Vehicle-bridge parameter synchronous inversion identification method based on depth time sequence model

Abstract

The invention discloses a vehicle-bridge parameter synchronous inversion identification method based on a depth time sequence model. The method comprises the following steps: collecting bridge response data; an MFNI model is constructed; an MFNI model is trained; using the trained MFNI model to extract vehicle moving force features, and obtaining a dynamic force time history of each axle of the vehicle on the bridge; the DMF-corr block is used for correcting the output of the decoder; extracting wheelbase and vehicle speed information from vehicle moving force features output by the MFNI model through a time sequence feature derivation method; obtaining a vehicle-induced response sample; constructing a PTAD model, and capturing damage sensitive features in the vehicle-caused response sample through the PTAD model; acquiring damage indexes through the damage sensitive features, evaluating the bridge state according to the damage indexes, and quantifying the damage degree of the bridge. According to the invention, accurate and efficient detection of bridge structural damage is realized, a new technical means is provided for structural health monitoring in the field of bridge engineering, and the method has important significance for promoting intelligent and refined development in the field of bridge engineering.

Description

Technical Field

[0001] The present invention belongs to the field of bridge structure health monitoring, and in particular relates to a vehicle-bridge parameter synchronous inversion identification method based on a deep time series model. Background Art

[0002] In the field of bridge engineering, with the continuous increase in traffic volume and the extension of bridge service life, real-time and accurate health monitoring of bridge structures is particularly important. Traditional monitoring methods often focus on the static analysis of a single parameter, which makes it difficult to fully capture the dynamic response of bridge structures under vehicle loads and their complex interactions with vehicle parameters.

[0003] Traditional vehicle-bridge interaction analysis methods usually rely on prior knowledge of parameters such as vehicle speed, number of axles, and wheelbase. The acquisition of these parameters often relies on complex vehicle detection systems or manual measurements, which is not only costly but also inefficient. In addition, existing moving force identification (MFI) technologies are mostly based on time domain, frequency domain or interpretive methods. These methods are affected to varying degrees by the dynamic characteristics of bridge structures, the diversity of vehicle types, and environmental factors (such as road roughness), which limits their accuracy and robustness in practical applications. In recent years, with the rapid development of deep learning technology, especially its outstanding performance in time series analysis and pattern recognition, it has provided new ideas for the synchronous inversion identification of vehicle-bridge parameters. Deep neural networks, with their powerful nonlinear mapping capabilities and automatic feature extraction capabilities, have shown great potential in processing dynamic response data of complex systems. However, most of the existing deep learning-based MFI methods have failed to effectively solve the problem of high-precision identification in the absence of detailed vehicle parameters (such as vehicle speed and axle configuration).

[0004] In the field of bridge engineering, structural damage state detection is also one of the key technologies to ensure the safe operation of bridges. Traditional bridge damage detection methods mostly rely on regular manual inspections or model-based damage identification technologies, which have limitations such as long detection cycles, high costs, susceptibility to human factors, and large model errors. Especially when the bridge structure is complex and the sensors are sparsely deployed, the accuracy and efficiency of traditional methods often cannot meet actual needs. In recent years, with the rapid development of deep learning technology, especially its wide application in the field of time series analysis, models based on deep neural networks have shown great potential in structural health monitoring. In particular, the Transformer model has achieved great success in the fields of natural language processing and computer vision, and its powerful sequence modeling capabilities have provided new ideas for the processing of time series data. However, directly applying the Transformer model to the field of structural health monitoring still faces many challenges, such as high model complexity, large amount of computation, and insufficient ability to capture local features. In order to overcome the above challenges, researchers proposed an improved model based on PatchTST (Patch-based Time Series Transformer), which reduces the length of the input sequence by introducing a patch mechanism and emphasizes local semantic information, thereby reducing the amount of computation while ensuring model performance. This innovation provides a new solution for the processing of time series data and a new idea for bridge structural health monitoring. However, most of the existing deep learning-based bridge damage detection methods rely on a large amount of labeled data, which is often difficult to achieve in practical applications. Summary of the invention

[0005] Purpose of the invention: In order to overcome the deficiencies in the prior art, a vehicle-bridge parameter synchronous inversion identification method based on a deep time series model is provided, which realizes accurate and efficient detection of bridge structure damage, provides new technical means for structural health monitoring in the field of bridge engineering, and is of great significance for promoting the intelligent and refined development of the field of bridge engineering.

[0006] Technical solution: To achieve the above purpose, the present invention provides a vehicle-bridge parameter synchronous inversion identification method based on a deep time series model, comprising the following steps:

[0007] S1: Collect bridge response data and preprocess them; the data include displacement, acceleration, etc. The preprocessing includes denoising, outlier removal, etc. to ensure the quality of input data;

[0008] S2: construct the MFNI model, including the encoder, decoder, and dynamic moving load correction DMF-corr block;

[0009] S3: The finite element model of the target bridge is modified based on the preprocessed bridge response data to generate simulation response samples for MFNI model training; by optimizing the mean absolute error loss function, the model can accurately identify the vehicle movement force from the bridge response;

[0010] Conduct parameter sensitivity analysis to evaluate the impact of different parameters on model performance, further optimize model structure and parameter settings, and improve model robustness and recognition accuracy;

[0011] S4: Use the trained MFNI model to process the new bridge response data, extract the vehicle movement force characteristics, and obtain the dynamic force time history of each axle of the vehicle on the bridge;

[0012] S5: Use the DMF-corr block to correct the decoder output to ensure the accuracy of the recognition result; through the correction function, the model can more accurately determine the number of vehicle axles and correct the load time history of the non-bridge acting time step;

[0013] S6: Extract wheelbase and vehicle speed information from the vehicle mobility features output by the MFNI model through the time series feature derivation method;

[0014] S7: Integrate the MFNI model into the bridge health monitoring system to achieve real-time monitoring and identification of vehicle mobility; this method can be used in multiple fields such as bridge structure health monitoring, traffic flow analysis, overload detection, etc., to improve the intelligent level of bridge management and maintenance;

[0015] The MFNI model is used to collect and separate the vehicle-induced response signals in the bridge monitoring signals in real time, denoise the vehicle-induced response signals, and divide the vehicle-induced response samples using the sliding window technology.

[0016] S8: Build a PTAD model based on PatchTST and group convolution, and train the PTAD model using vehicle-induced response samples;

[0017] S9: Capture the damage-sensitive features in vehicle-induced response samples through the trained PTAD model;

[0018] S10: The damage index is obtained through the damage sensitive characteristics, the bridge status is evaluated according to the damage index, and the degree of bridge damage is quantified.

[0019] Furthermore, in step S2, the encoder adopts an extended causal convolution structure to capture local feature information and interactive dependency patterns in the input data; the decoder uses a recurrent neural network model with a stacked extended structure to capture complex time-dependent patterns; and the DMF-corr block is used to correct and modify the decoding results to improve recognition accuracy.

[0020] Furthermore, the load correction function of the DMF-corr block in step S2 is implemented by adding a ReLU activation function after the output of the fully connected layer, and its specific form is Among them, z is the output of the neural network before correction.

[0021] Furthermore, the generation of the simulation response sample in step S3 includes:

[0022] A1: Modify the finite element model to randomly generate road roughness data;

[0023] A2: Generate a vehicle model with random key parameters to be identified, and use it to generate a large number of simulation training samples.

[0024] Furthermore, the time series feature derivation method in step S6 solves the wheelbase and vehicle speed information by using the corrected load time history results, specifically including:

[0025] B1: Calculate the time step difference τ of different axles acting on the bridge based on the cross-correlation function i , the formula is as follows:

[0026]

[0027] B2: Use the first-order difference to obtain the time steps of the axle on and off the bridge, so as to obtain the average action time Δt of each axle load on the bridge;

[0028] B3: Substitute and solve the following linear non-homogeneous equations to obtain the wheelbase a i And the estimated vehicle speed v:

[0029]

[0030] Furthermore, the damage index in step S10 includes a long-term damage index DI long-term and short-term damage index DI short-term By calculating the reconstruction error of the test data in the PTAD model, the error vector θ composed of the average REI value of each sensor in the sliding window is obtained, and compared with the reference error vector θ0 in the healthy state, DI is calculated. long-term and DI short-term ;

[0031] The calculation formula is as follows:

[0032]

[0033] Furthermore, in step S10, the bridge state is evaluated according to the damage index to quantify the degree of bridge damage, specifically including: statistically establishing short-term and long-term damage thresholds with a confidence level of 95% respectively based on the DIs values ​​obtained in the healthy state;long-term Compare with the preset damage threshold to determine whether the bridge has potential damage; at the same time, regularly analyze the short-term damage index DI short-term , to assess the extent of damage in real time;

[0034] According to DI long-term and DI short-berm The results of the bridge damage are used to quantify the damage extent and provide a basis for subsequent maintenance decisions.

[0035] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0036] 1. This invention constructs an advanced vehicle-bridge parameter synchronous inversion identification framework by integrating deep learning and time series analysis technology. This model can make full use of the rich information contained in the bridge vibration signal, combined with the dynamic characteristics of the vehicle during driving, to achieve synchronous and high-precision inversion of key parameters such as vehicle moving load and bridge structure state. This innovation not only overcomes the limitations of traditional methods in dealing with complex dynamic systems, but also significantly improves the utilization efficiency of monitoring data and the accuracy of parameter identification.

[0037] 2. Developing a new method that can accurately identify and synchronously invert key parameters (such as vehicle axle weight, wheelbase, vehicle speed and bridge dynamic response) in the process of vehicle-bridge interaction without relying on specific vehicle information is of great significance for improving the intelligent level and practical application effect of bridge health monitoring. The method of the present invention is aimed at this technical problem. By constructing a specially designed deep neural network architecture, it can achieve efficient and accurate identification of the vehicle-bridge interaction process, providing strong support for the safety assessment and maintenance management of bridge structures.

[0038] 3. The present invention provides a method for bridge damage detection using unlabeled data (i.e., unsupervised learning). It also takes into account the important role of vehicle-bridge interaction (VBI) in the dynamic response of bridges, integrates VBI data into the damage detection model, and realizes the synchronous inversion and identification of vehicle-bridge parameters, which will further improve the accuracy and real-time performance of damage detection.

[0039] 4. The present invention aims to achieve accurate and efficient detection of bridge structure damage, provide new technical means for structural health monitoring in the field of bridge engineering, and is of great significance for promoting the intelligent and refined development of bridge engineering. The method framework of the present invention has an automated processing flow from the original signal to the final recognition result, and then provides real-time synchronous changes in vehicle and bridge parameters, providing strong theoretical method support for bridge safety assessment, damage warning and maintenance decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the overall process of the method of the present invention;

[0041] Figure 2 This is a structural diagram of a deep learning network involved in the bridge dynamic weighing system of the present invention;

[0042] Figure 3 This is a structural diagram of a deep learning network involved in the abnormal state diagnosis of a bridge according to the present invention;

[0043] Figure 4 This is the preliminary verification effect diagram of the bridge dynamic weighing system;

[0044] Figure 5 It is a preliminary verification effect diagram for the diagnosis of long-term abnormal conditions;

[0045] Figure 6 This is a preliminary verification effect diagram for short-term abnormal state diagnosis. DETAILED DESCRIPTION

[0046] 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.

[0047] The present invention provides a vehicle-bridge parameter synchronous inversion identification method based on a deep time series model, such as Figure 1 As shown in Figure 1, it includes five stages: data acquisition, bridge dynamic weighing network training, load weighing network deployment, vehicle-induced response sample acquisition, and bridge abnormal state diagnosis. Figure 1 to Figure 3 , the method of the present invention is described, comprising the following steps:

[0048] 1) Data acquisition stage

[0049] The bridge response data is collected and preprocessed; the data includes displacement, acceleration, etc. The preprocessing includes denoising, outlier removal, etc. to ensure the quality of the input data; this embodiment uses db8 wavelet to perform 4-layer decomposition to achieve the purpose of signal noise reduction.

[0050] 2) Bridge dynamic weighing network training stage

[0051] 2-1) Construct the MFNI model, including the encoder, decoder and dynamic moving load correction DMF-corr block;

[0052] like Figure 2As shown in the figure, the encoder adopts an extended causal convolution structure to capture local feature information and interactive dependency patterns in the input data; the decoder uses a recurrent neural network model with a stacked extended structure to capture complex time-dependent patterns; the DMF-corr block is used to correct and modify the decoding results to improve recognition accuracy; Figure 2 The encoder-decoder surrogate model with multiple time resolutions is shown, thereby achieving accurate mapping of multi-source structural responses and vehicle load inputs;

[0053] 2-2) Based on the preprocessed bridge response data, the finite element model of the target bridge is modified to generate simulation response samples for MFNI model training. The mean absolute error loss function is optimized by the gradient back propagation algorithm so that the model can accurately identify the vehicle movement force from the bridge response;

[0054] The generation of simulation response samples includes:

[0055] A1: Modify the finite element model to randomly generate road roughness data;

[0056] A2: Generate a vehicle model with random key parameters to be identified, and use it to generate a large number of simulation training samples.

[0057] Conduct parameter sensitivity analysis to evaluate the test performance of the MFNI model under different vehicle parameter conditions, further optimize the model structure and parameter settings, and improve the robustness and recognition accuracy of the model in actual application scenarios;

[0058] 2-3) Use the trained MFNI model to process the real-time collected bridge response data, extract the vehicle movement force characteristics, and obtain the dynamic force time history of each axle of the vehicle on the bridge;

[0059] 2-4) Use the DMF-corr block to correct the decoder output to ensure the accuracy of the recognition result; through the correction function, the model can more accurately determine the number of vehicle axles and correct the load time history of the non-bridge action time step;

[0060] The load correction function of the DMF-corr block is implemented by adding a ReLU activation function after the output of the fully connected layer. Its specific form is: Among them, z is the output of the neural network before correction.

[0061] 3) Load weighing network deployment phase

[0062] The wheelbase and vehicle speed information are extracted from the vehicle mobility features output by the MFNI model through the time series feature derivation method;

[0063] The time series feature derivation method uses the modified load time history results to solve the wheelbase and vehicle speed information, including:

[0064] B1: Calculate the time step difference τ of different axles acting on the bridge based on the cross-correlation function i , the formula is as follows:

[0065]

[0066] B2: Use the first-order difference to obtain the time steps of the axle on and off the bridge, so as to obtain the average action time Δt of each axle load on the bridge;

[0067] B3: Substitute and solve the following linear non-homogeneous equations to obtain the wheelbase a i And the estimated vehicle speed v:

[0068]

[0069] 4) Vehicle-induced response sample acquisition stage

[0070] The MFNI model is integrated into the bridge health monitoring system to achieve real-time monitoring and identification of vehicle mobility. This method can be used in multiple fields such as bridge structure health monitoring, traffic flow analysis, overload detection, etc., to improve the intelligent level of bridge management and maintenance.

[0071] The MFNI model is used to collect and separate the vehicle-induced response signals in the bridge monitoring signals in real time, denoise the vehicle-induced response signals, and divide the vehicle-induced response samples using the sliding window technology.

[0072] 5) Bridge abnormality diagnosis stage

[0073] Reference Figure 3 , specifically including the following steps:

[0074] 5-1) If Figure 3 As shown in (a) and (b), a PTAD model is constructed based on PatchTST and group convolution, and the PTAD model is trained using vehicle-induced response samples, which is specifically achieved by optimizing the mean square error loss function through gradient back propagation;

[0075] 5-2) Capture the sensitive features of bridge damage hidden in the input vehicle-induced response samples through the trained PTAD model;

[0076] 5-3) Obtain damage indicators through damage-sensitive features, evaluate bridge status based on the damage indicators, and quantify the degree of bridge damage;

[0077] Injury indicators include long-term injury index DI long-term and short-term damage index DI short-termBy calculating the reconstruction error of the test data in the PTAD model, the error vector θ composed of the average REI value of each sensor in the sliding window is obtained, and compared with the reference error vector θ0 in the healthy state, DI is calculated. long-term and DI short-term ;

[0078] The calculation formula is as follows:

[0079]

[0080] The bridge status is evaluated based on the damage index and the degree of bridge damage is quantified, including: statistically analyzing the DIs values ​​obtained under the healthy state to establish short-term and long-term damage thresholds with a confidence level of 95%; and quantifying the damage degree of the bridge based on the long-term damage index DIs. long-term Compare with the preset damage threshold to determine whether the bridge has potential damage; at the same time, regularly analyze the short-term damage index DI short-term , to assess the extent of damage in real time;

[0081] According to DI long-term and DI short-term The calculation results of the indicators are used to quantify the degree of bridge damage and provide a basis for subsequent maintenance decisions.

[0082] In order to verify the effectiveness and effect of the method of the present invention, this embodiment is verified through multiple experiments, and the specific data and analysis are as follows:

[0083] Figure 4 This is the preliminary verification effect diagram of the bridge dynamic weighing system. Figure 4 It can be seen from the results of load weighing the input bridge response using the MFNI model that the static load reference value and the load prediction value are almost consistent, which shows that the proposed MFNI method can be trained using simulation data to complete the vehicle load weighing task.

[0084] Figure 5 This is a preliminary verification effect diagram for long-term abnormal state diagnosis. Figure 5 It can be seen that the DI extracted by the PTAD model long-term The indicator can effectively detect bridge damage by inputting vehicle-induced response samples into the trained PTAD model to extract sensitive features, thereby completing anomaly detection and quantification.

[0085] Figure 6 This is a preliminary verification effect diagram for short-term abnormal state diagnosis. Figure 6 It can be seen that the DI extracted by the PTAD model short-term The indicator can effectively detect bridge damage by inputting vehicle-induced response samples into the trained PTAD model to extract sensitive features, thereby completing anomaly detection and quantification.

Claims

1. A vehicle-bridge parameter synchronous inversion identification method based on a deep time series model, characterized in that: The steps include: S1: Collect bridge response data and preprocess them; S2: construct the MFNI model, including the encoder, decoder, and dynamic moving load correction DMF-corr block; S3: Modify the finite element model of the target bridge based on the preprocessed bridge response data and generate simulation response samples for MFNI model training; S4: Use the trained MFNI model to process the real-time collected bridge response data, extract the vehicle movement force characteristics, and obtain the dynamic force time history of each axle of the vehicle on the bridge; S5: Correct the output of the decoder using the DMF-corr block; S6: Extract wheelbase and vehicle speed information from the vehicle mobility features output by the MFNI model through the time series feature derivation method; S7: The vehicle-induced response signals in the bridge monitoring signals are collected and separated in real time through the MFNI model, the vehicle-induced response signals are denoised, and the sliding window technology is used to divide the vehicle-induced response samples; S8: Build a PTAD model based on PatchTST and group convolution, and train the PTAD model using vehicle-induced response samples; S9: Capture the damage-sensitive features in vehicle-induced response samples through the trained PTAD model; S10: The damage index is obtained through the damage sensitive characteristics, the bridge status is evaluated according to the damage index, and the degree of bridge damage is quantified.

2. According to claim 1, a vehicle-bridge parameter synchronous inversion identification method based on a deep time series model is characterized in that: In step S2, the encoder adopts an extended causal convolution structure to capture local feature information and interactive dependency patterns in the input data; the decoder uses a recurrent neural network model with a stacked extended structure to capture complex time dependency patterns; and the DMF-corr block is used to correct and modify the decoding results to improve recognition accuracy.

3. The method for synchronous inversion and identification of vehicle-bridge parameters based on a deep time series model according to claim 2 is characterized in that: The load correction function of the DMF-corr block in step S2 is implemented by adding a ReLU activation function after the output of the fully connected layer, and its specific form is: Among them, z is the output of the neural network before correction.

4. The method for synchronous inversion and identification of vehicle-bridge parameters based on a deep time series model according to claim 1 is characterized in that: The generation of the simulation response sample in step S3 includes: A1: Modify the finite element model to randomly generate road roughness data; A2: Generate a vehicle model with some random key parameters to be identified and use it to generate simulation training samples.

5. The method for synchronous inversion and identification of vehicle-bridge parameters based on a deep time series model according to claim 1 is characterized in that: The time series feature derivation method in step S6 uses the corrected load time history results to solve the wheelbase and vehicle speed information, specifically including: B1: Calculate the time step difference τ of different axles acting on the bridge based on the cross-correlation function i , the formula is as follows: B2: Use the first-order difference to obtain the time steps of the axle on and off the bridge, so as to obtain the average action time Δt of each axle load on the bridge; B3: Substitute and solve the following linear non-homogeneous equations to obtain the wheelbase a i And the estimated vehicle speed v:

6. The method for synchronous inversion and identification of vehicle-bridge parameters based on a deep time series model according to claim 1 is characterized in that: The damage index in step S10 includes a long-term damage index DI long-term and short-term damage index DI short-term By calculating the reconstruction error of the test data in the PTAD model, the error vector θ composed of the average REI value of each sensor in the sliding window is obtained, and compared with the reference error vector θ0 in the healthy state, DI is calculated. long-term and DI short-term ; The calculation formula is as follows:

7. The method for synchronous inversion and identification of vehicle-bridge parameters based on a deep time series model according to claim 1 is characterized in that: In the step S10, the bridge state is evaluated according to the damage index to quantify the degree of bridge damage, specifically including: statistically analyzing the DIs values ​​obtained under the healthy state to establish short-term and long-term damage thresholds with a confidence level of 95% respectively; According to the long-term injury index DI long-term Compare with the preset damage threshold to determine whether the bridge has potential damage; At the same time, regularly analyze the short-term damage index DI short-term , to assess the extent of damage in real time; According to DI long-term and DI short-term The results of the bridge damage are used to quantify the extent of damage and provide a basis for subsequent maintenance decisions.

Images