Bridge dynamic weighing identification method based on time sequence deep learning network

Abstract

The invention discloses a bridge dynamic weighing identification method based on a time sequence deep learning network, and the method comprises the steps: carrying out the data statistics, and obtaining a vehicle load spectrum; according to the vehicle load spectrum, using a Monte Carlo random sampling technology to extract a vehicle information sample, and forming random traffic flow information; establishing and correcting a bridge finite element model; according to vehicle information of a single vehicle in the random traffic flow, an axle coupling model is established based on the corrected bridge finite element model, bridge response of vehicle bridge crossing is simulated by using the axle coupling model, and a vehicle information mapping bridge response data set is established; building a neural network framework, and training a neural network based on the vehicle information mapping bridge response data set to obtain a vehicle information identification agent model; and identifying vehicle information from the actually measured vehicle-induced bridge response by using the proxy model. According to the method, the vehicle information can be accurately estimated from the vehicle-caused bridge displacement, and the vehicle space coordinate information does not need to be acquired priori.

Description

Technical Field

[0001] The present invention belongs to the field of bridge health monitoring and relates to dynamic weighing of bridges, and specifically to a dynamic weighing identification method for bridges based on a time series deep learning network. Background Art

[0002] The bridge dynamic weighing technology regards the bridge as a "scale" and uses the bridge response generated by the vehicle passing over the bridge to identify the static load of the vehicle. This technology is of great significance for monitoring the health status of the bridge and monitoring the phenomenon of vehicle overloading.

[0003] However, the dynamic weighing technology of bridges is still difficult to apply in engineering practice. The main reason is that it is impossible to accurately perceive the real-time spatial coordinates of the vehicles on the bridge deck, and it is also difficult to identify the static load of the vehicles under multi-vehicle conditions. This makes it difficult to use the influence line method to identify the static load of the vehicles in engineering practice. Although there are neural network-based algorithms that can identify the static load of vehicles without the spatial position of the vehicles, such methods are currently only used under numerical simulation and laboratory test conditions. Further research is still needed to realize the engineering application of vehicle static load identification. Summary of the invention

[0004] Purpose of the invention: In order to overcome the shortcomings of the prior art, a bridge dynamic weighing identification method based on a time series deep learning network is provided, which can accurately estimate vehicle information from vehicle-induced bridge displacement without the need to obtain vehicle spatial coordinate information a priori.

[0005] Technical solution: To achieve the above purpose, the present invention provides a bridge dynamic weighing identification method based on a time series deep learning network, comprising the following steps:

[0006] S1: Perform data statistics to obtain vehicle load spectrum;

[0007] S2: Based on the vehicle load spectrum, the Monte Carlo random sampling technique is used to extract vehicle information samples to form random vehicle flow information;

[0008] S3: Establish and modify the finite element model of the bridge;

[0009] S4: According to the vehicle information of a single vehicle in the random traffic flow, a vehicle-bridge coupling model is established based on the modified bridge finite element model. The vehicle-bridge coupling model is used to simulate the bridge response of the vehicle passing the bridge, and a vehicle information mapping bridge response data set is established;

[0010] S5: Build a neural network framework, train the neural network based on the vehicle information mapping bridge response dataset, and obtain the proxy model for vehicle information recognition;

[0011] S6: Using a surrogate model to identify vehicle information from measured vehicle-induced bridge responses.

[0012] Furthermore, in step S1, the dynamic weighing system data of the local bridges of the same type is used for statistics, and the statistical content includes lanes, vehicle types, speeds, axle loads, wheelbases and vehicle headway.

[0013] Furthermore, in step S3, a finite element model of the bridge is established according to the drawings of the monitored bridge, and the finite element model is corrected according to the modal parameters of the on-site modal test of the bridge.

[0014] Furthermore, the dynamic equation of the vehicle-bridge coupling system in the vehicle-bridge coupling model in step S4 is:

[0015]

[0016] Among them, M b , C b and K b They are the mass, damping and stiffness matrices of the bridge model, which can be obtained by finite element software modeling; and b are the acceleration, velocity and displacement vector of the bridge respectively; M v , C v and K v are the mass, damping and stiffness matrices of the vehicle model respectively; C vb , C bv , and C bb is the coupled damping matrix; K vb , K bv , and K bb is the coupled stiffness matrix; F v and F b They are the vehicle load component and bridge load component of the vehicle-bridge coupling system respectively.

[0017] Furthermore, in step S5, a convolutional neural network is used to build a dynamic weighing technology framework for bridges. The network framework accepts a vehicle-induced bridge response signal as input, performs normalization on each sample, extracts signal features through convolutional layers, residual blocks, pooling layers and other layers, and distinguishes vehicle information from the bridge response.

[0018] Beneficial effects: Compared with the prior art, the present invention can simulate the data set of vehicle information mapping bridge response through the modified finite element numerical model, use the data set to generate a proxy model of the neural network, and finally use the model to identify the vehicle from the measured vehicle-induced bridge response. Because the method is trained through a time-series deep learning network, it does not need to extract the real-time spatial coordinates of the vehicle, and can adaptively identify the number of vehicle axles, axle dead weight, lane and vehicle speed from the bridge response. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1is a flow chart of the method of the present invention;

[0020] Figure 2 This is the encoder structure diagram of the convolutional neural network;

[0021] Figure 3 This is the decoder structure diagram of the convolutional neural network;

[0022] Figure 4 This is a sample data graph of a bridge response under random traffic flow. DETAILED DESCRIPTION

[0023] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0024] like Figure 1 As shown, the present invention provides a bridge dynamic weighing identification method based on a time series deep learning network, comprising the following steps:

[0025] S1: Perform data statistics to obtain vehicle load spectrum;

[0026] Traffic load has great regional characteristics and is also related to the traffic capacity of the tested bridge. In order to make the statistical results more accurately express the vehicle load distribution of the tested bridge, the dynamic weighing system data of the same type of local bridges should be used. The statistical content includes lanes, vehicle types, speeds, axle loads, wheelbases and vehicle headway.

[0027] S2: According to the vehicle load spectrum, the Monte Carlo random sampling technique is used to extract vehicle information samples to form random traffic flow information. The bridge response sample data under the action of random traffic flow obtained in this embodiment is as follows: Figure 4 As shown;

[0028] S3: Establish and modify the finite element model of the bridge;

[0029] A finite element model of the bridge is established according to the drawings of the monitored bridge. The finite element model is modified according to the modal parameters of the on-site modal test of the bridge. By modifying the stiffness, mass and constraint conditions of the structure, the model is made to have the same natural frequency as the actual bridge.

[0030] S4: According to the vehicle information of a single vehicle in the random traffic flow, a vehicle-bridge coupling model is established based on the modified bridge finite element model. The vehicle-bridge coupling model is used to simulate the bridge response of the vehicle passing the bridge, and a vehicle information mapping bridge response data set is established;

[0031] The dynamic equation of the vehicle-bridge coupling system in the vehicle-bridge coupling model is:

[0032]

[0033] Among them, M b , C b and K b They are the mass, damping and stiffness matrices of the bridge model, which can be obtained by finite element software modeling; and b are the acceleration, velocity and displacement vector of the bridge respectively; M v , C v and K v are the mass, damping and stiffness matrices of the vehicle model respectively; C vb , C bv , and C bb is the coupled damping matrix; K vb , K bv , and K bb is the coupled stiffness matrix; F v and F b They are the vehicle load component and bridge load component of the vehicle-bridge coupling system respectively.

[0034] S5: Build a neural network framework, train the neural network based on the vehicle information mapping bridge response dataset, and obtain the proxy model for vehicle information recognition;

[0035] A convolutional neural network is used to build a technical framework for dynamic weighing of bridges. The network framework accepts vehicle-induced bridge response signals as input, performs normalization on each sample, extracts signal features through convolutional layers, residual blocks, pooling and other layers, and distinguishes vehicle information from bridge responses.

[0036] like Figure 2 As shown in the figure, the structure and operation process of the encoder of the convolutional neural network are shown; Figure 3 As shown, the structure and operation process of the decoder of the convolutional neural network are demonstrated.

[0037] S6: Based on the measured vehicle-induced bridge response, a proxy model is used to identify vehicle information from the measured vehicle-induced bridge response, such as the number of axles, axle dead weight, vehicle speed, and vehicle lane.

Claims

1. A bridge dynamic weighing identification method based on time series deep learning network, characterized in that: The steps include: S1: Perform data statistics to obtain vehicle load spectrum; S2: Based on the vehicle load spectrum, the Monte Carlo random sampling technique is used to extract vehicle information samples to form random vehicle flow information; S3: Establish and modify the finite element model of the bridge; S4: According to the vehicle information of a single vehicle in the random traffic flow, a vehicle-bridge coupling model is established based on the modified bridge finite element model. The vehicle-bridge coupling model is used to simulate the bridge response of the vehicle passing the bridge, and a vehicle information mapping bridge response data set is established; S5: Build a neural network framework, train the neural network based on the vehicle information mapping bridge response dataset, and obtain the proxy model for vehicle information recognition; S6: Using a surrogate model to identify vehicle information from measured vehicle-induced bridge responses.

2. According to the method of claim 1, the method is characterized in that: In the step S1, the dynamic weighing system data of the local bridges of the same type is used for statistics, and the statistical content includes lanes, vehicle types, speeds, axle loads, wheelbases and vehicle headway.

3. According to the method of claim 1, the method is characterized in that: In the step S3, a finite element model of the bridge is established according to the drawings of the monitored bridge, and the finite element model is corrected according to the modal parameters of the on-site modal test of the bridge.

4. According to the method of claim 1, the method is characterized in that: The dynamic equation of the vehicle-bridge coupling system in the vehicle-bridge coupling model in step S4 is: Among them, M b , C b and K b They are the mass, damping and stiffness matrices of the bridge model, which can be obtained by finite element software modeling; and b are the acceleration, velocity and displacement vector of the bridge respectively; M v , C v and K v are the mass, damping and stiffness matrices of the vehicle model respectively; C vb , C bv , and C bb is the coupled damping matrix; K vb , K bv , and K bb is the coupled stiffness matrix; F v and F b They are the vehicle load component and bridge load component of the vehicle-bridge coupling system respectively.

5. According to the method of bridge dynamic weighing identification based on time series deep learning network in claim 1, it is characterized in that: In step S5, a convolutional neural network is used to build a dynamic weighing technology framework for bridges. The network framework accepts the vehicle-induced bridge response signal as input, performs normalization on each sample, extracts signal features through layers, and distinguishes vehicle information from the bridge response.

6. The bridge dynamic weighing identification method based on time series deep learning network according to claim 5 is characterized in that: In the network framework, signal features are extracted through layer blocks, and the layer blocks include convolutional layers, residual blocks, and pooling layers.

7. The bridge dynamic weighing identification method based on time series deep learning network according to claim 3 is characterized in that: The finite element model in step S3 is modified by modifying the stiffness, mass and constraint conditions of the structure so that the model and the actual bridge have the same natural frequency.

Images