CN211731391U - An intelligent online detection device for track surface defects - Google Patents

Abstract

The utility model relates to an intelligent online detection device for track surface defects. The device is composed of a track surface accurate defect detection system, a track surface defect detection model online update system, a train set data exchange system, a track defect location system, an abnormality processing system and a power supply system. . The utility model combines the research of convolutional neural network to realize non-destructive and non-contact detection. The collected track surface image is made into a track data set by a series of data image processing methods and sent to the neural network model for intelligent detection of track surface defects; the involved shock absorption device is used to prevent the collected image from being blurred due to the bumping and shaking of the train. , to achieve high-precision detection and reduce manual intervention.

Description

An intelligent online detection device for track surface defects

technical field

The utility model belongs to the technical field of intelligent online detection of track surface defects, in particular to an intelligent online detection device for track surface defects.

Background technique

At present, railway lines are the most important infrastructure for running vehicles. Due to the harsh natural environment and constant train loads, the surface state of the train tracks is always changing, and deformation and damage are constantly occurring. In order to ensure the safety of passengers, the health status of the track must be checked and confirmed frequently. The traditional track inspection methods are mainly manual inspection method, eddy current coil inspection and ultrasonic inspection. The manual detection method is mainly completed by the patrolling workers, and its detection efficiency is low, and it is easily affected by environmental factors and considered factors; ultrasonic detection and eddy current coil detection will contact with the surface defects of the rail, and physical and chemical changes may occur, and further expansion defective area.

SUMMARY OF THE INVENTION

The purpose of the present utility model is to provide an intelligent online detection device for track surface defects in view of the above-mentioned deficiencies of the prior art. It is used to detect the defect state of the track surface in real time, thereby improving the safety of the railway and reducing the maintenance cost of the railway; through the positioning system of the utility model, the precise position of the track surface defect can be sent to the railway center to ensure that the damaged track can be obtained. Repair in time.

The purpose of this utility model is to realize through the following technical solutions:

An intelligent online detection device for track surface defects, characterized in that it is composed of a track surface accurate defect detection system, a track surface defect detection model online update system, a train set data exchange system, a track defect location system, an exception handling system and a power supply system;

Wherein, the track surface defect detection system includes an image acquisition subsystem and a track detection and classification subsystem. The image acquisition system consists of a base 17 , a shock absorber 20 connected to the base 17 , a calibration component 40 , a light source 50 and an image The collector 30 is composed; the track classification detection subsystem is composed of an image processor 60 connected with the image collector 30, a track identification module 70 and a track classification module 80;

The track surface defect detection model online update system is composed of a human-computer interaction interface 90, an image labeling and neural network training center 100, and a model parameter updating module 110 connected to the track identification module 70 and the track classification module 80. The human-computer interaction interface 90 The neural network training center 100 is connected to the model parameter updating module 110 through image marking;

The train set data interaction system is composed of a signal transceiver module 120, a data synchronization module 130, a data standardization module 140 and a central database 150. The signal transceiver module 120 is connected to the central database 150 through the data synchronization module 130 and the data standardization module 140;

The track defect locating system consists of a relative position estimation module 170 connected with the speed/acceleration sensor 160, a GPS/Beidou 180, an absolute position estimation module 190 connected with the GPS/Beidou 180, and a relative position estimation module 170 and an absolute position The fusion positioning module 200 connected to the calculation module 190 is composed of;

The anomaly handling system is composed of an anomaly detection module 250, an anomaly classification module 260, an anomaly alarm device 270, and an anomaly clearing device 280 connected to each system. device 270 is connected;

The power supply system is composed of a train power system module 210, a conventional power supply module 220, an emergency power supply module 230 and a power supply adapter module 240. The train power system module 210 passes through the conventional power supply module 220, the emergency power supply module 230 and the power supply adapter module 240 respectively. connected.

Further, the shock absorbing device 20 is composed of a protective cover 1 fixed to the train, a hydraulic damper cylinder I2, a longitudinal shock absorbing spring 13, a piston rod 14, a camera 5, a camera fixing device 6, and a transverse shock absorbing spring. I9. The light source fixing part 10, the light source 11, the longitudinal damping spring II12, the transverse damping spring II13, and the axial damping spring 15 are composed;

The camera 5 is fixed in the camera fixing device 6, and the camera fixing device 6 is connected with the protective cover 1 through the connecting member 7, the pin member 8 and the hydraulic damper; the light source 11 is connected with the protective cover 1 through the light source fixing member 10. fixed.

Compared with the prior art, the beneficial effects of the present utility model are:

1. Since the data set used in the neural network training stage of the present utility model contains various defect morphological targets under different environmental conditions, the training data is more comprehensive and balanced, so the algorithm has strong robustness, and has certain characteristics. Anti-interference ability; 2. A system for real-time online updating of the neural network model is designed according to the detection results of the present utility model, which improves the effectiveness of the neural network model; The acquired image is blurred and the influence of the detection accuracy is reduced. A shock absorption device for the image acquisition system is designed; 4. The utility model comprehensively improves the detection accuracy and recall rate, and can greatly reduce the cost of railway maintenance and improve the detection efficiency.

Description of drawings

Fig. 1 is the embodiment block diagram of a kind of track surface defect intelligent online detection method, device and system of the present invention;

2 is a schematic flowchart of an embodiment of an intelligent online detection method for track surface defects of the present invention;

3 is a schematic flowchart of a preferred embodiment of the execution process of step S110;

4 is a schematic flowchart of a preferred embodiment of the execution process of step S120;

5 is a flowchart of a preliminary procedure for removing pseudo-edges in an embodiment of the present invention;

6 is a schematic flowchart of a preferred embodiment of the execution process of step 140;

FIG. 7 is a schematic diagram of the real-time update flow of the neural network model in the embodiment of the present utility model

8 and 9 are schematic structural diagrams of the shock absorbing device of the system in the embodiment of the present invention.

In the figure, 1. Protective cover 2. Hydraulic damper cylinder I 3. Longitudinal damping spring I 4. Piston rod I 5. Camera 6. Camera fixing device 7. Connecting piece 8. Pin piece 9. Lateral reduction Shock spring I 10. Light source fixture 11. Light source 12. Longitudinal damping spring Ⅱ 13. Transverse damping spring Ⅱ 14. Hydraulic damper hydraulic cylinder Ⅱ 15. Axial damping spring 16. Piston rod Ⅱ 17. Base 20. Damping device 30. Image collector 40. Calibration component 50. Light source 60. Image processor 70. Track identification module 80. Track classification module 90. Machine interaction interface 100. Neural network training center 110. Model parameter update module 120 . Signal transceiver module 130. Data synchronization module 140. Data standardization module 150. Central database 160. Speed/acceleration sensor 170. Relative position estimation module 180. GPS/Beidou 190. Absolute position estimation module 200. Fusion positioning module 210. Train power System module 220. Conventional power supply module 230. Emergency power supply module 240. Power supply adapter module 250. Abnormal detection module 260. Abnormal classification module 270. Abnormal alarm 280. Abnormal clearing.

Detailed ways

Embodiments of the present invention will be described in detail below, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to like elements or elements having the same function throughout. The embodiments described below with reference to the accompanying drawings are exemplary, and are intended to be used to explain the present invention, but should not be construed as a limitation of the present invention. Based on the embodiments of the present invention, those skilled in the art will Other embodiments obtained under the premise of no creative work all belong to the protection scope of the present invention.

In order to solve the above technical problems, the utility model provides an intelligent online device for track surface defects, which is mainly used for real-time detection of the defect state of the track surface, thereby improving the safety of the railway and reducing the maintenance cost of the railway. Through the positioning system of the utility model, the precise position of the track surface defect can be sent to the railway center, so as to ensure that the damaged track can be repaired in time.

As shown in Figure 1, the intelligent online detection device for track surface defects of the present utility model is composed of a track surface accurate defect detection system, a track surface defect detection model online update system, a train set data exchange system, a track defect location system, an abnormal processing system and a power supply system. System Components. Wherein, the track surface defect detection system includes an image acquisition subsystem and a track detection and classification subsystem, and the image acquisition system includes a base 17 , a damping device 20 connected to the base 17 , a calibration component 40 , a light source 50 and an image collector 30; the track classification detection subsystem includes an image processor 60 connected to the image collector 30, a track identification module 70, and a track classification module 80; the track surface defect detection model online update system includes a human-computer interaction interface 90, image markers and neural networks The training center 100 and the model parameter updating module 110 connected with the track identification module 70 and the track classification module 80; the train set data interaction system includes a signal transceiver module 120, a data synchronization module 130, a data standardization module 140, and a central database 150; track defect location The system includes a speed/acceleration sensor 160, a relative position estimation module 170, a GPS/Beidou 180, an absolute position estimation module 190, and a fusion positioning module 200 connected to the relative position estimation module 170 and the absolute position estimation module 190; the power supply system includes train power The system module 210, the conventional power supply module 220, the emergency power supply module 230 and the power adapter module 240; in addition, the abnormality processing system includes an abnormality detection module 250, an abnormality classification module 260, an abnormality alarm 270 and an abnormality clearing 280 connected to each system; In order to prevent the failure of each system to obtain and detect images normally, it is necessary to back up the equipment to take over the work of the faulty equipment; in addition, it is also necessary to back up all the data received by the data exchange center to prevent the system from being unable to work normally due to data loss. .

The hydraulic damper cylinder II 14 and the hydraulic damper cylinder I and the piston rod II 16 and the piston rod I 4 have different dimensions.

On the basis of the implementation of the above-mentioned track surface defect detection system, the present utility model also proposes an intelligent detection method for track surface defects, as shown in FIG. 2 , the track defect detection method includes:

A. Use the image collector installed on the train to take continuous images of the track: In order to quickly obtain a complete, clear and accurate track image, the track image acquisition system is first designed to be composed of an image acquisition device 30 , a light source 50 and a shock absorption device 20 . The image collector 30 adopts a high-speed area array CCD camera with sufficiently high resolution, and a fixed-focus lens with an appropriate focal length is installed; the light source 50 adopts an auxiliary light source of suitable shape and type with adjustable brightness and illumination angle. The acquired image is transmitted to the image processor 60 .

B. Identify and extract the track in the acquired image.

The image is preprocessed to remove the useless information of the image and enhance the useful real information, enhance the detectability of the target and reduce the data throughput in the subsequent process, thereby improving the reliability of the data; in order to complete the accurate detection of track defects , it is necessary to use the image processor 60 to preliminarily process the image transmitted by the image collector 30. As shown in the figure, step B specifically includes:

B1. Grayscale processing is performed on the image transmitted by the image collector 30: grayscale processing is performed on the original RGB track image by using the grayscale weighting method. The calculation formula of the weighted image gray value f(i, j) is shown in (1).

f(i,j)=αR(i,j)+βG(i,j)+γB(i,j) (1)

In formula (1), α, β, γ are the coefficients of the gray value calculation formula, i represents the row label of the image where the pixel is located, j represents the column label of the picture where the pixel is located, and R(i, j) represents the original image of the pixel. The red pixel component of , G(i, j) represents the green component of the pixel in the original image, and B(i, j) represents the blue component of the pixel in the original image.

B2. Segmentation of the region of interest on the image obtained by the above operation: Considering the large color difference between the track and the road surface, in order to efficiently extract the track surface from the image, the present utility model adopts an adaptive threshold segmentation method based on a Gaussian distribution model. ; In the ROI area, calculate the mean value and variance of the grayscale of each pixel point, and then calculate the segmentation threshold according to formula (2), and use this as a basis to segment the grayscale image to obtain a binary image.

In formula (2), μ _f represents the mean value of pixel gray level, σ _f represents the variance of pixel point gray level, and _IT represents the segmentation threshold.

B3. Perform morphological processing on the image obtained by the above operation: in order to eliminate noise and highlight the highlight part of the track, find the obvious maximum value area in the track image, and use the binary image of the structuring element of T×T pixels to process the image. Dilation and erosion operations are performed, and the edge information of the image is obtained by making the difference between the two images; thus, the track in the captured image is identified.

In order to achieve accurate detection of track defects, it is necessary to roughly extract the identified tracks first.

C. Preliminary processing and positioning of the track is performed on the image of the extracted track.

Further, in order to achieve precise positioning of the track, it is first necessary to perform preliminary processing, as shown in the figure, step C specifically includes:

C1, the filter processing of track picture: due to the uncertainty of the track environment, the discrete noise with a large value often occurs in the picture, for the protection of the edge of the image, the utility model first adopts median filtering, and then carries out bilateral filtering; In bilateral filtering, the output value of a pixel depends on the weighted combination of neighboring pixel values, and the calculation method is shown in formula (3).

In formula (3), the weighting coefficient h(i, j, k, l) depends on the product of the definition kernel and the range kernel.

Among them, the defined core domain is expressed as formula (4).

The range kernel is expressed as Equation (5).

The weight function is expressed as formula (6).

h(i,j,k,l)=dv(6)

C2. Use the canny operator to perform edge detection on the track image.

C3. Perform preliminary false edge removal on the image after the above operation: use a threshold method to perform preliminary false edge removal, wherein the calculation method of the threshold _ST is as shown in formula (7).

In formula (7), n is the total number of columns in the picture, I _max is the maximum pixel value, and _Is is the pixel reduction multiple.

Further, the flow chart of the preliminary pseudo-edge removal algorithm is shown in FIG. 5 .

C4. Further pseudo-edge removal is performed on the picture after the above operation, and the present invention proposes a constraint condition (8) to remove the pseudo-edge.

In the formula (8), STD _i is the standard deviation of the pixel value of the ith row of the track image, where i=(0, 1, 2, ..., m-1), n represents the total number of columns of the input image, and m represents the input image The total number of rows, c is a dynamic factor that can be adjusted.

C5. Track fitting and positioning: Since the operation of removing the false edge may remove the real edge, 0 needs to be further restored; this paper performs linear fitting on the two groups of edge points retained after the false edge cleaning, and the two straight lines are obtained by linear fitting. The actual edge can be approximately recovered after intercept and slope.

D. Defect detection of the accurately positioned track: After the track is accurately positioned, inception-v3 is selected as the basic network structure, and the convolutional neural network is used to classify the track image.

An ideal model is obtained by using the data of the present invention and a small learning rate for migration training, so that the data set of the present invention is applied to a large neural network, and the model trained by big data can be applied to its own model.

Among them, the recall rate (P) and the recall rate (P) are important indicators to evaluate the detection and recognition effects, and their definitions are shown in equations (9) and (10), respectively.

In equations (9) and (10), TP represents the number of accurately detected tracks, FP represents the number of faulty tracks detected incorrectly, and FN represents the number of actually defective tracks but detected as good tracks.

Among them, the RGB color mode is a color standard in the industry, which is obtained by changing the three color channels of red (R), green (G), and blue (B) and superimposing them on each other. of color.

On the basis of the above-mentioned track defect detection method based on neural network and machine vision, the present utility model also proposes a method for locating the global position of the defect track, as shown in FIG. 6 , the positioning method includes:

Considering the particularity of the railway, the utility model adopts the method of integrating global positioning system (GPS/Beidou, etc.) and inertial sensor (IMU) data and combining with high-precision map; because the train driving environment is relatively complex and there are often tunnels crossing , jungle and other signal shielding road sections, so GPS/Beidou, etc. will have obvious multi-path reflection problems and no signal problems, resulting in large errors in the received GPS/Beidou and other positioning information; IMU is a high-frequency inertial sensor. The detected speed, acceleration and other data are used to calculate the travel displacement information of the train in real time. However, due to the accumulation of integral errors in the process of calculating the train displacement, the effective positioning of the train cannot be realized. and other sensor fusion technology to fuse GPS/Beidou and other inertial sensor data to achieve relatively accurate positioning of the train; in addition, in order to achieve high-precision positioning of the train, the present utility model also utilizes the high-precision map positioning method and the above-mentioned GPS/Beidou. The positioning method fused with the IMU mutually corrects the positioning error technology to achieve the precise positioning of the train.

In order to improve the accuracy of the track surface defect detection results, the utility model also proposes a method for real-time updating of the neural network model, as shown in Figure 7, the principle of the method is as follows:

After the results detected by the track surface defect system are confirmed by the railway maintenance personnel, the detection images can be divided into two categories: one is the images with real defects on the track surface, which are marked as "1", and the other is the images with defects on the track surface. The track surface is determined to have no defects, and this type of image is marked as "0"; there is also a class of images with an incorrect detection result, which will also be classified according to whether there are actual track defects after other inspection systems or manual verification. It is marked as "0" or "1". Send the confirmed and labeled images to the neural network training center, and re-supervised the training of the neural network model; finally, use the retrained model to update the track surface defect detection system, so as to continuously improve the detection system. accuracy.

In order to improve the working accuracy of the detection device and prevent the detection device from being paralyzed due to the damage of the image acquisition equipment, the utility model proposes a method for improving the working accuracy and efficiency of the detection device, the method comprising:

Install a set of image acquisition subsystems at the front end of the train and the rear end of the train;

Use the two systems before and after to verify the recognition results to improve the accuracy of the detection system; and, if one of the image acquisition subsystems fails and cannot acquire orbital images, the other set of image acquisition subsystems can continue to work without paralyze the entire system;

If the data sent by the inter-train data exchange center on the previous train is the track defect detection system of the train, and the result of the detection of the device at a certain position is that the track surface is defective, then the train will carefully detect when it passes here. The defect situation of the track at this place is compared and analyzed with the detection result of the previous train, so as to improve the accuracy of the system defect detection.

In order to ensure that each system and device of the present utility model can work normally, the present utility model proposes a power supply system that provides electrical energy for each system and device. The power supply method of the power supply system is as follows:

Obtain electrical energy directly from the train power system module 210 to charge the conventional power supply module 220;

At the same time, in order to prevent the whole system from being unable to work due to the failure of the conventional power supply module 220, the present invention is also equipped with an emergency power supply module 220, which is also charged by the train power system module 210;

Considering the inconsistent voltage required by all systems and devices of the present invention, the conventional power supply module 220 and the emergency power supply module 220 also need to be connected to the power adapter module 240 to output various required voltages to meet the requirements of each system and device.

Finally, in order to prevent blurring of the collected images due to the vibration during the running of the train, the present invention also provides a shock absorption device for an image acquisition module, as shown in FIGS. 8 and 9 , the shock absorption device 20 includes:

Longitudinal damping spring Ⅰ3, longitudinal damping spring Ⅱ12, transverse damping spring Ⅰ9, transverse damping spring Ⅱ13, axial damping spring 15, hydraulic damper hydraulic cylinder 2, piston rod 4, use hydraulic damper to absorb consumption factors. The energy of vertical, horizontal and axial vibration of the device caused by the acceleration, deceleration or turbulence of the train, so as to maintain the stability of the image acquisition module;

The camera 5 is fixed in the camera fixing device 6, and the camera fixing device 6 is connected with the protective cover 1 through the connecting member 7, the pin member 8 and the hydraulic damper, thereby realizing the shock absorption of the image acquisition module;

The light source 11 and the light source fixing member 10 are used to fix the light source 11 on the protective cover 1 by using the light source fixing member 10 to prevent the clear image from being unable to be captured due to insufficient light;

The protective cover 1, which is fixed on the train, is used for placing and protecting the shock-absorbing spring and protecting the image acquisition device, so as to prevent the damage of the image acquisition device and the shock-absorbing spring due to weather.

The above are only some or preferred embodiments of the present utility model, and neither the text nor the accompanying drawings can therefore limit the protection scope of the present utility model. Equivalent structural transformations made by the contents of the accompanying drawings, or direct/indirect applications in other technically related fields are all included within the protection scope of the present invention.

Claims (2)

1. An intelligent online detection device for track surface defects, characterized in that: by a track surface accurate defect detection system, a track surface defect detection model online update system, a train set data exchange system, a track defect location system, an abnormality processing system and a power supply system constitute; Wherein, the track surface defect detection system includes an image acquisition subsystem and a track detection and classification subsystem, and the image acquisition system consists of a base (17), a damping device (20) connected to the base (17), and a calibration component (40), a light source (50) and an image collector (30); the track detection and classification subsystem consists of an image processor (60) connected with the image collector (30), a track identification module (70) and a track classification The module (80) is composed; The track surface defect detection model online updating system consists of a human-computer interaction interface (90), an image labeling and neural network training center (100), and a model parameter updating module connected with the track identification module (70) and the track classification module (80). (110) is composed, the human-computer interaction interface (90) is connected with the neural network training center (100) and the model parameter updating module (110) through image marking; The train set data interaction system is composed of a signal transceiver module (120), a data synchronization module (130), a data standardization module (140) and a central database (150). The data standardization module (140) is connected with the central database (150); The track defect locating system consists of a relative position estimation module (170) connected with a speed/acceleration sensor (160), a GPS/Beidou (180), an absolute position estimation module (190) connected with the GPS/Beidou (180), and a fusion positioning module (200) connected with the relative position estimation module (170) and the absolute position estimation module (190); The anomaly handling system is composed of an anomaly detection module (250) connected to each system, an anomaly classification module (260), an anomaly alarm (270), and an anomaly clearing (280) connected to the anomaly alarm (270). The module (250) is connected to the abnormality alarm (270) through the abnormality classification module (260); The power supply system is composed of a train power system module (210), a conventional power supply module (220), an emergency power supply module (230) and a power supply adapter module (240), and the train power system module (210) passes through the conventional power supply module (220) respectively ) and the emergency power supply module (230) are connected to the power adapter module (240). 2. An intelligent online detection device for track surface defects according to claim 1, characterized in that: the shock absorbing device (20) consists of a protective cover (1) fixed to the train, a hydraulic damper cylinder I ( 2), longitudinal damping spring I (3) 3, piston rod I (4), camera (5), camera fixing device (6), transverse damping spring I (9), light source fixture (10), Light source (11), longitudinal damping spring II (12), transverse damping spring II (13), and axial damping spring (15); The camera (5) is fixed in the camera fixing device (6), and the camera fixing device (6) is connected with the protective cover (1) through the connecting piece (7), the pin piece (8) and the hydraulic damper; The light source (11) is fixed to the protective cover (1) by the light source fixing member (10).

Images