Rail body defect detection system using deep learning

Abstract

A rail body defect detection system according to one embodiment of the present disclosure can comprise a camera unit, a sensor unit, and a detection unit. The camera unit can comprise multiple cameras arranged in the lower part of a train to capture an image of the rail body. The sensor unit can comprise a position sensor that provides position information of the rail body. Based on the rail body image and the position information, the detection unit can determine whether the rail body is defective and provide a detection result. In the rail body defect detection system according to the present disclosure, the detection unit determines, on the basis of data obtained by mixing rail body images taken by a plurality of cameras contained in the camera unit and position information of the rail body depicted by the corresponding cameras, whether a defect exists in the rail body, thereby quickly and accurately identifying a location with a high risk of accident on the rail body.

Description

Reference to related registration(s)

[0001] This application claims priority from Korean patent application No. 10-2024-0157927, which was filed with the Korean Intellectual Property Office on November 8, 2024, and the disclosure of which is incorporated herein by reference in its entirety. Background 1. Area

[0002] The present disclosure relates to a system for the preventive detection of rail body defects. 2. Description of the state of the art

[0003] Even if a sudden, dangerous situation arises during train operation, the train's weight and speed make sudden braking impossible. Due to the characteristics of a train's braking distance, the train's speed is controlled by railway signaling systems, thus limiting and safely managing sudden, dangerous situations.

[0004] There is a growing need for safety measures that proactively detect and verify risks through deep learning by utilizing data on the condition of the track bed that can be collected by moving trains, and various studies are being conducted to reduce the risk of train accidents by implementing more effective response methods. [Relevant prior art documents][Patent documents]

[0005] Korean application no. 10-2024-0126268 (published on August 20, 2024) overview

[0006] One aspect of the present disclosure may provide a rail body defect detection system in which a detection unit determines whether a rail body is defective based on data obtained by combining rail body images taken by a plurality of cameras contained in a camera unit and position information of the rail bodies depicted by the relevant cameras, thereby enabling faster and more accurate identification of a high-risk location in a rail body.

[0007] A rail body defect detection system according to one embodiment of the present disclosure may comprise a camera unit, a sensor unit, and a detection unit. The camera unit may comprise multiple cameras arranged in the lower part of a train to capture a rail body image corresponding to an image of the rail body. The sensor unit may comprise a position sensor that provides position information of the rail body. Based on the rail body image and the position information, the detection unit can determine whether the rail body is defective and provide a detection result.

[0008] The camera unit can comprise a first camera unit and a second camera unit. The first camera unit can include first rail body cameras that capture first rail body images of a top, a first side, and a second side of a first rail body below the rail bodies. The second camera unit can comprise second rail body cameras that capture second rail body images of a top, a first side, and a second side of a second rail body below the rail bodies.

[0009] The sensor unit can comprise a first sensor unit and a second sensor unit. The first sensor unit can provide position information of the first rail body, acquired by the first rail body cameras contained within the first camera unit. The second sensor unit can provide position information of the second rail body, acquired by the second rail body cameras contained within the second camera unit.

[0010] The track defect detection system may also include a receiving unit. The receiving unit can receive an acoustic signal transmitted by the track while the train is in motion.

[0011] The rail body defect detection system can also include an interval adjustment unit. The interval adjustment unit can adjust a time interval during which the majority of rail body cameras operate based on the acoustic signal.

[0012] The sound signal can include a sound signal with a specific frequency and an abrupt sound signal that have already been learned, and the time interval can decrease as the strength of the sound signal increases.

[0013] The rail body defect detection system can further include an image evaluation unit, an acoustic evaluation unit, and a weighting unit. The image evaluation unit can provide an image evaluation based on the rail body image provided by the camera unit. The acoustic evaluation unit can provide an acoustic evaluation based on the intensity of the acoustic signal. The weighting unit can provide image-weighted and acoustic-weighted evaluations by applying weighting values ​​to the image and acoustic evaluations, respectively.

[0014] The track defect detection system can further comprise a comparison unit and a decision unit. The comparison unit can compare the sum of the image-weighted assessment and the sound-weighted assessment with a predetermined reference assessment and provide a comparison result. Based on the comparison result, the decision unit can determine whether the train should be stopped.

[0015] The track defect detection system may further comprise a first reset unit, a second reset unit, and a third reset unit. The first reset unit can reset the position information to a last saved position prior to the malfunction of the track defect detection system when the system is restarted due to a malfunction. The second reset unit can reset the position information to an initial value when the track defect detection system receives a start signal. The third reset unit can reset the position information to the initial value when the track defect detection system receives an audio signal corresponding to the start signal.

[0016] The investigation unit may also include an artificial intelligence unit. The artificial intelligence unit can learn the track geometry, position information, and sound signal and provide a result indicating whether the track is defective.

[0017] In addition to the technical problems mentioned above in the present disclosure, further features and advantages of the present disclosure are described below or can be clearly understood by a person skilled in the field of the present invention from this description and explanation. Brief description of the drawings Fig. Figure 1 is a diagram illustrating a rail body defect detection system according to embodiments of the present disclosure. Fig. 2 is a diagram illustrating a camera unit used in the rail body defect detection system. Fig. 1 is included. Fig. Figure 3 is a diagram illustrating a sensor unit used in the rail body defect detection system. Fig. 1 is included. Fig. 4 and Fig. Figure 5 shows the operation of the camera and sensor units used in the rail body defect detection system according to Fig. 1 is included. Fig. 6 and Fig. Figure 7 shows an interval adjustment unit used in the rail body defect detection system according to Fig. 1 is included. Fig. Figure 8 is a diagram showing a unit of assessment and a unit of weighting used in the rail body defect detection system according to Fig. 1 is included. Fig. Figure 9 is a diagram representing a comparison unit and a decision unit used in the rail body defect detection system of Fig. 1 is included. Fig. 10 and Fig. 11 are diagrams that illustrate an operating mode of a reset unit used in the rail body defect detection system of Fig. 1 is included. Fig. 12 is a diagram representing an investigation unit used in the rail body defect investigation system of Fig. 1 is included. Detailed description

[0018] The description should note that when adding reference numerals to components in the drawings, the same reference numerals refer to the same components, even if the components are shown in different drawings.

[0019] The terms used in this description should be understood as follows.

[0020] The singular terms used herein also include the plural form unless expressly stated otherwise, and the scope of this disclosure is not limited by the terms used herein.

[0021] It is understood that the terms “include” or “have” do not exclude the presence or addition of one or more other features, numbers, processes, components, parts or combinations thereof mentioned in the description.

[0022] In the following, embodiments of the present disclosure are described in detail with reference to the accompanying drawings.

[0023] Fig. Figure 1 is a diagram illustrating a rail body defect detection system according to embodiments of the present disclosure, Fig. 2 is a diagram illustrating a camera unit used in the rail body defect detection system. Fig. 1 is included, Fig. Figure 3 is a diagram showing a sensor unit used in the rail body defect detection system of Fig. 1 is included, and Fig. 4 and Fig. Figure 5 shows the operation of camera and sensor units used in the rail body defect detection system of Fig. 1 is included.

[0024] With reference to the Fig. 1 to 5, a rail body defect detection system 10 according to an embodiment of the present disclosure can comprise a camera unit 100, a sensor unit 200 and a detection unit 300.

[0025] The camera unit 100 can comprise multiple cameras arranged in a lower part of a train to capture a rail body image SI corresponding to an image of a rail body. In one embodiment, the camera unit 100 can comprise a first camera unit 110 and a second camera unit 120. The first camera unit 110 can comprise first rail body cameras that capture first rail body images of a top, a first side, and a second side of a first rail body R1 below the rail bodies. As, for example, in the Fig. 4 and Fig. As shown in Figure 5, the track bodies can comprise the first track body R1 and a second track body R2, and a train can be positioned and travel on the first track body R1 and the second track body R2. Furthermore, a plurality of first track body cameras can comprise a 1_1 track body camera C1_1, a 1_2 track body camera C1_2, and a 1_3 track body camera C1_3. If the 1_1 track body camera C1_1, the 1_2 track body camera C1_2, and the 1_3 track body camera C1_3 are connected to the lower part of the train, the first camera unit 110 can be positioned to surround the first track body R1. In this case, the 1_1 rail body camera C1_1 can image the top of the first rail body R1, and the 1_2 rail body camera C1_2 can image the first side face of the first rail body R1.Furthermore, the 1_3 rail body camera C1_3 can image the second side surface of the first rail body R1, which is opposite the first side surface.

[0026] The second camera unit 120 can include second trackside cameras that capture second trackside images of a top, a first side, and a second side of the second trackside R2 below the trackside. For example, the plurality of second trackside cameras can include a 2_1 trackside camera, a 2_2 trackside camera, and a 2_3 trackside camera. If the 2_1 trackside camera, the 2_2 trackside camera, and the 2_3 trackside camera are connected to the lower part of the train, the second camera unit 120 can be positioned to surround the second trackside R2. In this case, the 2_1 trackside camera can image the top of the second trackside R2, and the 2_2 trackside camera can image the first side of the second trackside R2.Furthermore, the 2_3 rail body camera can image the second side face of the second rail body R2, which is opposite the first side face.

[0027] Here, the second camera unit 120 can have the same structure as the first camera unit 110.

[0028] The sensor unit 200 can include a position sensor that provides position information PI of the rail body. In one embodiment, the sensor unit 200 can include a first sensor unit 210 and a second sensor unit 220. The first sensor unit 210 can provide position information PI of the first rail body R1, which was acquired by the first rail body cameras of the first camera unit 110. For example, the first sensor unit 210 can include a GPS sensor, and the first camera unit 110 can include a 1_1 rail body camera C1_1, a 1_2 rail body camera C1_2, and a 1_3 rail body camera C1_3. The images taken by the 1_1 rail body camera C1_1, the 1_2 rail body camera C1_2 and the 1_3 rail body camera C1_3 at a first time point T1 can be a 1_1 rail body image SI1_1, a 1_2 rail body image SI1_2 and a 1_3 rail body image SI1_3.In this case, the first sensor unit 210 can be used to determine the position information PI of the first rail body R1, on which the 1_1 rail body image SI1_1, the 1_2 rail body image SI1_2, and the 1_3 rail body image SI1_3 were recorded at the first time point T1. Here, the 1_1 rail body image SI1_1, the 1_2 rail body image SI1_2, and the 1_3 rail body image SI1_3, as well as the position information PI, can be mixed and transmitted as individual data to the determination unit 300.

[0029] The second sensor unit 220 can provide position information PI of the second rail body R2, which was acquired by the second rail body cameras encompassed by the second camera unit 120. For example, the second sensor unit 220 can include a GPS sensor, and the second camera unit 120 can include the 2_1 rail body camera, the 2_2 rail body camera, and the 2_3 rail body camera. The images acquired by the 2_1 rail body camera, the 2_2 rail body camera, and the 2_3 rail body camera at a second time point T2 can be a 2_1 rail body image SI2_1, a 2_2 rail body image SI2_2, and a 2_3 rail body image SI2_3.In this case, the second sensor unit 220 can be used to determine the position information PI of the second rail body R2, on which the 2_1 rail body image SI2_1, the 2_2 rail body image SI2_2, and the 2_3 rail body image SI2_3 were recorded at the second time point T2. Here, the 2_1 track image SI2_1, the 2_2 track image SI2_2, the 2_3 track image SI2_3, and the position information PI can be combined and transmitted as individual data to the detection unit 300.

[0030] The investigation unit 300 can use the rail body images SI and the position information PI to determine whether a rail body is defective and provide an investigation result PR. For example, the investigation unit 300 can determine the condition of the rail body based on the available rail body image SI and the position information PI. The investigation result PR provided by the investigation unit 300 can include a defect location and a defect type of the rail body.

[0031] In one embodiment, the detection unit 300 can further comprise an artificial intelligence unit 310. The artificial intelligence unit 310 can learn the track bed images SI, the position information PI, and the sound signals SS to provide the result PR regarding the presence of a track bed defect. For example, the artificial intelligence unit 310 can be trained by deep learning using big data, comprising a plurality of track bed images SI and position information PI, prior to the operation of the track bed defect detection system 10 according to the present disclosure. Here, the detection unit 300 can comprise an artificial intelligence server and a control server.

[0032] In the rail body defect detection system 10 according to the present disclosure, the detection unit 300 determines, based on the data obtained by mixing the rail body images SI recorded by the plurality of cameras in the camera unit 100 and the position information PI of the rail body depicted by the corresponding cameras, whether a rail body is damaged. This enables faster and more accurate identification of a high-risk location on the rail body.

[0033] The Fig. 6 and Fig. Figure 7 are diagrams illustrating an interval adjustment unit used in the rail body defect detection system. Fig. 1 is included.

[0034] With reference to the Fig. In one embodiment, the rail defect detection system 10, as described in Figures 1 to 7, can further comprise a receiving unit 310. The receiving unit 310 can receive the sound signal SS transmitted by the rail during the movement of a train. For example, when the train moves on the rail, noise can occur due to friction between the train and the rail, and the sound signal SS corresponding to the noise caused by friction between the train and the rail can be transmitted to the receiving unit 310. The receiving unit 310 can be implemented as a single unit, such as the camera unit 100 and the sensor unit 200, as shown in Figures 1 to 7. Fig. 5 is shown.

[0035] In one embodiment, the rail body defect detection system 10 can further comprise an interval adjustment unit 400. The interval adjustment unit 400 can set time intervals during which a plurality of rail body cameras are operated based on the sound signal SS. In another embodiment, the time intervals can decrease as the strength of the sound signal SS increases. For example, a preset reference sound signal strength can be a first reference strength. If the sound signal SS received by the receiver unit 310 is less than the first reference strength, multiple rail body cameras can be operated in each first time interval TI1 to image the rail body.

[0036] Conversely, if the strength of the sound signal SS received by the receiver unit 310 is greater than the first reference strength, the multiple trackside cameras can be operated in time intervals shorter than the first time interval TI1 to image the trackside. For example, a multiple of these time intervals can range from a first time T1 to a fifth time T5. If the strength of the sound signal SS at the first time T1 is less than the first reference strength, the multiple trackside cameras can be operated again at the second time T2, which is after a first time interval TI1 from the first time T1, to image the trackside.If the strength of the acoustic signal SS is greater than the first reference strength at the second time point T2, which is after the first time interval TI1 from the first time point T1, the track defect detection system 10 according to the present disclosure can operate the multiple track cameras again after the second time interval TI2, which is shorter than the first time interval TI1 from the second time point T2, to image the track. If, after this, the strength of the acoustic signal SS is also greater than the first reference strength at the fourth time point, the multiple track cameras can be operated again after a third time interval TI3, which is shorter than the second time interval TI2. A strong acoustic signal SS can be interpreted as an indication of a risk on the track, and in sections where a risk exists, it may be necessary to inspect the track at shorter intervals.

[0037] Fig. Figure 8 is a diagram showing a rating unit and a weighting unit used by the rail body defect detection system of Fig. 1 includes, and Fig. Figure 9 is a diagram showing a comparison unit and a decision unit used by the rail body defect detection system of Fig. 1 is included.

[0038] With reference to the Fig. In one embodiment, the rail body defect detection system 10 can further comprise an image evaluation unit 510, an acoustic evaluation unit 520, and a weighting unit 600. The image evaluation unit 510 can provide an image evaluation IP, which is determined based on a rail body image SI provided by the camera unit 100. For example, to calculate the image evaluation IP, the image evaluation unit 510 can use a predefined lookup table capable of determining evaluations based on the rail body image SI, and the image evaluation unit 510 can be configured to perform deep learning on the rail body image SI and then provide the image evaluation IP based on an input image.

[0039] The Sound Weighting Unit 520 can provide a sound weight SP determined based on the intensity of the sound signal SS. For example, the Sound Weighting Unit 520 can use a predefined lookup table capable of determining weights based on the intensity of the sound signal SS to calculate the sound weight SP, or the Sound Weighting Unit 520 can be configured to perform deep learning on the sound signal SS and then provide the sound weight SP based on an input sound.

[0040] The Weighting Unit 600 can provide an image-weighted score (IWP) and a sound-weighted score (SWP) by applying weighting values ​​to the image score (IP) and the sound score (SP). For example, the image score (IP) can be 60 points and the sound score (SP) 70 points. The weighting values ​​provided by the Weighting Unit 600 can include a first weight and a second weight. The first weight can be applied to the image score (IP), and the first weight can be 0.8. Additionally, the second weight can be applied to the sound score (SP), and the second weight can be 0.5. In this case, the image-weighted score (IWP) can be 60 * 0.8 = 48 points, and the sound-weighted score (SWP) can be 70 * 0.5 = 35 points.

[0041] In one embodiment, the track defect detection system 10 may further comprise a comparison unit 700 and a decision unit 800. The comparison unit 700 can compare the sum of the image-weighted rating IWP and the sound-weighted rating SWP with a predetermined reference rating REP to provide a comparison result CR. For example, the reference rating REP, which corresponds to the degree to which a train must be stopped, may be 70, and the sum of the image-weighted rating IWP and the sound-weighted rating SWP may be 83. In this case, the comparison unit 700 can provide a comparison result CR indicating that "the sum rating is greater than the reference rating REP".

[0042] Decision unit 800 can decide whether to stop the train ST based on the comparison result CR. For example, if comparison unit 700 provides a comparison result CR indicating that "the total score is greater than the reference score REP", decision unit 800 can provide a decision signal corresponding to the comparison result CR, indicating that "the train must be stopped".

[0043] The Fig. 10 and Fig. Figure 11 are diagrams illustrating the operation of a reset unit used in the rail body defect detection system. Fig. 1 is included, and Fig. 12 is a diagram illustrating an investigation unit that is part of the rail body defect investigation system. Fig. 1 is included.

[0044] With reference to the Fig.In embodiments 1 to 12, the rail body defect detection system 10 can comprise a first reset unit 910, a second reset unit 920, and a third reset unit 930. If the rail body defect detection system 10 is restarted due to a malfunction OS of the rail body defect detection system 10, the first reset unit 910 can set the position information PI to the last saved position LP before the malfunction of the rail body defect detection system 10.

[0045] Conversely, the second reset unit 920 can set the position information PI to a predetermined initial value INP when the rail defect detection system 10 receives a start signal STS, provided the system is not defective. The third reset unit 930 can set the position information PI to a predetermined initial value INP when the rail defect detection system 10 receives an audio signal STH, which corresponds to a start signal.

[0046] In the rail body defect detection system 10 according to the present disclosure, the detection unit 300 uses data obtained by mixing rail body images SI, which were recorded by a plurality of cameras in the camera unit 100, and position information PI of the rail body depicted by the corresponding cameras to determine whether a defect is present in the rail body, thereby quickly and accurately identifying a location with a high risk of accidents on the rail body.

[0047] As previously explained, the present revelation has the following effects.

[0048] In the rail body defect detection system according to the present disclosure, the detection unit uses data obtained by mixing rail body images captured by a plurality of cameras in the camera unit and position information of the rail body mapped by the corresponding cameras to determine whether a defect is present in the rail body, thereby quickly and accurately identifying a location with a high risk of accidents on the rail body.

[0049] Furthermore, additional features and advantages of the present disclosure can be newly identified through the embodiments of the present disclosure.

[0050] In addition to the technical tasks mentioned above in the present disclosure, further features and advantages of the present disclosure may be described below or are clearly understandable to those skilled in the field of the present disclosure from this description and explanation. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] KR 10-2024-0157927

[0001] KR 10-2024-0126268

[0005]

Claims

[1] Rail body defect detection system, comprising: a camera unit with a plurality of cameras arranged in the lower part of a train to capture a track body image corresponding to an image of a track body; a sensor unit with a position sensor that provides position information of the rail body; and an investigative unit that determines, based on the rail body image and position information, whether the rail body is defective and provides an investigative result. [2] Rail body defect detection system according to claim 1, wherein the camera unit comprises: a first camera unit with first rail body cameras that capture first rail body images of a top, a first side surface and a second side surface of a first rail body below the rail bodies; and a second camera unit with second rail body cameras that capture second rail body images of a top, a first side surface and a second side surface of a second rail body below the rail bodies. [3] Rail body defect detection system according to claim 1 or 2, wherein the sensor unit comprises: a first sensor unit that provides position information of the first rail body, which is captured by the first rail body cameras encompassed by the first camera unit; and a second sensor unit that provides position information of the second rail body, which is captured by the second rail body cameras encompassed by the second camera unit. [4] Rail body defect detection system according to one of claims 1 to 3, further comprising a receiving unit which receives a sound signal transmitted by the rail body during the journey of the train. [5] Rail body defect detection system according to one of claims 1 to 4, further comprising an interval adjustment unit which adjusts a time interval during which the multiple rail body cameras operate on the basis of the sound signal. [6] Rail body defect detection system according to claim 4 or 5, wherein the sound signal comprises a sound signal with a specific frequency and an abrupt sound signal that have already been learned, and the time interval decreases with increasing strength of the sound signal. [7] Rail body defect detection system according to any one of claims 1 to 6, further comprising: an image evaluation unit that provides an image evaluation based on the rail body image provided by the camera unit; a sound weighting unit that provides a sound weighting determined on the basis of the strength of the sound signal; and a weighting unit that provides image-weighted and sound-weighted assessments by applying weighting values ​​to the image and sound assessments. [8] Rail body defect detection system according to any one of claims 1 to 7, further comprising: a comparison unit that compares the sum of the image-weighted score and the sound-weighted score with a predetermined reference score and provides a comparison result; and a decision-making unit that determines, based on the comparison result, whether the train should be stopped. [9] Rail body defect detection system according to any one of claims 1 to 8, further comprising: a first reset unit that resets the position information to a last saved position before the malfunction of the rail body defect detection system when the rail body defect detection system is restarted due to a malfunction of the rail body defect detection system; a second reset unit that sets the position information to an initial value when the rail body defect detection system receives a start signal; and a third reset unit that resets the position information to the initial value when the rail body defect detection system receives an audio signal corresponding to the start signal. [10] Rail body defect detection system according to one of claims 1 to 9, wherein the detection unit further comprises an artificial intelligence unit that learns the rail body image, the position information and the sound signal and provides a result as to whether the rail body has a defect.

Images