Beam bridge moving load prediction method based on CNN and dynamic strain
Through the beam bridge moving load prediction method based on CNN and dynamic strain, the problems of low efficiency and high cost in vehicle parameter prediction of traditional bridge monitoring methods are solved, and the rapid and accurate prediction of bicycle moving parameters on the bridge are achieved, reducing monitoring costs And improve monitoring efficiency.
Patent Information
- Application Number
- CN202411363961.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-28
- Publication Date
- 2025-02-07
AI Technical Summary
Existing bridge monitoring methods are inefficient and costly in predicting multiple vehicle parameters. Traditional linear models are difficult to capture complex nonlinear relationships, and relying on a large amount of historical data leads to insufficient real-time and accuracy.
Using the beam bridge movement load prediction method based on CNN and dynamic strain, the dynamic strain response data is reconstructed as a feature vector through numerical simulation, and multi-scale feature extraction and mapping relationship analysis are used to realize accurate prediction of bicycle moving parameters on actual bridges. .
It realizes rapid and accurate prediction of the driving parameters of moving vehicles on the bridge, reduces monitoring costs, improves monitoring efficiency, and provides technical support for the operation management of bridge overload and speeding vehicles and the evaluation of safety and stability.
Smart Images
Figure CN119397828A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of structural health monitoring, and in particular relates to a method for predicting moving load of a beam bridge based on CNN and dynamic strain. Background Art
[0002] With the acceleration of urbanization, bridges, as important transportation infrastructure, are increasingly valued for their safety and stability. Traditional bridge monitoring methods mainly rely on manual inspection and simple sensor data collection, which are difficult to meet the needs of complex vehicle operation conditions in highways and urban traffic. These methods often consume a lot of manpower and material resources, and are easily affected by environmental factors, resulting in inaccurate monitoring results.
[0003] In recent years, with the development of sensor technology and data processing technology, structural health monitoring methods have gradually developed towards intelligence and automation. Dynamic monitoring technology has emerged to evaluate the health status and bearing capacity of bridges by collecting dynamic response information of bridges under vehicle loads. In particular, dynamic strain response objects can reflect the state changes of bridges under real-time loads.
[0004] However, existing technologies still have some shortcomings in analyzing and predicting dynamic strain data. For example, traditional methods usually rely on linear models, which are easily affected by various factors such as changes in the external environment, vehicle type and speed, and cannot effectively capture complex nonlinear relationships. In addition, existing prediction models often have insufficient data or rely too much on a large amount of historical data, resulting in deficiencies in real-time and accuracy.
[0005] With the rapid development of artificial intelligence and computer technology, artificial neural network technology has begun to be applied to the prediction of moving loads. Artificial neural networks have strong logical reasoning capabilities and are more suitable for inverse mechanical problems such as structural load prediction. They can also avoid the expensive installation and maintenance of traditional bridge health monitoring systems. However, traditional neural networks have slow convergence speeds, and the difficulty of collecting training samples is prominent. It is difficult to obtain sufficient and high-quality data, which further limits the performance of the model. Therefore, a new multi-task learning method for vehicle parameter prediction is proposed, which has very important practical significance and application location. Summary of the invention
[0006] In order to solve the above technical problems, the present invention proposes a beam bridge moving load prediction method based on CNN and dynamic strain, aiming to solve the problems of low efficiency and high cost in the prediction of various vehicle parameters of traditional bridge monitoring methods. The core content is to reconstruct the dynamic strain response data of the beam bridge under the action of moving load into a fixed-dimensional feature vector through numerical simulation, and perform multi-scale feature extraction and mapping relationship analysis through the CNN algorithm, so as to achieve accurate prediction and analysis of the moving parameters of a single vehicle on the actual bridge.
[0007] To achieve the above object, the present invention provides a method for predicting moving load of a beam bridge based on CNN and dynamic strain, comprising:
[0008] Obtain a sample set of moving loads for beam bridges;
[0009] Building a multi-task learning network, using the beam bridge moving load sample set to train the multi-task learning network, and obtaining a moving load recognition model;
[0010] The strain data are collected and input into a moving load identification model to obtain a prediction result of a single vehicle parameter on the bridge structure.
[0011] Optionally, obtaining the beam bridge moving load sample set includes:
[0012] Acquire vehicle driving parameters, and generate training set samples by permutation and combination of the vehicle driving parameters;
[0013] The training set samples are preprocessed to obtain a beam bridge moving load sample set.
[0014] Optionally, the vehicle driving parameters include: vehicle type, vehicle speed and vehicle position.
[0015] Optionally, the method for preprocessing the training set samples is:
[0016]
[0017] Among them, x′ r is the normalized data, x r is the training sample, x max , x min are the maximum and minimum values in the training samples respectively.
[0018] Optionally, the multi-task learning network includes: a convolutional layer, a pooling layer, a fully connected layer and an output layer;
[0019] The convolution layer is used to obtain a feature map;
[0020] The pooling layer is used to perform pooling processing on the feature map;
[0021] The fully connected layer is used to obtain a feature vector according to the convolution layer and the pooling layer, and predict the mobile load parameter classification label of the vehicle according to the feature vector;
[0022] The output layer is used to output the prediction results.
[0023] Optionally, the method for obtaining the feature map is:
[0024]
[0025] in, is the jth feature map of the convolutional layer of l, represents the i-th feature map of the convolutional layer l-1, m j represents the input feature subset, The weight matrix representing the convolution between the j-th feature map of the L-th layer and the i-th feature map of the L-1 layer is the convolution kernel. represents the additive bias coefficient of the j-th feature map of the l-th layer, represents the convolution operation, and f(·) is the activation function.
[0026] Optionally, a method for performing pooling processing on the feature map is:
[0027]
[0028] in, is the multiplicative bias of the j-th neuron in the l-th layer, It represents the additive bias coefficient of the j-th feature map of the l-th layer, and down() is the sampling function.
[0029] Optionally, the method for obtaining the feature vector is:
[0030] x l =f(w l x l-1 +b l )
[0031] Among them, x l-1 is the feature map output by layer l-1, w l is the weight coefficient of the fully connected layer l and b l is the bias coefficient of the fully connected layer l, and f(·) is the activation function.
[0032] Optionally, using the beam bridge moving load sample set to train the multi-task learning network to obtain the moving load identification model includes:
[0033] Based on the Adam optimization algorithm, the multi-task learning network is trained using the beam bridge moving load sample set to obtain the moving load identification model.
[0034] Compared with the prior art, the present invention has the following advantages and technical effects:
[0035] The present invention can quickly and accurately predict the driving parameters of moving vehicles on the bridge based on the dynamic strain response data collected from the actual bridge, and provide technical support for the operation management of overloaded and speeding vehicles on the bridge and the assessment of safety and stability. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0037] Figure 1 It is a flow chart of a method for predicting moving load of a beam bridge based on CNN and dynamic strain according to an embodiment of the present invention;
[0038] Figure 2 A finite element model diagram of a prefabricated segmental beam bridge according to an embodiment of the present invention;
[0039] Figure 3 A convolutional neural network prediction schematic diagram of an embodiment of the present invention;
[0040] Figure 4 A schematic diagram of the predicted vehicle lane effect according to an embodiment of the present invention;
[0041] Figure 5 A schematic diagram of the predicted vehicle loading type effect according to an embodiment of the present invention;
[0042] Figure 6 It is a schematic diagram of the predicted vehicle driving speed effect according to an embodiment of the present invention. DETAILED DESCRIPTION
[0043] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0044] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0045] The present invention proposes a beam bridge moving load prediction method based on CNN and dynamic strain. Figure 1 As shown, the specific steps include:
[0046] Obtain a sample set of moving loads for beam bridges;
[0047] Build a multi-task learning network and use the beam bridge moving load sample set to train the multi-task learning network to obtain a moving load recognition model;
[0048] The strain data is collected and input into the moving load identification model to obtain the prediction results of the single-vehicle parameters on the bridge structure.
[0049] Specifically, the present invention can quickly and reliably analyze and predict the driving parameters of moving vehicles on the bridge based on the dynamic strain response data collected from the actual bridge, and provide technical support for the operational management of overloaded and speeding vehicles on the bridge and the assessment of safety and stability.
[0050] Furthermore, obtaining a sample set of moving loads of a beam bridge includes:
[0051] Obtain vehicle driving parameters, and generate training set samples by permutation and combination of the vehicle driving parameters;
[0052] The training set samples are preprocessed to obtain the beam bridge moving load sample set.
[0053] Specifically, the finite element model of the precast segmental bridge structure was established using the finite element software ABAQUS. Figure 2 As shown, it is used as the basic model of the present invention to be used for numerical simulation calculation.
[0054] Furthermore, the vehicle driving parameters include: vehicle type, vehicle speed and vehicle position.
[0055] Specifically, permutations and combinations are used to generate multiple groups of training set samples in order to systematically simulate different vehicle parameters and their corresponding dynamic strain responses. The vehicle driving parameters, i.e., the influencing factors of the dynamic strain response of the precast segmental beam bridge, have a number p of 3, including vehicle loading type, vehicle driving lane, and vehicle driving speed.
[0056] Determine the range of vehicle lanes and vehicle speeds in the test parameters. Specifically, [X imin , X imax ](i=1,2,…,p), where X imin and X imax are the minimum and maximum limits of the ith parameter, respectively.
[0057] According to urban bridge specifications, vehicle types are divided into standard categories and assigned different overload ratios.
[0058] Furthermore, the method for preprocessing the training set samples is:
[0059]
[0060] Among them, x′r is the normalized data, x r is the training sample, x max , x min are the maximum and minimum values in the training samples respectively.
[0061] The normalized samples are transformed according to the features to form the training sample set matrix x t , where n is the number of design samples, m is the number of strain sensors, and s is the normalized value of the response data collected by a single sensor when the vehicle passes the full bridge at a constant speed.
[0062] Furthermore, the multi-task learning network includes: a convolutional layer, a pooling layer, a fully connected layer, and an output layer;
[0063] Convolutional layer, used to obtain feature maps;
[0064] Pooling layer, used to perform pooling on feature maps;
[0065] The fully connected layer is used to obtain the feature vector according to the convolution layer and the pooling layer, and predict the vehicle's mobile load parameter classification label according to the feature vector;
[0066] The output layer is used to output the prediction results.
[0067] Specifically, the constructed beam bridge moving load prediction learning network is a multi-task learning model consisting of a two-dimensional convolution input layer, two one-dimensional convolution layers, a pooling layer, a fully connected layer and an output layer. The fully connected layer is used as a specific task layer to predict the vehicle's moving load parameter classification label.
[0068] Specifically, let the sample x t The input is sent to the convolution layer and convolved with a convolution kernel of size i*j*r. The output feature map is obtained as follows:
[0069]
[0070] Where: Represents the jth and ith feature maps of the convolutional layer l and l-1 respectively, and m j represents the input feature subset, The weight matrix representing the convolution between the j-th feature map of the L-th layer and the i-th feature map of the L-1 layer is the convolution kernel. represents the additive bias coefficient of the j-th feature map of the l-th layer, represents the convolution operation, f(·) is the activation function, and the ReLU activation function is used.
[0071] The obtained feature map is pooled to prevent overfitting of the data. The formula is as follows:
[0072]
[0073] Where: and They represent the multiplicative bias and additive bias of the j-th neuron in the l-th layer respectively, and down() is the sampling function.
[0074] The high-dimensional features obtained in the previous two steps are flattened and mapped into a feature vector of fixed length. The formula is as follows:
[0075] x l =f(w l x l-1 +b l )
[0076] Where: x l-1 is the feature map output by layer l-1, w l and b l are the weight coefficient and bias coefficient of the fully connected layer l, respectively, and f(·) is the activation function.
[0077] During model training, the Adam optimization algorithm was used to train the network model. The training batch size was set to 25, the maximum number of training times was 500, the initial learning rate was set to 0.001, and the remaining parameters of the optimization algorithm were based on the default parameters. The prediction effect of the model training process was determined by the coefficient of determination R 2 To calculate, the calculation formula is as follows:
[0078]
[0079] Where: n is the number of samples, is the calculated value, y i is the true value, is the average of the true values.
[0080] Furthermore, the strain response data of the bridge structure health monitoring system is collected in real time by installing strain gauges at key positions of the bridge. The strain response data is used as a validation set to evaluate the performance of the multi-task learning network. After training, the multi-task learning network can perform feature extraction and pattern recognition on the input strain data, thereby realizing accurate identification of single vehicle parameters on the bridge structure, including the prediction and analysis of parameters such as vehicle speed, vehicle model and vehicle position.
[0081] The following is a detailed description with examples:
[0082] Step 1: Use the finite element software ABAQUS numerical simulation analysis software to establish a three-dimensional finite element model of the prefabricated segmental beam bridge structure under the impact of moving loads as the basis for subsequent working condition simulation, such as Figure 2 shown.
[0083] Step 2: Select the number of vehicle driving parameters p, i.e., the influencing factors of the dynamic strain response of the precast segmental beam bridge, including vehicle loading type, vehicle driving lane, and vehicle driving speed.
[0084] Furthermore, according to the relevant specifications of urban bridges and the actual operation of a specific project, the vehicle load, driving speed, and lane of action are comprehensively considered and reasonably simplified, and the lane driving is reasonably simplified and divided into four categories: the first lane, the second lane, the third lane, and the fourth lane. At the same time, the vehicle speed range is set at [10,25] m / s, and the specific speed is divided into a level of 3 m / s. The detailed classification of vehicle loading types is shown in Table 1:
[0085] Table 1
[0086]
[0087] Step 3: The experiment is simulated according to the training set scheme designed in step 2, and the obtained sample data is normalized according to the following formula:
[0088] Among them, x r , x max , x min are the training samples, the maximum value and the minimum value in the training samples respectively.
[0089] Furthermore, the normalized samples are transformed according to the features to form the training sample set matrix x t , where n is the number of design samples, m is the number of strain sensors, and s is the normalized value of the response data collected by a single sensor when the vehicle passes the full bridge at a constant speed. In this embodiment, the total number of design samples n is 240 after permutation and combination according to the vehicle parameter types divided in step 2; m is 9, representing 9 strain measurement points; s is 356, representing the normalized data extracted under one working condition. Finally, 70% of the samples are randomly selected from all the constructed samples as training sets, and 30% of the samples are selected as test sets.
[0090] Step 4: The multi-task learning network process for moving load parameter prediction based on CNN and dynamic strain is as follows:
[0091] 1. Input the sample into the convolution layer and perform convolution operation with the convolution kernel of size i*j*r. Then the output feature map is obtained, and its formula is:
[0092]
[0093] in Represents the jth and ith feature maps of the convolutional layer l and l-1 respectively, and mj represents the input feature subset, The weight matrix representing the convolution between the j-th feature map of the L-th layer and the i-th feature map of the L-1 layer is the convolution kernel. represents the additive bias coefficient of the j-th feature map of the l-th layer, represents the convolution operation, f(·) is the activation function, and the ReLU activation function is used.
[0094] 2. Perform a pooling operation on the obtained feature map to prevent overfitting of the data. The formula is:
[0095]
[0096] in and They represent the multiplicative bias and additive bias of the j-th neuron in the l-th layer respectively, and down() is the sampling function.
[0097] 3. Flatten the high-dimensional features obtained in the previous two steps and map them into a feature vector of fixed length. The formula is:
[0098] x l =f(w l x l-1 +b l ),
[0099] where x l-1 is the feature map output by layer l-1, w l and b l are the weight coefficient and bias coefficient of the fully connected layer l respectively.
[0100] 4. Model training The Adam optimization algorithm is used to train the network model. The training batch size is set to 25, the maximum number of training times is 500, the initial learning rate is set to 0.001, and the other parameters of the optimization algorithm follow the default parameters. The prediction effect of the model training process is calculated using the determination coefficient R^2, and the calculation formula is as follows:
[0101]
[0102] Where n is the number of samples, is the calculated value, y i is the true value, is the average of the true values.
[0103] Furthermore, the multi-task learning network architecture for mobile load parameter prediction based on CNN and dynamic strain is as follows: Figure 3 shown.
[0104] The strain response data of the specified position of the bridge structure health monitoring system is obtained as the validation set, and it is input into the multi-task learning network for beam bridge moving load prediction based on CNN and dynamic strain that has been trained in step 4, and the results of single-vehicle parameter identification on the bridge structure are output.
[0105] Furthermore, the influence of moving vehicles on the dynamic strain of the beam bridge is compared with the predicted value output by the neural network model. The prediction results of vehicle loading type, vehicle speed and vehicle action lane are shown in the figure below. Figure 4-Figure 6 The prediction effect of the model on each indicator is shown in the following table:
[0106]
[0107] In the above-mentioned implementation case, the present invention, based on a certain engineering example, establishes a finite element analysis model, which can quickly and reliably analyze and predict the driving parameters of moving vehicles on the bridge according to the dynamic strain response data collected from the actual bridge, and provide technical support for the operation management of overloaded and speeding vehicles on the bridge and the evaluation of safety and stability.
[0108] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A method for predicting moving load of beam bridge based on CNN and dynamic strain, characterized in that: include: Obtain a sample set of moving loads for beam bridges; Building a multi-task learning network, using the beam bridge moving load sample set to train the multi-task learning network, and obtaining a moving load prediction model; The strain data are collected and input into a moving load prediction model to obtain a prediction result of a single vehicle parameter on the bridge structure.
2. The method for predicting moving load of a beam bridge based on CNN and dynamic strain according to claim 1 is characterized in that: Obtaining the beam bridge moving load sample set includes: Acquire vehicle driving parameters, and generate training set samples by permutation and combination of the vehicle driving parameters; The training set samples are preprocessed to obtain a beam bridge moving load sample set.
3. The method for predicting moving load of a beam bridge based on CNN and dynamic strain according to claim 2 is characterized in that: The vehicle driving parameters include: vehicle type, vehicle speed and vehicle position.
4. The method for predicting moving load of a beam bridge based on CNN and dynamic strain according to claim 2 is characterized in that: The method for preprocessing the training set samples is: Among them, x r is the normalized data, x r is the training sample, x max , x min are the maximum and minimum values in the training samples respectively.
5. The method for predicting moving load of a beam bridge based on CNN and dynamic strain according to claim 1 is characterized in that: The multi-task learning network includes: a convolutional layer, a pooling layer, a fully connected layer and an output layer; The convolution layer is used to obtain a feature map; The pooling layer is used to perform pooling processing on the feature map; The fully connected layer is used to obtain a feature vector according to the convolution layer and the pooling layer, and predict the mobile load parameter classification label of the vehicle according to the feature vector; The output layer is used to output the prediction results.
6. The method for predicting moving load of a beam bridge based on CNN and dynamic strain according to claim 5 is characterized in that: The method for obtaining the feature map is: in, is the jth feature map of the convolutional layer of l, represents the i-th feature map of the convolutional layer l-1, m j represents the input feature subset, The weight matrix representing the convolution between the j-th feature map of the L-th layer and the i-th feature map of the L-1 layer is the convolution kernel. represents the additive bias coefficient of the j-th feature map of the l-th layer, represents the convolution operation, and f(·) is the activation function.
7. The method for predicting moving load of a beam bridge based on CNN and dynamic strain according to claim 6 is characterized in that: The method for performing pooling processing on the feature map is: in, is the multiplicative bias of the j-th neuron in the l-th layer, It represents the additive bias coefficient of the j-th feature map of the l-th layer, and down() is the sampling function.
8. The method for predicting moving load of a beam bridge based on CNN and dynamic strain according to claim 7 is characterized in that: The method for obtaining the feature vector is: x l =f(w l x l-1 +b l ) Among them, x l-1 is the feature map output by layer l-1, w l is the weight coefficient of the fully connected layer l and b l is the bias coefficient of the fully connected layer l, and f(·) is the activation function.
9. The method for predicting moving load of a beam bridge based on CNN and dynamic strain according to claim 1 is characterized in that: Using the beam bridge moving load sample set to train the multi-task learning network to obtain the moving load prediction model includes: Based on the Adam optimization algorithm, the multi-task learning network is trained using the beam bridge moving load sample set to obtain the moving load prediction model.