TUNNEL DEFECT DETECTION AND MANAGEMENT SYSTEM BASED ON VIBRATION SIGNAL OF MOVING TRAIN

Abstract

A tunnel defect detection and management system based on a vibration signal of a moving train. This system identifies the defects in a subway tunnel structure and soil behind a wall through the acquisition, transmission and analysis of an on-board acceleration signal. A signal acquisition sensor is mounted on the moving train. A signal acquisition module and a signal transmission system are mounted in the train to preprocess and compress the signal. A data processing and analysis server performs data analysis to quickly identify the defects of the tunnel and the auxiliary structure thereof, and determine a defect location and type. A tunnel management platform releases real-time detection information and health status of the tunnel, alarms for the defects, and releases the defect data to relevant personnel to take measures.

Claims

1. (canceled)

2. A tunnel defect detection and management system based on a vibration signal of a moving train, comprising: sensors mounted on a train to acquire a vibration signal of a train in service as the train passes through a tunnel; a signal transmission system on the train including a data acquisition module, a data processing module including a processor and a memory, and a wireless transmission module, the data acquisition module receiving the vibration signal in real-time, the data processing module encoding the vibration signal for transmission, the wireless transmission module wirelessly transmitting the vibration signal to a server over a network; and the server including a processor and a non-transitory storage medium including a data analysis system and a tunnel health management platform accessible via a mobile device, the data analysis system including software analyzing and processing the vibration signal to identify a defect of a tunnel and to determine defect data including a type and a location of the defect for display via the tunnel health management platform.

3. The system of claim 2, wherein the software of the data analysis system further decodes and decompresses the vibration signal data to obtain original acceleration, speed and location data X(t,a.sub.x,a.sub.y,a.sub.z,v,s), wherein t represents a time, v represents a speed, s represents a location, a.sub.x represents an X-axis acceleration, a.sub.y represents a Y-axis acceleration, and a.sub.z represents a Z-axis acceleration.

4. The system of claim 3, wherein the software of the data analysis system further denoises and enhances a signal-to-noise ratio (SNR) of the vibration signal.

5. The system of claim 4, wherein the software of the data analysis system further performs dimensionality reduction to construct a feature vector F(t,α) as a sample set of machine learning (ML), wherein α represents a feature vector after dimensionality reduction.

6. The system of claim 5, wherein the software of the data analysis system includes a cyclic neural network classifier to determine the defect, and a decision tree classifier to determine the location and the type of the defect and a defect magnitude.

7. The system of claim 6, wherein the software of the data analysis system further supplements newly acquired data to a sample set to continue to train the cyclic neural network classifier and the decision tree classifier, so as to continuously improve accuracy of the cyclic neural network classifier and the decision tree classifier.

8. The system of claim 2, wherein the sensors include a plurality of acceleration sensors, speed sensors and positioning sensors.

9. The system of claim 8, wherein the acceleration sensors and the speed sensors are mounted on an axle, a bogie and in a carriage of the train.

10. The system of claim 8, wherein the sensors are fixed to the train by a magnetic support.

11. The system of claim 2, wherein the sensors are wireless sensors with a sampling frequency of 2 kHz.

12. The system of claim 2, wherein the network is a 5G network.

13. A tunnel defect detection method, comprising steps of: acquiring, with sensors mounted on a train in service, a vibration signal of the train as the train passes through a tunnel; transmitting with a wireless transmission module the vibration signal to a server through a network; analyzing and processing the vibration signal by the server to identify a tunnel defect and determine a type and a location of the tunnel defect; and outputting defect data including the type and the location of the tunnel defect to a tunnel health management platform accessible via a mobile device.

14. The method of claim 13, wherein analyzing and processing the vibration signal includes decoding and decompressing the vibration signal data to obtain original acceleration, speed and location data X(t,a.sub.x,a.sub.y,a.sub.z,v,s), wherein t represents a time, v represents a speed, s represents a location, a.sub.x represents an X-axis acceleration, a.sub.y represents a Y-axis acceleration, and a.sub.z represents a Z-axis acceleration.

15. The method of claim 14, wherein analyzing and processing the vibration signal includes denoising and enhancing a signal-to-noise ratio (SNR) of the vibration signal.

16. The method of claim 15, wherein analyzing and processing the vibration signal includes performing dimensionality reduction to construct a feature vector F(t,α) as a sample set of machine learning (ML), wherein α represents a feature vector after dimensionality reduction.

17. The method of claim 16, wherein analyzing and processing the vibration signal includes determining, by a cyclic neural network classifier, whether there is the tunnel defect and determining, by a decision tree classifier, the location and type of the tunnel defect.

18. The method of claim 17, further comprising training the cyclic neural network classifier through an initial sample such that the cyclic neural network classifier is able to determine a defect, and training the decision tree classifier such that the decision tree classifier is able to determine a defect location, a defect type and a defect magnitude.

19. The method of claim 18, further comprising supplementing newly acquired data to a sample set to continue to train the cyclic neural network classifier and the decision tree classifier, so as to continuously improve accuracy of the cyclic neural network classifier and the decision tree classifier.

20. The method of claim 13, wherein the sensors include a plurality of acceleration sensors, speed sensors and positioning sensors.

21. The method of claim 20, wherein the acceleration sensors and the speed sensors are mounted on an axle, a bogie and in a carriage of the train.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0025] FIG. 1 is a structural diagram of a tunnel defect detection system based on a vibration signal of a moving train.

[0026] FIG. 2 is a schematic diagram of the tunnel defect detection system based on a vibration signal of a moving train.

[0027] FIG. 3 shows a machine learning (ML)-based vibration data analysis algorithm.

REFERENCE NUMERALS

[0028] 1. train; 2. bogie; 3. wheelset; 4. sensor; 5. signal transmission system; 6. data processing system; 7. mobile terminal; 8. personal computer (PC) terminal; 9. track; 10. floating track slab; 11. tunnel; 12. soil; 13. track defect; 14. track slab defect; 15. tunnel defect; 16. soil discontinuity defect behind the segment.

DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] The present disclosure is further described below with reference to the embodiments and accompanying drawings.

Embodiment 1

[0030] As shown in FIG. 1, the present disclosure provides a tunnel defect detection and management method based on a vibration signal of a moving train. FIG. 2 shows a flowchart of the method. Sensors 4 are used to acquire vibration signals of a train 1 in service and a bogie 2 and a wheelset 3 thereof, and the train vibration data is wirelessly transmitted to an on-board signal transmission system 5 for preprocessing and compression. Then the data is transmitted to a cloud server through a 5G network. The data is analyzed and processed to identify defects of a tunnel 11 and an auxiliary structure thereof and determine an approximate location of the defect. Then the defect data is released on a management cloud platform to provide a reference for real-time understanding of the health status of the tunnel. The present disclosure greatly improves the tunnel detection efficiency and reduces the detection cost and operation risk.

[0031] As shown in FIGS. 1 and 2, the shield tunnel defect detection and management method uses four subsystems, namely a signal acquisition system 20, a signal transmission system 5, a data processing system 6 and a tunnel health management platform 30.

[0032] When the subway train 1 in service runs in the shield tunnel, the signal acquisition system 20 forms a coupled vibration system with a tunnel structure and a stratum, and uses sensors 4 mounted on the train 1 to acquire a vibration signal transmitted to the train 1. The sensor 4 include a plurality of acceleration sensors, speed sensors and positioning sensors; the acceleration sensors and the speed sensors are mounted on an axle of the wheelset 3, the bogie 2 and in a carriage, and are fixed by a magnetic base and a strapping; the positioning sensors are mounted in the carriage, and are fixed by a magnetic support. The sensors 4 are wireless sensors with a sampling frequency of 2 kHz. The sensor has a built-in rechargeable battery, which can be recycled and has sufficient power to support real-time monitoring for a long time. The sensor automatically sleeps to save power when the subway train 1 is out of service at night.

[0033] The sensors 4 send the data to an acquisition module 21 in the carriage in real time after the data is acquired. Then the acquisition module 21 transmits the data to the server for analysis. The signal transmission system 5 includes a data receiving module 21, a data processing module 22, a data wireless transmission module 23 and a power supply module. The signal transmission system 5 is packaged in a box and can be mounted under a seat in a carriage of the same train as the sensor 4 to avoid affecting a passenger. The data receiving module 21 receives the measurement data transmitted by the sensor 4 in real time. The data processing module 22 includes a microprocessor, a memory and an encoder. The data processing module 22 caches certain data, preliminarily organizes and compresses the data, and re-encodes the data. The data transmission module 22 uploads the encoded data to the server through a 5G network or the Internet for data processing and analysis. This part can be powered directly from the carriage, or by a storage battery if there is no available power source.

[0034] The data is transmitted through a network to the data processing system 6 for analysis. The data processing system 6 includes a high-performance computing processor, an ultra-large-capacity memory, a network module, a power supply module and analysis software. The network module provides a stable network speed and as much bandwidth as possible, and stably receives data transmitted through the Internet. Then a plurality of central processing units (CPUs) and graphics processing unit (GPUs) perform parallel computing and quickly process a large amount of data, analyze whether there are defects in the tunnel and its auxiliary structure or the soil, and assess the health of the subway tunnel. The defects include those occurring in a track 9, a floating track slab 10, the tunnel 11 and the soil 12. The main types of defects include but are not limited to track defect 13, track slab defect 14, track fasteners defect, steel spring defect, tunnel lining crack or concrete spalling, and soil discontinuity defect 16 behind lining wall. The analyzed data is stored in the ultra-large-capacity memory, which can be stored stably for a long time. The analysis result is released on the subway tunnel health management platform 30 to inform relevant personnel 32. Meanwhile, the processing system should have a stable power supply to avoid data loss caused by sudden power failure.

[0035] As shown in FIG. 3, in the analysis process, existing professional data analysis software is used to decode and decompress the train vibration signal data, further denoise, enhance a signal-to-noise ratio (SNR), and then perform modal analysis and wavelet transform (steps 40, 41, 42). In addition, a machine learning (ML) method such as a convolutional neural network (CNN) is used to quickly identify the tunnel defect 15 (crack, concrete spalling), initially determine the location of the tunnel defect 15, and assess the health of the tunnel (steps 43, 44, 45, 46, 48). Finally, the assessment result is released on the tunnel health management platform 30 (step 47). Specifically, a cyclic neural network classifier and a decision tree classifier are introduced. The cyclic neural network classifier is trained through an initial sample (43,44) such that the cyclic neural network classifier is able to determine a defect (45), and the decision tree classifier is trained (48) such that the decision tree classifier is able to determine a defect location, a defect type and a defect magnitude (46). The cyclic neural network is used to quickly determine whether there is a tunnel defect (45). If yes, the decision tree classifier is used to initially determine the location and type of the tunnel defect (46) and output the location and type of the tunnel defect to assess the health of the tunnel (47). Further, a CNN classifier and a decision tree classifier are introduced. The CNN classifier is trained through an initial sample (43,44) such that the CNN classifier is able to determine a defect (45), and the decision tree classifier is trained such that the decision tree classifier is able to determine a defect location, a defect type and a defect magnitude (46). The CNN classifier is used to quickly determine whether there is a tunnel defect (45). If yes, the decision tree classifier is used to initially determine the location and type of the tunnel defect (46) and output the location and type of the tunnel defect to assess the health of the tunnel (47). The tunnel health management platform 30 releases information of the tunnel and its auxiliary structures, such as health status, defect location and defect assessment. It can be installed as a mobile application (APP) at a mobile terminal 7 or a personal computer (PC) terminal 8 of relevant personnel 32 in a subway operation and maintenance company for real-time understanding of the tunnel health status. When a heavy defect of the tunnel and its auxiliary structure is found, the relevant personnel 32 can take measures in time and use more sophisticated and professional testing instruments or methods to detect the tunnel. This avoids threats to the safety of people's property due to the inability to detect the tunnel defect in time, reduces the cost of defect detection, and improves the efficiency of operation and maintenance.