A heavy haul railway corrugation identification method, device and equipment based on target detection

Abstract

The present application provides a heavy haul railway corrugation identification method, device and equipment based on target detection, comprising: sampling the time domain waveform of the acceleration of the wheel-rail contact surface when the heavy haul train runs through the heavy haul railway; coloring the time domain waveform according to the acceleration threshold to obtain a colored image; introducing the colored image into a corrugation identifier to determine the corrugation of the heavy haul railway; wherein the corrugation identifier is obtained by training the Faster-RCNN model, the coloring of the time domain waveform is realized through the acceleration threshold, the colored image can be led to the corrugation identifier, the corrugation on the heavy haul railway is identified through the color of the colored image, the target detection different from the deep learning convolution network is realized, and the corrugation identification accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of rail transit technology, and can be applied to the field of heavy haul railway quality detection, in particular to a heavy haul railway corrugation identification method, device, equipment and medium based on target detection. BACKGROUND

[0002] Railway corrugation, also known as corrugation, refers to the periodic wave-shaped irregularity phenomenon on the surface of the rail, which is one of the most urgent problems to be solved in heavy haul railway maintenance and repair work.

[0003] China's heavy haul railway is mainly for coal transportation. For a long time, heavy haul railways have been plagued by rail corrugation. Due to the high total weight of the trains running on heavy haul railways, the high volume and high traffic density, rail corrugation occurs quite commonly and develops rapidly, especially in small-radius curve sections.

[0004] When rail corrugation occurs, the vertical vibration of the wheelset caused by rail irregularities increases, causing abnormal stress on the wheelset, which damages the rail and even destroys the track subgrade, greatly reducing the service life of the rail and seriously affecting the economic benefits and operational safety of heavy haul railways.

[0005] Therefore, accurate positioning of the heavy haul rail corrugation section is an important prerequisite for heavy haul railway maintenance and repair, which helps to timely discover and rectify defect locations and is of great significance to improving the safety of China's heavy haul railway transportation.

[0006] In the prior art, deep learning methods have been applied in the field of rail corrugation research, such as a subway rail corrugation identification method based on one-dimensional convolutional neural networks. However, this method is suitable for subways, as the load of subways is lighter than that of heavy haul rails, and the frequency of corrugation occurrence is lower. Moreover, the image used in the subway rail corrugation identification method based on one-dimensional convolutional neural networks is a grayscale image, and the result obtained is usually whether there is corrugation in a section, but it cannot further determine the location of the corrugation, resulting in low accuracy of corrugation identification, which is suitable for scenarios with low requirements for the accuracy of corrugation identification.

[0007] However, the running scenario of heavy haul rails is to be solved in this paper. Since the load of heavy haul rails is the largest in the rail transit system, corrugation occurs more frequently. When the degree of corrugation is severe, accidents may occur on heavy haul rails. Therefore, a high-precision rail corrugation identification method is needed to solve the problem of low accuracy of corrugation identification in the prior art. SUMMARY

[0008] In view of the above problems of the prior art, the purpose of the present document is to provide a target detection-based heavy haul railway corrugation identification method, device, equipment and medium to solve the problem of low accuracy of rail corrugation identification in the prior art.

[0009] To solve the above technical problems, the specific technical solutions of the present document are as follows:

[0010] On the one hand, the present document provides a target detection-based heavy haul railway corrugation identification method, comprising:

[0011] Sampling the time-domain waveform of the acceleration of the wheel-rail contact surface when the heavy haul train passes through the heavy haul railway;

[0012] Coloring the time-domain waveform according to the acceleration threshold to obtain a colored image;

[0013] Importing the colored image into a corrugation identifier to determine the corrugation of the heavy haul railway;

[0014] Wherein, the corrugation identifier is obtained by training the Faster-RCNN model.

[0015] As an embodiment of the present document, before importing the colored image into the corrugation identifier to determine the corrugation of the heavy haul railway, it comprises:

[0016] Cutting the colored image according to the length of the heavy haul railway to obtain several pieces of the colored image and importing them into a classification network;

[0017] The classification network divides several pieces of the image into normal sections, partial corrugation sections and full corrugation sections according to the acceleration threshold;

[0018] Importing the colored image corresponding to the partial corrugation section into the corrugation identifier.

[0019] As an embodiment of the present document, before importing the colored image into the corrugation identifier model to determine the corrugation of the heavy haul railway, it comprises:

[0020] Using geometric transformation method on the colored image to obtain enhanced colored image;

[0021] Importing the enhanced colored image into the corrugation identifier.

[0022] As an embodiment of the present document, the method further comprises:

[0023] According to the length of the colored image after cutting, determining the positions of the normal section, the partial corrugation section and the full corrugation section on the heavy haul railway.

[0024] As an embodiment herein, the coloring the time-domain waveform according to the acceleration threshold value further comprises:

[0025] The acceleration threshold value comprises a first vertical acceleration threshold value and a second vertical acceleration threshold value;

[0026] In the time-domain waveform, when the absolute value of the vertical acceleration of the wheel-rail contact surface is greater than zero and less than the first vertical acceleration threshold value, the vertical acceleration is colored as a first color;

[0027] When the absolute value of the vertical acceleration of the wheel-rail contact surface is greater than the first vertical acceleration threshold value and less than the second vertical acceleration threshold value, the vertical acceleration is colored as a second color;

[0028] When the absolute value of the vertical acceleration of the wheel-rail contact surface is greater than the second vertical acceleration threshold value, the vertical acceleration is colored as a third color.

[0029] As an embodiment herein, the importing the colored image to a corrugation identifier to determine the corrugation of the heavy haul railway further comprises:

[0030] The corrugation identifier identifies the colored image through a candidate box;

[0031] The colored image including the first color, the second color and the third color in the candidate box is determined as the corrugation of the heavy haul railway.

[0032] As an embodiment herein, the method comprises:

[0033] Determining the length of the section corresponding to the heavy haul railway of the colored image;

[0034] Determining the length of the candidate box according to the vertex coordinates of the candidate box;

[0035] Determining the proportional relationship between the length of the candidate box and the length of the section corresponding to the heavy haul railway of the colored image;

[0036] Determining the position of the corrugation in the heavy haul railway according to the proportional relationship and the position of the partial corrugation section in the heavy haul railway.

[0037] In another aspect, the present disclosure also provides a heavy haul railway corrugation identification device based on target detection, comprising:

[0038] A sampling unit for sampling the time-domain waveform of the acceleration of the wheel-rail contact surface when a heavy haul train passes through a heavy haul railway;

[0039] A coloring unit for coloring the time-domain waveform according to an acceleration threshold value to obtain a colored image;

[0040] The identification unit is configured to input the colored image into a corrugation identifier to determine the corrugation of the heavy haul railway, wherein the corrugation identifier is trained by a Faster-RCNN model.

[0041] In another aspect, the present disclosure provides a computer device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the target detection-based heavy haul railway corrugation identification method according to any one of the embodiments when executing the computer program.

[0042] In another aspect, the present disclosure provides a computer-readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the target detection-based heavy haul railway corrugation identification method according to any one of the embodiments.

[0043] According to the above technical solution, the coloring of the time-domain waveform is realized by using the acceleration threshold value, and the colored image can be input into the corrugation identifier to identify the corrugation on the heavy haul railway by using the color of the colored image, so that the target detection different from the deep learning convolution network is realized, and the corrugation identification accuracy is improved.

[0044] In order to make the above and other objects, features and advantages of the present disclosure more apparent, specific preferred embodiments will be described in detail below with the accompanying drawings, and will be described in detail as follows. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without any creative work on the basis of these drawings.

[0046] Figure 1 Fig. 1 shows a whole system diagram of a target detection-based heavy haul railway corrugation identification method according to an embodiment of the present disclosure.

[0047] Figure 2 Fig. 2 shows a step schematic diagram of a target detection-based heavy haul railway corrugation identification method according to an embodiment of the present disclosure.

[0048] Figure 3 Fig. 3 shows a colored image schematic diagram of a target detection-based heavy haul railway corrugation identification method according to an embodiment of the present disclosure.

[0049] Figure 4 Fig. 4 shows a partial corrugation section identification result diagram of a target detection-based heavy haul railway corrugation identification method according to an embodiment of the present disclosure.

[0050] Figure 5 A pre-filtering frequency domain feature map of a target detection-based heavy haul railway corrugation identification method is shown;

[0051] Figure 6 A post-filtering frequency domain feature map of a target detection-based heavy haul railway corrugation identification method is shown;

[0052] Figure 7 A pre-filtering and post-filtering time domain acceleration comparison map of a target detection-based heavy haul railway corrugation identification method is shown;

[0053] Figure 8 A schematic diagram of a target detection-based heavy haul railway corrugation identification device is shown;

[0054] Figure 9 A detailed schematic diagram of a target detection-based heavy haul railway corrugation identification device is shown;

[0055] Figure 10 A coloring unit schematic diagram of a target detection-based heavy haul railway corrugation identification device is shown;

[0056] Figure 11 A data flow diagram of a target detection-based heavy haul railway corrugation identification overall system is shown;

[0057] Figure 12 A computer device schematic diagram is shown.

[0058] Explanation of the drawing symbols:

[0059] 101, database;

[0060] 102, operation server;

[0061] 103, control terminal;

[0062] 801, sampling unit;

[0063] 802, coloring unit;

[0064] 803, identification unit;

[0065] 901, cutting unit;

[0066] 902, classification unit;

[0067] 8021, first color module;

[0068] 8022, second color module;

[0069] 8023, third color module;

[0070] 1202, computer device;

[0071] 1204, processor;

[0072] 1206, memory;

[0073] 1208, drive mechanism;

[0074] 1210, input / output module;

[0075] 1212, input device;

[0076] 1214, output device;

[0077] 1216, presentation device;

[0078] 1218, graphical user interface;

[0079] 1220, network interface;

[0080] 1222, communication link;

[0081] 1224, communication bus. DETAILED DESCRIPTION

[0082] The technical solutions in the embodiments herein will be described clearly and completely below with reference to the accompanying drawings in the embodiments herein. Obviously, the described embodiments are only part of the embodiments herein, rather than all the embodiments. Based on the embodiments herein, all the other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection herein.

[0083] It should be noted that the terms "first", "second", and the like in the specification and claims of the present document and the above-described accompanying drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present document described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, device, product, or apparatus that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to the process, method, product, or apparatus.

[0084] Today's rail transit trains include subways, trams, maglev trains, high-speed trains and heavy haul trains, etc. Each train runs on its corresponding railway, and China's heavy haul trains are mainly for coal transportation, and in order not to affect people's daily travel, the railway where the heavy haul train runs is often in a place where people rarely go. Due to the large land area of China, the distance between coal-producing areas and demand areas is far, so it is extremely important to ensure the safety of heavy haul trains in a long driving distance.

[0085] Among the factors affecting the safety of heavy haul trains, in addition to unexpected accidents and irresistible factors, there is also the influence of heavy rail quality on heavy haul trains.

[0086] In the process of heavy load heavy haul train vehicle braking and starting, due to the aggravation of wheel-rail interaction, the steel rail wave-shaped loss (referred to as wave abrasion) is serious. Wave abrasion can cause a series of problems, such as abnormal vibration and noise pollution of the vehicle track, and even in areas with serious wave abrasion, the up and down acceleration caused by wave abrasion may damage the roadbed, greatly reducing the service life of the steel rail, and may also cause vehicle axle hanger fracture, primary steel spring fatigue fracture, and steel rail fastener spring bar fracture, etc. At present, grinding the rail is one of the main means to control wave abrasion, and before grinding, it is very important to accurately identify the state of the rail and determine the wave abrasion in heavy haul railway. After determining the wave abrasion, the location of the wave abrasion can be determined through GPS or time domain signal, and the track maintenance personnel are notified to maintain the wave abrasion, greatly improving the efficiency of wave abrasion maintenance and increasing the safety of heavy haul railway.

[0087] In summary, the existing deep learning method has been applied in the field of rail wave abrasion research, but the existing method mainly uses classification networks such as support vector machine, convolution and neural network, and the classification network has low accuracy in identifying existing rail wave abrasion, and needs to obtain clear rail field images. Due to the sectional nature of the classification network, it cannot accurately locate the position of the wave abrasion, but can only roughly give the interval where the wave abrasion occurs, so there is an urgent need for a method that can accurately identify wave abrasion.

[0088] As Figure 1 shown, a whole system diagram of heavy haul railway wave abrasion identification based on target detection includes a database 101, an operation server 102 and a control terminal 103.

[0089] The database 101 is used to store the time-domain waveform of the vertical acceleration of a certain section of heavy haul railway when the heavy haul train and the heavy haul railway rub against each other. The acceleration waveform is non-uniform. When the heavy haul railway has corrugation, the heavy haul train will be jolted due to the corrugation, so the peak value of the corresponding time-domain waveform is large, and the peak value is related to the degree of corrugation. When the heavy haul train travels on a normal section of heavy haul railway, the corresponding time-domain waveform is relatively flat because the surface of the heavy haul railway is relatively smooth. The database 101 is connected to the axle box sensor on the heavy haul train. The connection can be wireless or wired. The database 101 can be configured on each heavy haul train, or it can be configured in the command center of the heavy haul train through wireless transmission. The topology of the database 101 is not limited in this article.

[0090] The operation server 102 is used to obtain the time-domain waveform. It should be noted that the time-domain waveform can be a band-pass filtered waveform or an original time-domain waveform of the heavy haul railway. After the operation server receives the original time-domain waveform, it needs to be band-pass filtered to remove the noise and noise in the original time-domain waveform before identifying the corrugation. Of course, the time point of the operation process of the operation server 102 and the time point of the time-domain waveform acquisition process can be different. For example, the time when the database 101 records the time-domain waveform is November 24, and the time when the operation server 102 identifies the corrugation of the heavy haul railway corresponding to the time-domain waveform is November 26. The time when the database 101 records the time-domain waveform and the time when the operation server 102 identifies the time-domain waveform is not limited in this article.

[0091] The control terminal 103 is used to control the operation server 102 to obtain which section of the time-domain waveform, and can be configured in the command center of the heavy haul railway. For example, the first time-domain waveform is the time-domain waveform of the heavy haul railway from city A to city B, and the second time-domain waveform is the time-domain waveform of the heavy haul railway from city C to city D. When the command center needs to remove the corrugation on the heavy haul railway from city C to city D, the command center controls the operation server 102 to obtain the second time-domain waveform through the control terminal 103. The command center directs the heavy haul railway maintenance personnel to the designated location to repair the heavy haul railway according to the result identified by the operation server 102.

[0092] It should be noted that the corrugation identification method described in this article can be applied to any type of rail transit, not just heavy haul railways. Those skilled in the art can identify other types of rails based on the method described in this article. In order to reduce the length, this article will not be described again. In order to make the method described in this article more detailed, only the heavy haul railway is described in the entire identification method.

[0093] In the process of heavy haul train running, it is often affected by the wave-shaped wear of the steel rail (referred to as wave wear). When the wave wear is strong, it can even cause damage to the heavy haul railway subgrade and heavy haul train wheelset, seriously affecting the economic benefits and operation safety of the heavy haul railway. In the prior art, the accuracy of identifying wave wear by using a convolutional network is low, and only whether wave wear exists in a certain section can be identified. It is suitable for light rail or subway with low load and light wave wear. Affected by the image gray scale, the position of the wave wear in the heavy haul railway section cannot be located. Therefore, how to more accurately identify and locate the wave wear is a problem to be solved.

[0094] To solve the above problems, the embodiment of the present application provides a heavy haul railway wave wear identification method based on target detection, which can improve the accuracy of wave wear identification. Figure 2 It is a step schematic diagram of a heavy haul railway wave wear identification method based on target detection provided by the embodiment of the present application. The present specification provides the method operation steps as described in the embodiment or flowchart, but more or fewer operation steps can be included based on conventional or non-creative labor. The order of steps listed in the embodiment is only one of the many step execution orders, and does not represent the only execution order. In actual system or device product execution, the method order can be executed in sequence or in parallel as shown in the embodiment or the drawing. Specifically, as shown in the drawing, the method can include: Figure 2

[0095] Step 201, sampling the time domain waveform of the acceleration of the wheel-rail contact surface when the heavy haul train runs through the heavy haul railway.

[0096] Step 202, coloring the time domain waveform according to the acceleration threshold to obtain a colored image.

[0097] Step 203, introducing the colored image into a wave wear identifier to determine the wave wear of the heavy haul railway, wherein the wave wear identifier is obtained by training a Faster-RCNN model.

[0098] It should be noted that the heavy haul railway wave wear identification method and device based on target detection of the present application can be used in the field of heavy haul railway and can also be used in the field of rail transportation other than heavy haul railway. The application field of the heavy haul railway wave wear identification method and device based on target detection is not limited in the present application.

[0099] ​The colored image in accordance with the running rule of the heavy haul train can be obtained by coloring the acceleration caused by the actual corrugation to the time domain waveform, and the corrugation on the heavy haul railway can be clearly identified by the interlacing of various colors on the image. Because the corresponding running mileage of the heavy haul train in the time domain waveform is difficult to distinguish by the naked eye in a one-dimensional gray image which color interlacing position exists corrugation, the colored image is imported into the corrugation identifier to determine the corrugation position in the time domain waveform through the target recognition function of the corrugation identifier, and the corrugation position on the heavy haul railway can be accurately positioned through the coordinates of the corrugation position.

[0100] It should be noted that the acceleration threshold is obtained through a large number of verification tests by designers, for example, driving the heavy haul train at the position where the corrugation exists to determine the acceleration of the axle box of the train. In order to facilitate description and be more intuitive, the axle box acceleration in this paper can be the vertical acceleration of the axle box. According to the physical knowledge, the vertical acceleration is the main acceleration that causes the vehicle to appear jolt, so the vertical acceleration of the axle box is mainly determined. For example, during the test process, the designer found that when the heavy haul train just started, the vertical acceleration of the axle box was greater than the first acceleration, and the speed was defined as the first vertical acceleration threshold. When the heavy haul train drives on the corrugation, the vertical acceleration of the axle box is always greater than the second acceleration, and the speed is defined as the second vertical acceleration threshold. Therefore, the acceleration threshold in this paper does not exist regular formulation. The acceleration threshold represents the natural law of the corrugation.

[0101] As an embodiment of this paper, before the colored image is imported into the corrugation identifier to determine the corrugation of the heavy haul railway in step 203, it includes:

[0102] The colored image is cut according to the length of the heavy haul railway to obtain a plurality of pieces of the colored image, and is imported into the classification network.

[0103] The classification network divides a plurality of pieces of the image into normal sections, partial corrugation sections and full corrugation sections according to the acceleration threshold.

[0104] The colored image corresponding to the partial corrugation section is imported into the corrugation identifier.

[0105] In this step, in order to improve the efficiency of corrugation identification, the colored image of the heavy haul railway is selectively imported into the corrugation identifier.

[0106] According to the average speed of the heavy-haul train, the colored image is cut into several equal-length colored images. For example, if the average speed of the heavy-haul train is 60 km / h, each piece of the colored image can be cut into 60 m. According to the time sampling frequency of the axle box acceleration, the horizontal coordinates of each piece of the colored image are divided into a certain number of points. For example, if the sampling frequency of the axle line acceleration is 2000 Hz, each piece of the colored image can be divided into 6000 points. According to the length and the number of points of the colored image, the pixel of the colored image is determined to be 300*500. Since there are at least two acceleration thresholds in this paper, the colored image includes at least three colors, so the dimension of the colored image is at least 300*500*3. By increasing the dimension of the image, the ability of the wave abrasion recognizer to identify wave abrasion can be enhanced to solve the problem of low accuracy of the convolutional network in identifying the segmented wave abrasion interval.

[0107] Each piece of the colored image is imported into a classification network, which can roughly classify the colored image according to the color of the colored image. Since the acceleration threshold in this paper divides the colored image into three colors, for convenience of explanation, the three colors are defined as the first color, the second color and the third color. For example, the first color corresponds to the color of the heavy-haul train running on the normal heavy-haul railway, the second color is the color when the heavy-haul train just contacts the wave abrasion, and the third color is the color when the heavy-haul train runs on the wave abrasion for a period of time.

[0108] For example, when a piece of the colored image has only the first color, the corresponding heavy-haul railway is normal.

[0109] When a piece of the colored image has the first color, the second color and the third color on the entire horizontal axis, the corresponding heavy-haul railway is full wave abrasion, which is regarded as a full wave abrasion section. According to the previous cutting method, the 60 m range of the heavy-haul railway is all wave abrasion.

[0110] When a piece of the colored image has the first color on a part of the horizontal axis and the first color, the second color and the third color on the remaining part, the corresponding heavy-haul railway is partially wave abrasion, and the remaining part is normal, which is regarded as a partially wave abrasion section.

[0111] The classification network only needs to classify the full corrugation section and the partial corrugation section, and send the partial corrugation section to the corrugation identifier. Because the length of the full corrugation section is fixed, when cutting the colored image, each colored image has its label. For example, the section interval of heavy-duty railway No. A is between 1000 meters and 1060 meters. Of course, the coordinates can be used to identify the heavy-duty railway section corresponding to each colored image, or the specific numerical length can be used to identify the heavy-duty railway section corresponding to each colored image. In this way, the efficiency of heavy-duty railway operation and maintenance personnel in handling corrugation sections can be greatly improved.

[0112] Although each section is relatively short for the entire heavy-load railway, it still takes a lot of operation and maintenance time for the operation and maintenance personnel to complete the removal. Therefore, it is necessary to accurately identify the corrugation positions in some corrugation sections and remove them.

[0113] As an embodiment of this invention, the step of obtaining a plurality of colored images and importing them into the classification network further includes:

[0114] A geometric transformation method is applied to several slices of the shaded image to obtain an enhanced shaded image.

[0115] The enhanced colored image is fed into the classification network.

[0116] In this step, in order to eliminate the sample asymmetry problem caused by the fact that the corrugation in some corrugation sections accounts for a small proportion of the entire section, it is necessary to enhance the corrugation image. In target recognition, the corrugation image is used as a negative sample, that is, a defect sample, for data enhancement. The main means used include geometric transformation method, which uses horizontal flipping, up and down flipping, diagonal flipping, up and down displacement, and positive and negative 1.25 times lateral scaling of the image in the geometric transformation method to enhance the image greater than the second occurrence acceleration threshold.

[0117] The number of enhanced negative samples is not only comparable to that of positive samples in magnitude, but also the richness of samples is improved, which is conducive to the training and learning of the corrugation identifier.

[0118] Similarly, because the length of the normal section is much longer than the length of the corrugated section in the entire heavy-load railway, the same method can be used to enhance the sample size of the full corrugated section and the partial corrugated section in the time domain waveform corresponding to the entire heavy-load railway, so that the number of negative samples is the same as the number of positive samples, for example, the sum of the number of full corrugated sections and the number of partial corrugated sections is the same as the number of normal sections.

[0119] As an embodiment of this invention, the method further comprises determining positions of the normal section, the partial corrugation section and the full corrugation section on the heavy-haul railway according to the length of the colored image after cutting.

[0120] In this step, because the cutting size has been determined when cutting the colored image, for example, each cut image is 60m, that is, the length and position coordinates corresponding to the colored image are marked after cutting, and the type of the colored image is determined, such as a normal segment, a partial corrugated segment, and a full corrugated segment.

[0121] When the rendered image of the piece is a normal segment, no processing is required.

[0122] When the colored image is a full corrugated section, the identifier of the colored image is derived.

[0123] When the colored image is a partial corrugation section, the identifier of the colored image is derived, and the colored image is imported into the corrugation identifier.

[0124] Through this method, the efficiency of corrugation recognition can be enhanced, the calculation time can be greatly reduced, and timeliness can be achieved.

[0125] like Figure 3 A colored image schematic diagram of a method for identifying heavy-haul railway corrugation based on target detection is shown. As an embodiment of this invention, coloring the time domain waveform according to the acceleration threshold further includes:

[0126] The acceleration threshold includes a first vertical acceleration threshold and a second vertical acceleration threshold;

[0127] In the time domain waveform, when the absolute value of the vertical acceleration of the wheel-rail contact surface is greater than zero and less than the first vertical acceleration threshold, the vertical acceleration is colored in a first color;

[0128] When the absolute value of the vertical acceleration of the wheel-rail contact surface is greater than the first vertical acceleration threshold and less than the second vertical acceleration threshold, coloring the vertical acceleration into a second color;

[0129] When the absolute value of the vertical acceleration of the wheel-rail contact surface is greater than the second vertical acceleration threshold, the vertical acceleration is colored in a third color.

[0130] It should be noted that in this step, the first color can be defined as blue, and the first vertical acceleration threshold can be 3g, the second color can be defined as green, and the second vertical acceleration threshold can be 8g, the third color can be defined as red, and the RGB three-color diagram composed of blue, green and red can clearly represent the time domain waveform of the heavy haul railway, and because the contour boundary of the time domain waveform is distinguished from the background by color and shape, the capture success rate of the target detection model Faster-RCNN model is higher, and the accuracy is higher. Through the data characteristics of the contour shape and the wave grinding position of the time domain waveform, the wave grinding can be distinguished more accurately than the existing convolutional network.

[0131] It should be noted that the first vertical acceleration threshold and the second vertical acceleration threshold can be adjusted according to the actual measurement data, and this paper only makes an exemplary description, and the specific numerical value is not limited in this paper.

[0132] As an embodiment of the present paper, the step 203 imports the colored image into the wave grinding identifier to determine the wave grinding of the heavy haul railway, further comprising:

[0133] The wave grinding identifier identifies the colored image through the candidate box.

[0134] The colored image including the first color, the second color and the third color in the candidate box is determined as the wave grinding of the heavy haul railway.

[0135] In this step, the wave grinding identifier is based on the Faster-RCNN model training, and the candidate box is used to identify the wave grinding waveform, and can make it possible to directly see the wave grinding position.

[0136] Because one of the factors that restrict the accuracy of wave grinding identification in this paper is the candidate box of the wave grinding identifier, this paper illustrates how to train a more accurate candidate box of the wave grinding identifier.

[0137] Compared with the deep learning classification network, the target detection gives the understanding of the foreground and background of the input image, that is, to separate the target to be detected from the background, and to determine the type and position of the target. In the present application, the background is the background corresponding to the time domain waveform, and the foreground is the time domain waveform itself, because the time domain waveform in this paper has been colored, so it is more obvious relative to the background.

[0138] In training the model, the training set and test set need to be constantly input into the Faster-RCNN, for example, the user can input the colored image corresponding to the partial wave abrasion section into the Faster-RCNN model as the training set, and input the colored image with the candidate box identified into the Faster-RCNN model as the test set, and through the feature extraction, candidate box extraction, and bounding box regression, the accuracy of the target detection is improved.

[0139] At the same time, the Faster-RCNN includes the bounding box loss into the loss function of the whole model:

[0140]

[0141] Where i is the index information of the generated candidate box, p i is the predicted probability of the candidate box i being the target object, and here the target object to be detected is the wave abrasion time domain waveform in the partial wave abrasion section. is the reference standard of the classification category of the target object. t i is the vector containing 4 position coordinates of the predicted candidate box. is the reference standard of the target object coordinates. L cls is the classification loss, which judges whether the target object is foreground or background. L reg is the candidate box regression loss. means that if the target object is predicted as foreground, the corresponding candidate box regression loss is calculated. N cls is the number of classification categories. N reg is the number of candidate boxes in which the target object is predicted as foreground. λ is the weight coefficient, which adjusts the order of magnitude balance of the classification loss and the regression loss. By adjusting the details of the candidate box offset, more accurate target position information can be obtained. In addition, the Faster-RCNN model has the advantage that compared with the convolutional network, the Faster-RCNN model has low restriction on the size of the input image, and the model can adjust the size of the input image through zero padding and pooling, so it is very suitable for training the model with colored images as input data, and can further identify the wave abrasion target of the colored image corresponding to the partial wave abrasion section identified by the classification network.

[0142] The Faster-RCNN model is exemplarily given in this paper to compare the improvement effect of the recognition accuracy of the prior art, and to obtain which parameters of the Faster-RCNN model need to be adjusted by the wave abrasion recognizer.

[0143] The various parameters in the Faster-RCNN model are set as shown in Table 1.

[0144] Table 1

[0145] Parameter Value Learning rate 0.0001 Learning rate decay coefficient 0.3 Learning rate decay period 10 Training times 50

[0146] Using the above parameter setting, the Faster-RCNN model is completed, that is, the rail corrugation identifier is obtained, and the partial rail corrugation section is identified, and the effect shown in Table 2 can be obtained.

[0147] Table 2

[0148] Evaluation index Value Mean average precision 0.933 Convergent loss 0.05 Convergent loss standard deviation 0.0076

[0149] The results show that the Faster-RCNN model for target detection using the optimal parameters is trained for 50 times, and the cross-validation in the validation set can achieve a mean average precision of 93.3%, the loss converges to 0.05 and has good stability, so the rail corrugation identifier based on the Faster-RCNN model can be used for target recognition of the partial rail corrugation section of the heavy haul railway.

[0150] Similarly, in order to more intuitively represent the technical effects of the classification network in the present application, the training process of the classification network and the final effect are exemplarily described.

[0151] The feature data set of the colored and segmented time domain waveform data is used as input, and the rail corrugation categories of the heavy haul railway are used as output, for example, normal section, full corrugation section and partial corrugation section.

[0152] In the present application, the feature data set input into the classification network contains three types of samples, namely normal section, partial corrugation section and full corrugation section, and the classification of the axle box vertical acceleration data is constructed according to the training set and validation set ratio of 7:3 in the feature data set. After multiple experimental parameter adjustment, the optimal parameter values of the classification network are obtained as shown in Table 3.

[0153] Table 3

[0154] Parameter Value Learning rate 0.0001 Learning rate decay coefficient 0.3 Learning rate decay period 30 Training times 150

[0155] The segmented colored image is classified using the above parameter-adjusted classification network, and the results shown in Table 4 are obtained.

[0156] Table 4

[0157] Evaluation index Value Average precision 0.98 Average recall 0.967 Convergent loss 0.02 Convergent loss standard deviation 0.0063

[0158] According to the test results on the verification set, the classification network with the optimal parameters can reach a prediction accuracy of 98%, a recall rate of 93.3%, and a loss convergence of 0.02 and good stability after 150 training and cross-validation in the verification set. Therefore, the classification network can be used for classification and identification of the entire section of the heavy haul railway, and can accurately classify part of the corrugated section and further identify it using the target detection Faster-RCNN model.

[0159] As Figure 4 A partial corrugated section identification result diagram of a heavy haul railway corrugation identification method based on target detection, as an embodiment of the present text, the method further comprises:

[0160] Determine the length of the colored image representing the heavy haul railway.

[0161] Determine the length of the candidate box according to the vertex coordinates of the candidate box.

[0162] Determine the proportional relationship between the length of the candidate box and the length of the colored image.

[0163] According to the proportional relationship and the position of the partial corrugated section in the heavy haul railway, determine the position of the corrugated waveform in the heavy haul railway.

[0164] In Figure 4 , the abscissa is the mileage of the colored image corresponding to the section of the heavy haul railway, such as 386.7km to 387.2km. In order to facilitate the selection of the candidate box, the coordinates of the candidate box may not be consistent with the abscissa of the colored image. When the coordinates of the candidate box of the colored image are (10, -20), (60, -20), (60, 20) and (10, 20) in counterclockwise order from the lower left corner, the width of the candidate box can be calculated as 50. When the maximum value of the candidate box coordinate system is 100, then according to the proportion 387.2-386.7 / 50 / 100, the length of this section is 0.25km, and the corresponding corrugated position in the entire heavy haul railway is 386.7km-(386.7+0.25)km.

[0165] Through the proportional relationship between the length of the candidate box and the length of the colored image, and the position of the colored image in the entire heavy haul railway, the corrugated position can be clearly obtained.

[0166] As Figure 5 A filter before frequency domain feature map of a heavy haul railway corrugation identification method based on target detection, as an embodiment of the present text, the time domain waveform obtained by the axle box can also be filtered, which specifically includes:

[0167] Use Fourier transform on the original time domain waveform to obtain the frequency domain waveform.

[0168] Bandpass filtering is performed on the frequency domain waveform to obtain a denoised frequency domain waveform.

[0169] The denoised frequency domain waveform is subjected to inverse Fourier transform to obtain a time domain waveform.

[0170] In this step, if Figure 5 As shown in the figure, by performing a Fourier transform on the axlebox vertical acceleration data, we can see that the data energy is distributed across different frequency bands. However, the characteristics we want to observe cannot be observed solely from the raw data. We need to filter the data in the frequency domain and conduct targeted data analysis. Using the Fourier transform and the inverse Fourier transform to construct a bandpass filter, we can automatically complete the frequency domain filtering process of the data given the upper and lower frequency limits of the passband.

[0171] Record the vertical acceleration data of the axle box x i ,i=0,1,2,…N-1, perform Fourier transform on it to obtain X(k) and its corresponding frequency f(k).

[0172]

[0173]

[0174] The wavelength of periodic wear of heavy-duty rails is generally between 40mm and 1000mm. Therefore, based on the transformation relationship between wavelength, frequency, and speed, we calculate the upper and lower limits of the frequency of the bandpass filter in the variable speed section, which are 15Hz to 500Hz. Then, according to the following formula, the inverse Fourier transform is used to obtain the vertical acceleration data of the axle box after bandpass filtering. The frequency components after filtering are as follows: Figure 6 The frequency domain characteristic diagram after filtering of a heavy-haul railway corrugation recognition method based on target detection is shown as follows:

[0175]

[0176] The filtered data contains the main data features of corrugation defects, such as Figure 7 A heavy-haul railway corrugation identification method based on target detection is shown in the comparison diagram of time domain acceleration before and after filtering. Therefore, in order to accurately identify corrugation, the axle box vertical acceleration data can be bandpass filtered.

[0177] By the above method, the accurate identification of the wave abrasion of the heavy haul railway can be realized, and according to the classification of the classification network, the identification efficiency can be increased, when a higher precision identification result is required, the clutter and noise in the time domain waveform can be removed through filtering means to obtain the identification result meeting the expected expectation, when the identification result is determined, the wave abrasion position can be determined according to the coordinates, and the operation and maintenance personnel are guided to remove the wave abrasion, thereby reducing the labor cost and improving the safety of the heavy haul railway.

[0178] As shown in Figure 8 A schematic diagram of a heavy haul railway wave abrasion identification device based on target detection, comprising:

[0179] The sampling unit 801 is configured to sample the time domain waveform of the acceleration of the wheel-rail contact surface when the heavy haul train passes through the heavy haul railway.

[0180] The coloring unit 802 is configured to color the time domain waveform according to the acceleration threshold to obtain a colored image.

[0181] The identification unit 803 is configured to input the colored image into a wave abrasion identifier to determine the wave abrasion of the heavy haul railway, wherein the wave abrasion identifier is trained by a Faster-RCNN model.

[0182] Through the above device, the actual condition of the heavy haul railway where the heavy haul train is located can be quickly obtained, and after the time domain waveform is obtained, the colored image can be determined according to the preset acceleration threshold, and when the colored image is determined, the image can be input into the wave abrasion identifier, and the wave abrasion of the heavy haul railway can be accurately identified by target recognition.

[0183] As shown in Figure 9 A detailed schematic diagram of a heavy haul railway wave abrasion identification device based on target detection, comprising:

[0184] The cutting unit 901 is configured to cut the colored image according to the length of the heavy haul railway to obtain a plurality of pieces of the colored image and input them into a classification network.

[0185] The classification unit 902 is configured to divide the plurality of pieces of the image into a normal section, a partial wave abrasion section and a full wave abrasion section according to the wave abrasion occurrence of the classification network.

[0186] The identification unit 803 is further configured to input the colored image corresponding to the partial wave abrasion section into the wave abrasion identifier.

[0187] As shown in Figure 10 A coloring unit schematic diagram of a heavy haul railway wave abrasion identification device based on target detection, comprising:

[0188] The acceleration threshold includes a first vertical acceleration threshold and a second vertical acceleration threshold;

[0189] The first color module 8021 is configured to color the vertical acceleration of the wheel-rail contact surface into a first color when the absolute value of the vertical acceleration is greater than zero and less than the first vertical acceleration threshold.

[0190] The second color module 8022 is configured to color the vertical acceleration of the wheel-rail contact surface into a second color when the absolute value of the vertical acceleration is greater than the first vertical acceleration threshold and less than the second vertical acceleration threshold.

[0191] The third color module 8023 is configured to color the vertical acceleration of the wheel-rail contact surface into a third color when the absolute value of the vertical acceleration is greater than the second vertical acceleration threshold.

[0192] The above-mentioned device can enhance the recognition effect of the time domain waveform according to the first vertical acceleration threshold and the second vertical acceleration threshold to obtain a colored image. The colored image can also be segmented and distinguished, and only the section where corrugation occurs is introduced into the corrugation identifier for target recognition, thereby increasing the efficiency of corrugation recognition and also increasing the precision and accuracy of corrugation recognition.

[0193] In order to make the data processing process clearer, this article also provides the data forwarding and processing process of interaction between data. Of course, those skilled in the art can adjust the data carrier according to actual needs, and this article does not limit this.

[0194] like Figure 11 The data flow diagram of the overall system for heavy-haul railway corrugation recognition based on target detection is shown, including:

[0195] Step 1101: Input the time domain waveforms of several sections of heavy-haul railway into the database 101.

[0196] Step 1102: Send a query to the control terminal 103 regarding the corrugation condition of heavy-haul railway A.

[0197] Step 1103 : The control terminal 103 sends a calculation instruction to the operation server 102 .

[0198] Step 1104 : The computing server 102 obtains the time domain waveform of heavy-haul railway A from the database 101 .

[0199] Step 1105 : The computing server 102 performs band-pass filtering and image processing to obtain a colored image.

[0200] Step 1106 : The computing server 102 performs data enhancement on the colored image and classifies the colored image to obtain a normal segment, a fully corrugated segment, and a partially corrugated segment.

[0201] Step 1107 : The computing server 102 sends the corresponding information of the entire corrugation section to the control terminal 103 .

[0202] Step 1108 : The computing server 102 performs target recognition on a portion of the model section, determines the corrugation position in the portion of the corrugation section, and sends the position to the control terminal 103 .

[0203] like Figure 12 As shown, a computer device provided in an embodiment of the present invention is shown. The computer device 1202 may include one or more processors 1204, such as one or more central processing units (CPUs), each of which may implement one or more hardware threads. The computer device 1202 may also include any memory 1206 for storing any type of information, such as code, settings, data, etc. For example, without limitation, the memory 1206 may include any one or more combinations of the following: any type of RAM, any type of ROM, a flash memory device, a hard disk, an optical disk, etc. More generally, any memory may use any technology to store information. Furthermore, any memory may provide volatile or non-volatile retention of information. Furthermore, any memory may represent a fixed or removable component of the computer device 1202. In one embodiment, when the processor 1204 executes associated instructions stored in any memory or combination of memories, the computer device 1202 may perform any operation of the associated instructions. The computer device 1202 also includes one or more drive mechanisms 1208, such as a hard disk drive mechanism, an optical disk drive mechanism, etc., for interacting with any memory.

[0204] Computer device 1202 may also include an input / output module 1210 (I / O) for receiving various inputs (via input devices 1212) and providing various outputs (via output devices 1214). A specific output mechanism may include a presentation device 1216 and an associated graphical user interface (GUI) 1218. In other embodiments, input / output module 1210 (I / O), input devices 1212, and output devices 1214 may not be included, and the computer device 1202 may simply be a computer device in a network. Computer device 1202 may also include one or more network interfaces 1220 for exchanging data with other devices via one or more communication links 1222. One or more communication buses 1224 couple the components described above together.

[0205] The communication links 1222 can be implemented in any manner, such as through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. The communication links 1222 can include any combination of hardwired links, wireless links, routers, gateway functionality, name servers, etc., governed by any protocol or combination of protocols.

[0206] Corresponding to the method of Figure 2 the embodiments herein further provide a computer readable storage medium, having stored thereon a computer program, which when executed by a processor, performs the steps of the above method.

[0207] The embodiments herein further provide a computer readable instruction, wherein when a processor executes the instruction, the program therein causes the processor to perform the method as shown in Figure 2 .

[0208] It should be understood that the size of the sequence number of the above processes in the various embodiments herein does not mean the order of execution, and the execution order of the processes should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments herein.

[0209] It should also be understood that in the embodiments herein, the term "and / or" is only a description of the association relationship of the associated objects, which means that there can be three relationships. For example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone. In addition, the character " / " in this paper generally represents that the front and rear associated objects are in an "or" relationship.

[0210] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in general terms in the above description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this paper.

[0211] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0212] In several embodiments provided herein, it should be understood that the disclosed system, apparatus and method can be implemented in other manners. For example, the apparatus embodiments described above are merely schematic. For example, the division of the units is only a logical function division. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electric, mechanical or other form.

[0213] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments provided herein.

[0214] In addition, each functional unit in each embodiment herein can be integrated into a processing unit, or each unit can be physically present separately, or two or more units can be integrated into one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0215] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions herein, essentially or the part of the prior art that contributes to the technical solutions, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment herein. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0216] The specific embodiments are applied herein to describe the principles and implementation manners of the embodiments herein. The above description of the embodiments is only to help understand the methods and core ideas thereof; meanwhile, for those skilled in the art, according to the ideas herein, the specific implementation manners and application ranges can be changed; and in view of the above, the content of the specification should not be understood as limiting the embodiments herein.

Claims

1. A heavy-haul railway corrugation identification method based on target detection, characterized in that: include: Sampling the time domain waveform of the acceleration of the wheel-rail contact surface when a heavy-haul train passes through a heavy-haul railway; Coloring the time domain waveform according to an acceleration threshold to obtain a colored image; Cutting the colored image according to the length of the heavy-haul railway to obtain a plurality of pieces of the colored image, and importing the pieces into a classification network; The classification network divides the plurality of colored images into normal segments, partial corrugation segments, and full corrugation segments according to the acceleration threshold; importing the colored image corresponding to the partial corrugation section into a corrugation identifier to further determine the corrugation position of the heavy-haul railway from the partial corrugation section; The corrugation identifier is obtained by training the Faster-RCNN model.

2. The method for identifying heavy-haul railway corrugation based on target detection according to claim 1, characterized in that: Before importing the colored image corresponding to the partial corrugation section into the corrugation identifier, the method includes: Applying a geometric transformation method to the shaded image corresponding to the partial corrugation section to obtain an enhanced shaded image; The enhanced colored image corresponding to the partial corrugation section is imported into the corrugation identifier.

3. The method for identifying heavy-haul railway corrugation based on target detection according to claim 1, characterized in that: The method further comprises: The positions of the normal section, the partial corrugation section and the full corrugation section on the heavy-haul railway are determined according to the length of the colored image after cutting.

4. The method for identifying heavy-haul railway corrugation based on target detection according to claim 3, characterized in that: The coloring of the time domain waveform according to the acceleration threshold further comprises: The acceleration threshold includes a first vertical acceleration threshold and a second vertical acceleration threshold; In the time domain waveform, when the absolute value of the vertical acceleration of the wheel-rail contact surface is greater than zero and less than the first vertical acceleration threshold, the vertical acceleration is colored in a first color; When the absolute value of the vertical acceleration of the wheel-rail contact surface is greater than the first vertical acceleration threshold and less than the second vertical acceleration threshold, coloring the vertical acceleration into a second color; When the absolute value of the vertical acceleration of the wheel-rail contact surface is greater than the second vertical acceleration threshold, the vertical acceleration is colored in a third color.

5. The method for identifying heavy-haul railway corrugation based on target detection according to claim 4, characterized in that: The step of importing the colored image corresponding to the partial corrugation section into a corrugation identifier to further determine the corrugation position of the heavy-haul railway from the partial corrugation section further comprises: The corrugation identifier recognizes the colored image through the candidate frame; The position of the colored image including the first color, the second color and the third color selected by the candidate box is determined as the corrugation position of the heavy-haul railway.

6. The method for identifying heavy-haul railway corrugation based on target detection according to claim 5, characterized in that: The method comprises: determining a section length of the shaded image corresponding to the heavy-haul railway; Determine the length of the candidate frame according to the vertex coordinates of the candidate frame; determining a proportional relationship between the length of the candidate frame and the length of the section of the colored image corresponding to the heavy-haul railway; The position of the corrugation mill in the heavy-haul railway is determined according to the proportional relationship and the position of the partial corrugation mill section in the heavy-haul railway.

7. A heavy-haul railway corrugation identification device based on target detection, characterized in that: include: A sampling unit, used for sampling the time domain waveform of the acceleration of the wheel-rail contact surface when the heavy-load train passes through the heavy-load railway; a coloring unit, configured to color the time domain waveform according to an acceleration threshold to obtain a colored image; a cutting unit, configured to cut the colored image according to the length of the heavy-haul railway to obtain a plurality of pieces of the colored image, and import the pieces into a classification network; a classification unit, configured for the classification network to classify the plurality of rendered images into normal segments, partially corrugated segments, and fully corrugated segments according to the acceleration threshold; The recognition unit is used to import the colored image into a corrugation identifier to further determine the corrugation position of the heavy-haul railway from the partial corrugation section; wherein the corrugation identifier is trained by a Faster-RCNN model.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the heavy-haul railway corrugation recognition method based on target detection according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for identifying heavy-haul railway corrugation based on target detection according to any one of claims 1 to 6 is implemented.

Images