A method for constructing a subway tunnel vehicle-induced vibration data set

Abstract

A method for constructing a subway tunnel vehicle-induced vibration dataset. First, based on field monitoring equipment and inspection, the operation subway tunnel vibration data and the corresponding structure state are obtained to construct the actual working condition-measured vibration dataset. The numerical simulation calculation method is used to obtain the subway tunnel simulation vibration response data to construct the virtual working condition-simulation vibration dataset. Then, the signal decomposition method is used to obtain the measured and simulation vibration sub-signals, and based on the Geraud similarity coefficient of each sub-signal and the influence parameter, the data sensitive sub-signal is determined to form a sensitive sub-signal set. Then, based on the conditional generative adversarial network, different working conditions are taken as conditions to build a vibration data sensitive sub-signal generation model, and the simulation sensitive sub-signal is changed into a sensitive sub-signal closer to the measured data. Finally, through signal reconstruction, the true sensitive sub-signal is changed into true vibration data, and is fused with the measured vibration data to construct a subway tunnel vehicle-induced vibration dataset.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of urban rail transit intelligent operation and maintenance, and relates to a subway tunnel vehicle-induced vibration data set construction method based on field monitoring and numerical simulation fusion. BACKGROUND

[0002] Subway tunnel vibration monitoring is an important control means for the safety state of subway operation. Abnormal conditions of tunnel structure such as segment damage, leakage, and track bed void will change the structural stiffness and support conditions, leading to changes in the vibration characteristics of the train-track bed-tunnel-soil system and causing changes in the tunnel vehicle-induced vibration response. Accelerometers, distributed vibration optical fibers and other monitoring technologies have been used in many cities to carry out long-term monitoring of structure vibration caused by subway tunnel driving. With the accumulation of vehicle-induced vibration monitoring data, the evaluation of the safety state of urban subway structures based on vibration signal characteristics has become an important direction for intelligent operation and maintenance of subways. The prerequisite for achieving this goal is to construct a high-quality subway tunnel vehicle-induced vibration data set.

[0003] However, the field monitoring data has problems such as asymmetry, high redundancy, and difficulty in labeling. The asymmetry refers to the fact that most actual operation tunnels are in a normal state, and the occurrence of abnormal conditions is relatively rare, resulting in an asymmetric distribution of data corresponding to different states in the measured data set. High redundancy refers to the fact that invalid data due to subway running intervals, repeated data caused by no significant changes in train operation and structure state, and other factors significantly increase the redundancy of long-term monitoring data. Labeling difficulty refers to the fact that the actual safety state evaluation of the structure is currently mainly based on manual inspection, which is time-consuming and labor-intensive. The evaluation of the structure is mainly qualitative, and it is difficult to quantitatively describe the hidden safety properties such as track bed void, making it difficult to match the field vibration measured data with the actual safety state of the structure. Therefore, it is difficult to construct a data foundation for intelligent operation and maintenance of subway structures solely relying on field monitoring.

[0004] Numerical simulation has the advantage of controllable boundary conditions, but the reliability of the calculation results depends on the checking and correction of measured data. Simply relying on numerical simulation methods to construct a subway tunnel vehicle-induced vibration data set has problems such as poor data reliability and incomplete matching with actual conditions.

[0005] In view of the above engineering practical problems and the advantages and disadvantages of the two technical means in data set construction, the present application proposes a subway tunnel vehicle-induced vibration data set construction method based on the fusion of the two. SUMMARY

[0006] In view of the problems in the prior art, the present application proposes a subway tunnel vehicle-induced vibration data set construction method based on fusion of field monitoring and numerical simulation. The method constructs a virtual working condition-simulation vibration data set based on numerical simulation calculation, obtains sensitive sub-signals of simulation vibration data through signal decomposition method and correlation determination, and based on conditional generative adversarial network, takes different working conditions as conditions, changes the simulation sensitive sub-signals into sensitive sub-signals closer to the measured data, finally through signal reconstruction, changes the true sensitive sub-signals into true vibration data, and fuses with the measured vibration data to construct a subway tunnel vibration data set.

[0007] In order to achieve the above-mentioned target, the present application adopts the following technical solutions:

[0008] Step 1. Using monitoring equipment such as accelerometers, vibration optical fiber sensors, etc., the subway tunnel vehicle-induced dynamic response in actual operation is monitored in real time, and combined with the field inspection results, the measured vibration data of the operating subway tunnel under different subway tunnel vibration influence parameters is obtained. On this basis, the actual working condition-measured vibration data set is constructed.

[0009] Step 2. According to the field structure design data and investigation, based on the numerical simulation model, the subway tunnel simulation vibration response data under the virtual influence parameters are obtained. The different working condition attributes and the corresponding simulation vibration response data are integrated to construct the virtual working condition-simulation vibration data set.

[0010] Step 3. Based on the wavelet decomposition method, the measured vibration data and the simulation vibration data are decomposed to obtain the measured and simulated vibration sub-signals. The Geraud similarity coefficient of each sub-signal and the influence parameter is calculated, and the relationship between the correlation coefficient and the threshold value is determined to determine the data sensitive sub-signals and form the sensitive sub-signal set.

[0011] Step 4. Based on the conditional generative adversarial network, the simulation vibration data sensitive sub-signal set is input into the generator. The generator makes distribution assumption and distribution parameter learning on the vibration data, excavates its potential distribution, and generates new labeled generated signals according to the existing data and its distribution characteristics. Then, the measured sensitive sub-signals and the simulation sensitive sub-signals are input into the discriminator at the same time, and the discriminator determines whether the input data comes from the measured data or the generated data.

[0012] Step 5. Based on the wavelet inverse transform method, the generated sub-signals and the corresponding wavelet coefficients are changed into true vibration data through signal reconstruction, and are fused with the measured vibration data to construct a subway tunnel vibration data set.

[0013] The present application has the following advantages:

[0014] As described above, the present invention describes a method for constructing a subway tunnel vibration dataset based on the fusion of field monitoring and numerical simulation. This method addresses the issues of asymmetry, high redundancy, and difficulty labeling subway tunnel vibration data. It proposes a technical approach to supplementing data from different operating conditions with numerical simulation. Sensitive sub-signals are selected using signal decomposition and correlation analysis. A conditional generative adversarial network (CGN) is used to transform the simulated sensitive sub-signals into those closer to the measured data. Signal reconstruction is then used to obtain the subway tunnel vibration dataset. This method combines the advantages of CGN in computational efficiency and data realism with the advantages of numerical simulation in terms of clear physical and mechanical concepts and high precision. This method constructs a high-quality, rich-condition subway tunnel vibration dataset, providing data support for research on subway tunnel structural status identification and operation and maintenance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a framework diagram of a method for constructing a subway tunnel vehicle-induced vibration dataset based on the fusion of field monitoring and numerical simulation in an embodiment of the present invention.

[0016] Figure 2 This is a diagram of the conditional generative adversarial network structure in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The purpose of the present invention is to construct a subway tunnel vibration dataset by integrating deep learning with numerical simulation, obtain subway tunnel vibration data under different working conditions through numerical simulation, obtain sensitive sub-signals of simulated vibration data by using signal decomposition and correlation judgment, and use conditional generative adversarial networks to transform the simulated sensitive sub-signals into sensitive sub-signals closer to measured data, thereby obtaining a subway tunnel vibration dataset to overcome the problem of imbalanced subway tunnel vibration data and the difficulty in obtaining tunnel vibration data under different subway tunnel vibration influencing parameters.

[0018] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0019] like Figure 1 As shown in FIG, a method for constructing a subway tunnel vibration dataset based on the integration of field monitoring and numerical simulation includes the following steps:

[0020] Step 1. Use monitoring equipment such as accelerometers and vibration fiber optic sensors to conduct real-time monitoring of the actuating force response of subway tunnel vehicles in actual operation. Combined with the results of on-site inspections, the actual measured vibration data A of the operating subway tunnel under different subway tunnel vibration influencing parameters are obtained. mon (t). On this basis, the actual working condition-measured vibration data set is constructed Dataset Each group of data is a time series obtained by continuous collection for t seconds at a preset sampling frequency of k Hz, with a total of k*t data points; each group of data is labeled with the corresponding {train speed V train , structure state C track , and underlying soil type T soil .

[0021] The measured data should involve as many subway tunnel working conditions as possible, including different train speeds V train , structure states C track , and soil types T soil . The measured data will serve as an important basis for numerical simulation model verification and conditional generative adversarial network construction.

[0022] Step 2. Based on the field structure design data and investigation, and based on the numerical simulation model, obtain the subway tunnel simulation vibration response data A sim (t) = M(V train , C track , T soil , t) under the virtual subway tunnel vibration influence parameters.

[0023] Integrate different working condition attributes and corresponding simulation vibration response data to construct a virtual working condition-simulation vibration data set The data set format is consistent with the actual working condition-measured vibration data set .

[0024] The numerical simulation model adopts a vehicle-track-tunnel-soil coupling subway tunnel dynamic numerical simulation model, including vehicles, tracks, tunnels, and soil, so as to consider different vehicle speeds, track types, soil types, and structure states. The model needs to be verified and optimized through measured data to ensure the reliability of the calculation results.

[0025] Step 3. Based on the wavelet decomposition method, decompose the measured vibration data A mon (t) and the simulation vibration data A sim (t), and based on the decomposed wavelet coefficients a mon , a sim , the sub-signals W mon , and W sim . By calculating the Jacard similarity coefficients of each sub-signal W mon , W sim , train speed V train , structure state C track , and underlying soil type T soil , the sub-signals with a correlation coefficient greater than r are considered sensitive. The value of r depends on the long-term data statistics of each subway tunnel.

[0026] The sensitive sub-signal sets of the measured and simulation vibration data are:

[0027]

[0028]

[0029] Since the main frequency component of subway tunnel is in the range of 0-200 Hz, different working conditions lead to changes in signal components of different frequency bands. To accurately express the vibration characteristics of subway tunnels under different working conditions, the wavelet decomposition method is used to obtain the approximation coefficients and detail coefficients by selecting a 6-layer wavelet decomposition. The wavelet coefficients greater than 200 Hz are discarded, and the remaining coefficients and detail coefficients can accurately and efficiently reflect the characteristics of subway tunnels.

[0030] Step 4. Based on the conditional generative adversarial network, the simulated vibration data sensitive sub-signal set L sim is generated. train , the structure state C track and the soil type T soil are input into the generator G. The generator G performs distribution assumption and distribution parameter learning on the vibration data, excavates its potential distribution, and generates new labeled generated signals L gen according to the existing data and their distribution characteristics. Figure 2 As shown in , the train speed V train , the structure state C track and the soil type T soil are used as labels and conditions in the conditional generative adversarial network. The measured vibration data sensitive sub-signal set L mon and the simulated vibration data sensitive sub-signal set L sim are input into the discriminator Z, which determines whether the input data comes from the measured data or the generated data, and finally obtains the labeled generated signals L gen that meet the requirements.

[0031] Step 5. Based on the wavelet inverse transform method, the generated sub-signal L gen and the corresponding wavelet coefficients are used for signal reconstruction to generate the true vibration data A gen (t).

[0032]

[0033] The reconstructed data A gen (t) and the measured vibration data A mon (t) have the same time length and sampling frequency. The subway tunnel vibration data set S A constructed by the fusion of the two includes n groups of data, each group of data has k*t data points; each group of data corresponds to {train speed V train , structure state C track and underlying soil type Tsoil} is a data tag; is a wavelet inverse transform function.

[0034] The method fuses on-site monitoring and numerical simulation, obtains virtual vibration data under a virtual working condition based on numerical simulation, and obtains virtual working condition-realistic vibration data similar to real subway tunnel vibration data distribution based on a conditional generative adversarial network, effectively solves the problem that rich working condition subway tunnel vibration data is difficult to obtain, and provides data support for subway tunnel structure state identification and operation and maintenance management research.

[0035] Compared with the prior art, the innovation of the present application is that, in view of the technical problem that subway tunnel vibration data is unbalanced and it is difficult to obtain tunnel vibration data under different subway tunnel vibration influence parameters, numerical simulation is used to supplement different working condition data, signal decomposition and correlation determination are used to obtain sensitive sub-signals of vibration data, a conditional generative adversarial network is used to convert simulated sub-signals into realistic sub-signals close to measured data, and signal reconstruction is used to obtain a subway tunnel vibration data set. The method combines the advantages of the conditional generative adversarial network in computational efficiency and data realism with the advantages of numerical simulation in clear physical and mechanical concepts and high precision, realizes the construction of a high-quality, rich-working-condition subway tunnel vibration data set, and can provide data support for subway tunnel structure state identification and operation and maintenance management research.

[0036] Of course, the above description is only for the preferred embodiments of the present application, and the present application is not limited to the above-described embodiments. It should be noted that any person skilled in the art can make all equivalent substitutions and obvious modifications under the teaching of the present application, and all such substitutions and modifications fall within the scope of the present application, and should be protected by the present application.

Claims

1. A method for constructing a subway tunnel vibration dataset based on the integration of field monitoring and numerical simulation, characterized in that: The steps include: Step 1. Use accelerometers and vibration fiber optic sensor monitoring equipment to monitor the actuating force response of subway tunnel vehicles in actual operation in real time, and combine the on-site inspection results to obtain the different train speeds V train , structural state C track and underlying soil type T soil Vibration measurement data of subway tunnel under operation mon (t); On this basis, the actual working condition-measured vibration data set is constructed Step 2. Based on the on-site structural design data and investigation, and based on the numerical simulation model, obtain the subway tunnel simulation vibration response data under the virtual subway tunnel vibration influencing parameters: A sim (t)=M(V train ,C track ,T soil ,t) (1) Integrate different working condition attributes and corresponding simulated vibration response data to build a virtual working condition-simulated vibration data set Dataset format and actual working conditions - measured vibration dataset consistent; Step 3. Decompose the measured vibration data A based on the wavelet decomposition method mon (t) and simulated vibration data A sim (t); based on the decomposed wavelet coefficient a mon , a sim With the sub-signal W mon and W sim ; The sensitive sub-signal sets of measured and simulated vibration data are: Step 4. Based on the conditional generative adversarial network, the simulated vibration data sensitive sub-signal set L sim The corresponding train speed V train , structural state C track and soil type T soil Input generator G; Step 5. Based on the inverse wavelet transform method, the generated sub-signal L gen The signal is reconstructed with the corresponding wavelet coefficients to generate the true vibration data A gen (t); 2. The method for constructing a subway tunnel vibration dataset based on the fusion of field monitoring and numerical simulation according to claim 1 is characterized in that: Step 3: In step 3, the wavelet decomposition adopts 6-layer wavelet decomposition, retains the wavelet coefficients with a frequency range of 0 to 200 Hz, and discards the wavelet coefficients greater than 200 Hz.

3. The method for constructing a subway tunnel vibration dataset based on the fusion of field monitoring and numerical simulation according to claim 1 is characterized in that: Step 4: The generator G makes distribution assumptions and distribution parameter learning on the vibration data, mines its potential distribution, and generates a new labeled generated signal L based on the existing data and its distribution characteristics. gen ; Afterwards, the measured vibration data sensitive sub-signal set L mon The sensitive sub-signal set L of the simulated vibration data sim At the same time, the discriminator Z is input, and the discriminator Z determines whether the input data comes from measured data or generated data, and finally obtains the generated signal L with the label that meets the requirements gen .

4. The method for constructing a subway tunnel vibration dataset based on the fusion of field monitoring and numerical simulation according to claim 1 is characterized in that: Step 5: The reconstructed data A gen (t) and the measured vibration data A mon (t) duration is consistent with the sampling frequency; the subway tunnel vibration dataset S is constructed by fusion of the two A , including n groups of data, each group of data has k*t data points; each group of data corresponds to {train speed V train , structural state C track and underlying soil type T soil } is the data label, is the inverse wavelet transform function.

Images