P distribution section to the jth Q distribution section, and min represents the minimum value operation.
2. The cable-stayed bridge state evaluation method based on the cable force and displacement distribution correlation modeling as claimed in claim 1, wherein in the second step, when the model is transformed in an unsupervised image, all available 27 main beam vertical displacements and 139 cable force channels of the cable are selected, and data distribution of each channel is integrated into a matrix according to rows and then used as two inputs of the network: the vertical displacement distribution matrix of the main beam and the cable force distribution matrix of the inhaul cable.
3. The cable-stayed bridge state evaluation method based on the cable force and displacement distribution correlation modeling as claimed in claim 1 or 2, characterized in that in step three, the original monitoring period of the vertical displacement of the main beam and the cable force of the cable is 2015-year, the sampling frequency is 3Hz, the data of 2006 2015-year is selected as a training set, the rest data are used as tests to verify the universality of the established model, the monitoring data of 3-8 points in the morning are adopted for the vertical displacement of the main beam and the cable force of the cable, the time window is 3 hours, the sliding step length is 10 minutes, 13 training samples are generated from the monitoring data of 5 hours each day, and finally, the training set and the testing set respectively comprise 2986 samples and 6236 samples.
4. The cable-stayed bridge state evaluation method based on the cable force and displacement distribution correlation modeling according to claim 1 or 2, characterized in that in the third step, the optimization algorithm is an ADAM algorithm, and the model update formula specifically is:
in which t represents the t-th time step, ω t Represents a parameter of the network model + t And
respectively representing the first and second order gradients of the error at the t-th time step, m t And v t Intermediate updating variables, beta, taking into account first and second moments of the gradient, respectively 1 And beta 2 Are respectively corresponding first-order and second-order attenuation factors, eta 0 Setting a default value of beta for the hyper-parameter as an initial learning rate, epsilon as a denominator non-zero control parameter 1 ＝0.9,β 2 ＝0.99,ε＝10 -6 ,η 0 ＝10 -3 。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010794104.XA CN111967185B (en) | 2020-08-10 | 2020-08-10 | Cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010794104.XA CN111967185B (en) | 2020-08-10 | 2020-08-10 | Cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111967185A CN111967185A (en) | 2020-11-20 |
| CN111967185B true CN111967185B (en) | 2022-08-02 |
Family
ID=73364975
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010794104.XA Active CN111967185B (en) | 2020-08-10 | 2020-08-10 | Cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111967185B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113570121A (en) * | 2021-07-09 | 2021-10-29 | 桂林理工大学 | Cable force trend prediction method for stayed cable based on convolutional neural network |
| CN114004467A (en) * | 2021-10-15 | 2022-02-01 | 河南工业大学 | Prefabricated bridge structure performance analysis method based on monitoring data |
| CN113865683B (en) * | 2021-12-01 | 2023-03-21 | 苏州博宇鑫交通科技有限公司 | Urban viaduct Liang Chaochong overload dynamic early warning method based on machine vision |
| CN114280061B (en) * | 2021-12-27 | 2022-10-14 | 交通运输部公路科学研究所 | Observation method for technical conditions of cable-stayed bridge cable beam anchoring area and monitoring window |
| CN114896844B (en) * | 2022-05-11 | 2023-06-23 | 昆明理工大学 | Arch bridge back buckling cable tension data processing method, system and storage medium |
| CN115048998B (en) * | 2022-06-13 | 2023-04-07 | 大连理工大学 | Cable-stayed bridge group cable force abnormity identification and positioning method based on monitoring data |
| CN117521188A (en) * | 2023-07-31 | 2024-02-06 | 中铁建大桥工程局集团第四工程有限公司 | Three-main-truss steel truss arch bridge elevation linear regulation and control method and system |
| CN116881819B (en) * | 2023-09-07 | 2023-11-14 | 成都理工大学 | Stay cable working state monitoring method based on isolated forest |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| LT5863B (en) * | 2010-10-22 | 2012-08-27 | Vilniaus Gedimino technikos universitetas | Method and device for bridge condition defining using dynamic method |
| CN105303462A (en) * | 2015-11-03 | 2016-02-03 | 国家电网公司 | EMD distance based coal consumption comparative analysis method for coal-fired power generation unit groups |
| CN108225811A (en) * | 2018-01-30 | 2018-06-29 | 四川天佑慧桥科技有限公司 | A kind of bridge structure safe assessment system based on vehicular load |
| CN110570329A (en) * | 2019-09-04 | 2019-12-13 | 四川智通路桥工程技术有限责任公司 | bridge construction overall process control and visual early warning system |
| CN111062080B (en) * | 2019-12-20 | 2022-07-12 | 哈尔滨工业大学 | Modeling method for time-space correlation depth learning of stay cable force and main beam vertical displacement |
| CN111310273B (en) * | 2020-03-16 | 2023-09-15 | 河北省交通规划设计研究院有限公司 | Full-bridge structure safety state monitoring method and system based on multi-source data |
-
2020
- 2020-08-10 CN CN202010794104.XA patent/CN111967185B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111967185A (en) | 2020-11-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111967185B (en) | Cable-stayed bridge state evaluation method based on cable force and displacement distribution correlation modeling | |
| CN111274737A (en) | Method and system for predicting remaining service life of mechanical equipment | |
| CN108873859B (en) | Bridge type grab ship unloader fault prediction model method based on improved association rule | |
| CN110457786B (en) | Ship unloader association rule fault prediction model method based on deep confidence network | |
| Yue et al. | Mechanics-Guided optimization of an LSTM network for Real-Time modeling of Temperature-Induced deflection of a Cable-Stayed bridge | |
| CN111691679B (en) | Digital twinning-based intelligent tensioning method for prestressed steel structure | |
| CN113900381B (en) | Steel structure remote health monitoring platform based on Internet of things and application method | |
| CN114282398A (en) | Bridge health monitoring system and method based on big data | |
| CN116150897A (en) | Machine tool spindle performance evaluation method and system based on digital twin | |
| CN114548375B (en) | Cable-stayed bridge girder dynamic deflection monitoring method based on two-way long-short-term memory neural network | |
| CN111062080B (en) | Modeling method for time-space correlation depth learning of stay cable force and main beam vertical displacement | |
| CN114186475B (en) | Slewing bearing service life prediction method based on Attention-MGRU | |
| CN116341272A (en) | Construction safety risk management and control system for digital distribution network engineering | |
| CN116451514A (en) | Bridge structure state evaluation method and device and electronic equipment | |
| Ge et al. | Dynamic real-time reliability prediction of bridge structures based on Copula–BHDLM and measured stress data | |
| CN113468635A (en) | Stay cable force identification method based on vibration signal and cyclic neural network | |
| CN116663126A (en) | Bridge temperature effect prediction method based on channel attention BiLSTM model | |
| CN114880767B (en) | Aero-engine residual service life prediction method based on attention mechanism Dense-GRU network | |
| CN116049958A (en) | Historical building structure monitoring data anomaly diagnosis and repair system | |
| Desjardins | Structural health assessment of bridges by long-term vibration monitoring and automated operational modal analysis | |
| CN113138058B (en) | Plate girder bridge hinge joint damage monitoring system | |
| CN109959493B (en) | Cable-stayed bridge cable damage real-time quantitative evaluation method based on static deflection modeling | |
| Chun et al. | An application of data fusion technology in structural health monitoring and damage identification | |
| CN113673010A (en) | Steel box girder evaluation method and system based on monitoring data | |
| CN112906793B (en) | Monitoring data repairing method and system for bridge health monitoring system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |