KR20190024447A - Real-time line defect detection system - Google Patents

Abstract

The present invention relates to a real-time railway defect detection system, comprising: a railway data obtaining unit installed at a lower portion of a railway vehicle to obtain railway data including state information of a railway surface, a sleeper, and a fastener located at the lower portion of the railway vehicle in real time during operation of the railway vehicle; and a railway information integration management server for collecting the railway data including the state information of the railway surface, the sleeper, and the fastener transmitted from the railway data obtaining unit in real time, and providing defect integrated information by detecting and analyzing defects for the railway surface, the sleeper, and the fastener using the collected railway data.

Description

[0001] REAL-TIME LINE DEFECT DETECTION SYSTEM [

The present invention relates to a real-time line defect detection system, and more particularly, to a system and method for real-time line defect detection, which is installed at a lower portion of a railway vehicle, and includes a rail surface, A rail data acquiring unit for acquiring data in real time and rail data including status information of a rail surface, a sleeper, and a fastener transmitted from the rail data acquiring unit in real time, and using the collected rail data, (3D) image can be acquired at a high speed by incorporating a rail information integration management server that provides defect comprehensive information by detecting and analyzing defects of each of the railroad ties, sleepers, and fasteners, Image analysis provides comprehensive defect information for rail surfaces, sleepers, and fasteners By providing the auto repair business representatives maintain that allows you to perform more efficient maintenance work relates to real-time track defect detection system.

Many trains have been developed continuously in order to secure the safety of trains and passengers, to increase the efficiency of driving, and to improve the driving convenience.

Train control devices such as ATP, ATC, and ATO are typical devices that secure the safety of trains, increase the efficiency of operation, and improve the convenience of operation.

Such a train operation control device is a device that enables automatic operation by exchanging information on train operation by ensuring the safety of the train by detecting the occupancy state of the preceding train and controlling the operation of the train.

As the technology developed, the speed of the train increased, the number of service times increased and the passenger traffic increased, safety became more important.

Safety of rails operated directly by trains is an important safety factor that causes major disasters such as derailment and overturning of trains during accidents.

The safety check for these rails is performed by using a rail test machine or a rail test machine.

The rail tester is a device that allows an expert to move along a rail and directly detect whether the rail is defective and has a working capacity of about 2 km per day.

The rail test car is a device for solving the problem of working ability and safety generated by the operation using the rail test machine. PROBE installed on the left and right side closely fits on the rail on which the water is sprayed and the ultrasonic wave is fired on the rail surface It analyzes the reflected ultrasonic waves, records the types, size and position of defects, and displays the status of the head, bottom, and abdomen of the rail as a graph. And has a working capacity of about 20 Km per hour.

However, the conventional rail detecting method as described above has a problem that the operation speed is slow, the skill level of the skilled person is dependent on the measured value, and the measurement value varies depending on the state of the rail, the state of the probe, and the like.

Korea Patent Publication No. 10-2009-0085214

SUMMARY OF THE INVENTION It is an object of the present invention to provide a railway vehicle which is installed on the right and left sides of a body of a railway vehicle and which can receive rail data including status information of a rail surface, A rail data acquisition unit for acquiring rail data, and rail data including status information of rail surfaces, sleepers, and fasteners transmitted from the rail data acquisition unit in real time, and using the collected rail data, By providing a rail information integration management server that provides defect comprehensive information by detecting and analyzing defects, high resolution images of rail surfaces, sleeper and fasteners can be obtained at high speed and can be obtained through three-dimensional (3D) image analysis Automatic maintenance of comprehensive defect information for rail surfaces, sleeper and fasteners To provide a rail flaw detection system in real time, which allows you to perform maintenance tasks more efficiently by providing business representatives.

According to another aspect of the present invention, there is provided a system for detecting a line defect in a railway vehicle, the railway vehicle including a railway surface, a tie rail and a fastener, A rail data acquiring unit that acquires rail data in real time; And a controller for collecting in real time the rail data including the status information of the rail surface, the sleepers, and the fasteners transmitted from the rail data acquiring unit, and detecting and detecting faults for the rail surface, the sleepers, and the fasteners using the collected rail data. Analyzing the defect information, and providing defect comprehensive information.

The present invention can acquire high-resolution images of rail surfaces, sleeper and fasteners at high-speed operation, and automatically collects defect comprehensive information for rail surfaces, sleepers, and fasteners through three-dimensional (3D) image analysis, To provide more efficient maintenance work.

FIG. 1A is a diagram illustrating left and right rail images captured using a rail data acquisition unit and a rail data acquisition unit attached to a railway vehicle according to an embodiment of the present invention.
1B schematically illustrates the relationship between the rail data acquisition unit and the rail information integration management server according to the present invention.
FIG. 2 illustrates a detailed configuration of a rail information integration management server according to the present invention.
FIG. 3 is an exemplary embodiment of the present invention, which shows the integrated information of rail defects analyzed by the rail information unified management server.
4 is a flowchart illustrating the analysis and maintenance of rail defects using the rail information integration management server according to the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1A is a diagram illustrating left and right rail images captured using a rail data acquisition unit and a rail data acquisition unit attached to a railway vehicle according to an embodiment of the present invention. FIG. And the relationship between the integrated management servers.

1A, the present invention includes a rail data acquisition unit 100 installed on the left and right sides of a body of a railway vehicle T10 to photograph a shape of the rail 10 while the railway vehicle T10 is traveling at high speed. To acquire rail data.

Here, the rail data means the state information of the rail surface 11, the sleepers 12 and the fasteners 13 constituting the rail 10, such as the state of cracks and dents of the rail surface 11, The damage and cracking state of the fastener 12 and the state of the fastener 13 such as breakage or failure due to each kind of the fastener 13 (e.g., e clip, fast clip, boss loom, etc.).

1B, the rail data acquisition unit 100 includes a camera unit 110, a laser unit 120, an illumination unit 130, and an illuminance sensing unit 140.

The camera unit 110 includes a 3D rail camera having high-speed photographing and high-resolution performance, and photographs rail surfaces to acquire rail data. The acquired rail data is stored in an internal memory, To the rail information unified management server 200 in real time.

The laser unit 120 includes a line laser, and the 3D rail camera irradiates a laser beam onto a rail surface to be photographed, and analyzes the reflected beam to acquire depth information.

Hereinafter, a method of detecting a defect in a rail using depth information will be briefly described.

First, the data acquired through the line laser can be expressed in a graph by obtaining the distance from the reference position using the laser triangulation method. When the data is combined by photographing every frame in the same manner, So that a range map can be generated.

In this case, since the obtained range map includes all the depth values of the rail surface, it is possible to attempt to detect defects using the depth map, but using the 3D data directly for the purpose of only that is very inefficient in terms of calculation amount .

Therefore, it is more effective to project the obtained 3D data on the XY plane again to create a two-dimensional image and to process it by utilizing it.

However, if the 3D data is projected onto the 2D plane, the depth values disappear. Therefore, in order to solve this problem, the depth values are displayed in different colors so that the depth values can be utilized in the 2D image.

On the other hand, for the detection of surface defects using the 3D profile, there is a method of detecting defects by accumulating defects by using moving averages in respective cross sections to obtain a range of defects, a method of converting a profile image into a Depth Map image, There are two methods for obtaining the above-mentioned values, which will be described in more detail below.

First, the moving average using method is a method of detecting a sudden lowered point compared to the height of a surrounding area and extracting it as a defect when a defect of a part of the surface is only a part of a single surface. For example, If the defect is detected in the same way as the image of the defect, it is possible to accumulate the defect in the position associated with the previous defect to obtain the width of the connected defect.

Secondly, the depth map image can be generated by using the acquired profile data. In this method, the depth map image having the depth value represented by the brightness difference can be generated. And the defect area and the area can be obtained by performing a blob labeling in which adjacent pixels having the same value are numbered by using the generated binary image.

Next, the illuminating unit 130 can be configured as a 2D image acquiring light, for example, a camera for acquiring a 3D image and a camera for acquiring a 2D image separately. In this case, The rail data can be easily obtained through the 3D rail camera or the line laser without being affected by the environment such as nighttime or weather deterioration.

The illuminance sensing unit 140 includes an illuminance sensor to sense the brightness of the light reflected by the rails and to allow the illuminator 130 to emit light of a predetermined brightness according to the sensed illuminance value.

The rail information integration management server 200 may be installed inside or outside the railway vehicle T10 and may operate and collect rail data transmitted by the rail data acquisition unit 100 through the network 5 , The collected rail data is used to detect and analyze defects for each of the rail surfaces / sleepers / fasteners to provide comprehensive information on the defects, thereby enabling more efficient rail maintenance work to be performed. Will be described later in Fig.

Here, the network 5 provides an interface capable of performing wireless communication with the outside, such as WiFi, Zigbee, RF, 3G, 4G, LTE, LTE-A, Internet) can be used.

FIG. 2 illustrates a detailed configuration of a rail information integration management server according to the present invention.

2, the rail information unified management server 200 according to the present invention includes a rail data collecting unit 210, a communication unit 220, a database unit 230, a rail defect detecting unit 240, (250), a maintenance management unit (260), and an integrated control unit (270).

The rail data collecting unit 210 receives the rail data transmitted in real time from the rail data acquiring unit 100, and collects the rail surface condition information, the sleeping condition information and the fastener condition information, respectively.

The communication unit 220 provides an interface capable of performing communication with the rail data acquisition unit 100 or an external terminal such as a maintenance manager terminal. Examples of the interface include a WiFi, a Zigbee, Wireless communication such as RF (RF), 3G, 4G, LTE, LTE-A, and Wireless Broadband Internet, PLC wire communication or the Internet, and Social Network Service (SNS).

The database unit 230 stores various operating programs necessary for various management of the railroad car T10 in advance or stores various kinds of information processed through the integrated control unit 270 such as rail data, Surface / sleeper / fastener defect information, etc.) using a certain storage medium.

In this case, the storage medium may be a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (such as SD or XD memory), a RAM (Random Access Memory), SRAM (Static Random Access Memory), ROM (Read Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM A disk, an optical disk, or the like can be used.

The rail defect detection part 240 includes a rail surface defect detection part 241, a sleeping defect detection part 242 and a fastener defect detection part 243. In this case, a defect detection algorithm To check whether or not each of the rail surface / sleeper / fastener is defective, and the degree of defects. Hereinafter, this will be described in detail.

The rail surface defect detection unit 241 analyzes the collected rail surface data to check the presence or absence of defects on the rail surface and the degree of defects.

In addition, in order to detect defects on the rail surface in the present invention, a rail surface defect sample image was used and the following image processing procedure was performed.

First, blurring was performed using a Gaussian filter to remove noise in the input data, and then segmentation was performed using a binary threshold. And then the noise of the segmentation portion is removed by an erosion operation and finally the defect of the rail surface is detected by giving a threshold value to the area width of the segmentation portion The results are shown in Table 1 below.

Table 1 below shows the classification of degree of defects in stages by using rail surface data obtained from high-speed railway (300 km / h) and ordinary railroad.

As shown in Table 1, Class 1 to Class 4 are classified according to the size of defects on the rail surface (pixel). As the number of classifications increases, the urgency of maintenance is high.

Next, the sleeping defect detector 242 detects the presence or absence of defects in the sleepers using the ART2 neural network model by a method of comparing and analyzing the preformed sleepers test bed with the collected sleeping data and the sleeping defect detection algorithm .

The reason for constructing the database using the sleeping test bed is that it is very difficult to construct the database including the sleeping room including the actual defects.

First, we will explain the procedure of building a sleeping test bed database to build a database by synthesizing randomly created crack images after acquiring normal sleeping images.

① Sleeping image acquisition process

In a laboratory designed to build a test bed, the rails are run in a vertical position with the sleepers, and the height of the sleepers and camera lenses is kept constant at 135 cm during the shooting. Only the regions are stored separately.

For example, seven normal sleepers were photographed seven times to obtain a total of 49 normal sleepers images.

② Crack image synthesis process

A crack image is created using the first seven snapshots of the normal sleepers used in the step of ① capturing the sleeping image, and the crack image is synthesized with the separated image of the sleeping area.

In order to synthesize a crack image on a normal sleeping image, two crack images are selected for each criterion and the cracks are divided into three parts (left (L), middle (M), right (R) Thereby generating a sleeping crack image.

③ Database building process

To classify the sleeping faults by grade, classify them as Class 4 (Class A ~ D) based on defect area.

Next, crack images are extracted from the concrete crack images in accordance with each coupling class classification criterion.

 A sleeping defect database is constructed by synthesizing two crack images for each grade (Class 4, Class A, B, C, D) in three regions (L, M, R regions) using one normal sleeper image.

In this case, the total number of sleeping defect data constructed is calculated as follows.

(Total number of sleeping defect data) = [number of normal sleepers (number of sleepers) × 7 (number of shots) = 49)] + number of defective sleepers images (number of sleepers) × 4 Crack location area) x 2 (crack type) = 1176 pieces] = 1225 pieces

Next, the ART2 neural network model used in the sleeping defect detection algorithm will be described.

As the simplest type of sleeper defect detection method, for example, a dark part of a sleeper area is detected as a defect, an area is calculated, a defect is determined, and a grade is divided according to a defect area.

This method has a short processing time by a simple method and has an advantage that the area of defects can be obtained and classified by area.

On the other hand, the sleeper pattern is extracted as a rectangular shape and the sleeper defect is examined in the extracted area.

In the case of a sleeping image obtained outdoors, even if the same sleeping image is seen due to the influence of the image angle, brightness, and ambient illumination, the image is seen in various colors or the shape is deformed, so that it can adaptively cope with various situations , A sleeping defect inspection method using a neural network that continuously learns itself (such as ART2 neural network) is applied.

The ART2 neural network is a neural network model capable of real-time learning that has the advantage of not affecting already learned patterns by forming a new cluster when an untrained pattern comes in.

The connection weights of the ART2 neural network responds evenly to the cluster generation by taking the average of all the input patterns. When the input vectors are input, the similar existing clusters with different characteristics are updated, Vectors can also cause weighting characteristics to be reduced by means of connection weight vectors and averages.

In order to train the sleeper image pattern on the ART2 neural network, the sleeper image is normalized to a certain size and then pattern data is generated and learned.

ART2 is divided into different clusters because they are slightly different from each other depending on the pattern inspection boundary parameter and type of sleeper.

In the course of training, the boundary parameter values are divided into several stages, and the threshold value at which the sleeping pattern is recognized is searched for in each cluster.

In practice, when no clustering is performed and no new clusters are generated beyond the boundary parameters, it is judged to be a defect, not a sleeping image.

Hereinafter, the above process is divided into a training process and a classification process.

First, the training process consists of the steps of inserting images -> setting the first image as a cluster reference ---> calculating the distance to the cluster ---> training the data ---> .

Next, the classification process includes the steps of loading the training data ---> inserting the image ---> calculating the distance to the cluster ---> finding the closest distance among the clusters ---> And determining an image rating.

Finally, the fastener defect detection unit 243 separates the background and the facility through line extraction through preprocessing on the collected fastener surface data, designates the ROI through projection, After performing pattern matching on the fastener element facilities in the ROI, the fastener defect according to the fastener type is checked based on the results of the recognized element facilities.

In this case, the pattern matching is performed by matching with the pre-stored test pattern in consideration of the fact that the shape, size, installation position, etc. of the collected fastener (such as e clip, fast clip, It means checking the presence or absence of defects according to the mutual matching degree of the patterns.

Referring again to FIG. 2, the defect comprehensive information providing unit 250 determines whether or not the rail surface / sleepers / sleepers detected by the rail surface defect detecting unit 241, the sleeper defect detecting unit 242 and the fastener defect detecting unit 243, And provides comprehensive information about the fastening hole defects, which will be described below with reference to FIG.

FIG. 3 is an exemplary embodiment of the present invention, which shows the integrated information of rail defects analyzed by the rail information unified management server.

In this case, the upper part of FIG. 3 shows the defect comprehensive information on the rail surface by item, the interruption figure of FIG. 3 shows the defect comprehensive information on the railroad by item, and the lower part of FIG. 3 shows the defect Comprehensive information is shown by item.

In this case, the defect comprehensive information includes, for example, object type (crack / dent / dropout), rail position (right / center / left), distance (KPR), width (mm) mm), a size (area, mm ² ), an acquisition date (May 27, 2015), a defective image (3D)

Next, the maintenance management unit 260 calculates the schedule, priority, maintenance, and the like of the rail-related maintenance work based on the defect comprehensive information for each of the rail surface / sleeper / fastener provided by the defect comprehensive information providing unit 250 Guides, and personnel information to create more efficient rail-related maintenance tasks.

Lastly, the integrated control unit 270 includes a rail data collecting unit 210, a communication unit 220, a database unit 230, a rail defect detecting unit 240, a defect comprehensive information providing unit 250, a maintenance management unit 260, Respectively.

4 is a flowchart illustrating the analysis and maintenance of rail defects using the rail information integration management server according to the present invention.

Hereinafter, the analysis and maintenance of rail defects using the rail information integration management server according to the present invention will be described with reference to FIGS. 2 and 4. FIG.

The rail data collecting unit 210 of the rail information integrated management server 200 has a first step S10 of receiving rail data transmitted in real time from the rail data acquiring unit 100. [

Next, there is a second step S20 of performing a preprocessing process on the rail data for each of the rail surfaces / sleepers / fasteners collected in the first step S10.

The preprocessing process refers to image processing such as unnecessary background removal, ROI selection, and contrast adjustment considering the characteristics of the rail data for each of the collected rail surfaces / sleepers / fasteners.

Next, each of the rail surface defect detection unit 241, the sleeping defect detection unit 242, and the fastener hole defect detection unit 243 is subjected to a third process of examining defects of the rail surface / sleepers / fastener using a defect detection algorithm S30).

Next, the defect comprehensive information providing unit 250 searches the rail surface / sleeper / fastener defects detected by the rail surface defect detecting unit 241, the sleeper defect detecting unit 242 and the fastener defect detecting unit 243, (Step S40).

Lastly, the maintenance management unit 260 manages the schedule, priority, and maintenance of the rail-related maintenance work based on the defect comprehensive information for each of the rail surfaces / sleeper / fasteners provided by the defect comprehensive information providing unit 250 And a fifth step (S50) of generating a guide, a responsible worker information, and the like and transmitting a maintenance command to the terminal of the maintenance worker in accordance with the degree of the defect.

While the present invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not limited to the disclosed embodiments. It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit of the invention.

100: rail data acquisition unit
110: camera unit 120: laser unit
130: illumination unit 140: illuminance detection unit
200: Rail information integration management server
210: rail data collecting unit 220: communication unit 230:
240: Rail defect detector
241: Rail surface defect detection part 242:
243: fastener hole defect detector
250: Defect Comprehensive Information Offering
260: Maintenance management department
270:
10: Rail
11: Rail surface
12: sleepers
13: Fastener
T10: Railway vehicles

Claims (11)

A rail data acquiring unit installed at a lower portion of the railway vehicle to acquire rail data including status information of a rail surface, a sleeper, and a fastener located at a lower portion of the railway car in real time during operation of the railway vehicle; And
The rail data including the state information of the rail surface, the sleepers and the fasteners transmitted from the rail data acquiring unit are collected in real time, and the rail surface, the sleepers, and the fasteners are subjected to defect detection and analysis And a rail information unified management server for providing defect integrated information. The rail data acquisition system according to claim 1,
A camera unit provided with a rail camera having high-speed shooting and high-resolution performance, and providing a picked-up image of a rail surface, a sleeper, and a fastener located at a lower portion of the railway vehicle; And
Line laser, and a laser beam is irradiated around the rail taken by the rail camera, and depth information for defect detection is acquired for each of the rail surface, the sleepers, and the fasteners through analysis of the reflected beam And a laser unit for detecting a defect in the real line.

The method of claim 1, wherein the rail information integration management server comprises:
A rail data collecting unit for receiving the rail data transmitted in real time from the rail data acquiring unit and collecting the rail surface condition information, the sleeping condition information and the fastener condition information, respectively;
A rail surface defect detector for analyzing the rail surface data collected by the rail data collecting unit and checking the presence or absence of defects on the rail surface and the degree of defects;
A sleeping defect detector for analyzing the sleeping data collected by the rail data collecting unit to check the presence or absence of defects in the sleepers and the degree of defects; And
And a fastener defect detector for analyzing the fastener data collected by the rail data collecting unit to check the presence or absence of defects in the fastener and the degree of defect. 4. The apparatus of claim 3, wherein the rail surface defect detector comprises:
Wherein a first image processing process is performed on the rail surface data, and then the presence or absence of a defect on the rail surface is checked through comparison with a rail surface defect sample image constructed in advance, system. [5] The method of claim 4,
A first process of performing blurring using a Gaussian filter to remove noise of input data;
A second step of performing segmentation using a binary threshold;
A third step of removing noise of a segmented part through an erosion operation; And
And a fourth step of providing a threshold value to the area width of the segmented part and detecting a defect on the rail surface by comparing with the threshold value. The apparatus as claimed in claim 3, wherein the sleeping defect detector comprises:
A system for comparing and analyzing the sleeping data collected in advance with a sleeping test bed, and an ART2 neural network model for checking the presence or absence of defects in the sleepers and the degree of defects. The sleeping test bed according to claim 6,
One normal sleeper image is divided into three regions (L, M, and R regions), and two crack images are classified into four regions (Class A, B, C, and D) Respectively, to generate a real-time line defect detection system.

4. The apparatus of claim 3, wherein the fastener hole defect detector comprises:
And checking the presence or absence of defects and the degree of defects of the fastener in consideration of the degree of matching between the collected fastener data and a pre-constructed test pattern. 9. The method according to claim 8,
Type, size, and installation position information of the fastening tool. The method of claim 3,
Further comprising a defect comprehensive information providing unit for providing defect comprehensive information on each of the rail surface, the sleeper, and the fastener detected by the rail surface defect detecting unit, the sleeper defect detecting unit and the fastener defect detecting unit, Detection system. 11. The method according to claim 10,
A defect size, a defect size (width / height / area), a 3D defect image, and a defect acquisition date information for each of the rail surface, the treadle and the fastener.

Images