A rail loss measurement and rail surface damage detection method and system

Abstract

The present invention discloses a rail loss measurement and rail surface damage detection method and system, which relates to the field of rail detection technology. The method involves acquiring BIM data of the rail, constructing a three-dimensional model of the rail, and obtaining the average curvature change value of the rail based on the three-dimensional model. If the value exceeds a set value, the rail is segmented accordingly. More than three sets of physical data acquisition devices are installed on the inspection vehicle for physical data acquisition, and at least one visual data acquisition device is installed on both sides for visual data acquisition. The processed three-dimensional model is used to adjust the angle of the data acquisition device and simultaneously correct the collected data. Based on the corrected collected data, rail loss data and rail damage data are output separately. A damage detection model is constructed and trained using the loss and damage data. The rail to be tested is then inspected based on the trained damage detection model. The present invention can improve the accuracy of rail damage detection in complex environments.

Description

Technical Field

[0001] The present invention relates to the technical field of rail detection, and in particular to a method and system for measuring rail loss and detecting rail surface damage. Background Art

[0002] As the core component of rail transportation, rails bear train loads, impacts and environmental effects for a long time, and may suffer damage such as wear, cracks, and breakage. Measuring the loss and damage of rails can not only ensure the safety of railway transportation, but also extend the service life of rails.

[0003] In rail defect detection, the traditional method is manual inspection. However, this method is gradually being replaced by other detection methods due to its low detection efficiency, high requirements, and the need for inspectors to have extensive practical experience. With the development of non-destructive testing and assessment technology, various non-destructive testing methods have been gradually applied to rail defect detection, and the detection effect has been significantly improved. Currently, common rail defect detection methods at home and abroad mainly include physical inspection methods and machine vision methods. Physical inspection methods mainly rely on relevant physical equipment to obtain different information such as the type, shape, location, and size of rail defects; machine vision methods extract shape features by performing pre-processing and image enhancement operations on the two-dimensional rail image, and then locate or segment the defects from the image.

[0004] However, the above methods have the following problems. Currently, most rail defect detection methods use a single method (physical inspection or machine vision) or a simple combination of the two methods. However, for rails in complex environments (rails with slopes or curves), their geometric shapes are complex, and changes in curvature may cause the coupling effect between the detection probe (physical inspection probe or camera) and the rail surface to deteriorate, resulting in blind spots in detection and large errors in detection data. In addition, it is impossible to efficiently combine physical inspection data and machine vision data, which ultimately affects the accuracy of rail damage detection. Summary of the Invention

[0005] In view of this, the present application provides a rail loss measurement and rail surface damage detection method and system to improve the accuracy of rail damage detection in complex environments.

[0006] The first aspect of the present application provides a method for measuring rail loss and detecting rail surface damage, comprising:

[0007] Acquire BIM data of the rail and establish a three-dimensional model of the rail;

[0008] Processing the three-dimensional model, specifically: obtaining the average curvature change value of the rail and determining whether it exceeds a first set value, if so, performing segmented processing on the three-dimensional model of the rail using a preset method; if not, taking no action;

[0009] M groups of physical data acquisition devices are set on the inspection vehicle, each group of physical data acquisition devices is used to output structural symmetrical laser for the rail and collect return data, M ≥ 3;

[0010] At least one visual data acquisition device is provided on each side of the rail on the inspection vehicle, and the visual data acquisition device is used to acquire image data of the rail;

[0011] Based on the processed 3D model, adjust the angles of all physical data acquisition devices and all visual data acquisition devices, and correct the corresponding return data and image data;

[0012] Outputting the rail loss data based on the corrected return data; outputting the rail damage data based on the corrected image data;

[0013] A damage detection model is constructed, and the damage detection model is trained using the loss data and damage data of the rail. Based on the trained damage detection model, damage detection is performed on the rail to be tested.

[0014] In a possible implementation of the first aspect, obtaining the average curvature change value of the rail includes:

[0015] Divide the rail into N sections at equal distances, and select three characteristic points for each of the N sections;

[0016] For each of the N rail segments, a fitting circle is constructed using three characteristic points, and the radius of the corresponding fitting circle is calculated, and the inverse of the radius of the fitting circle is used as the curvature of the corresponding rail segment. The three characteristic points are the starting point, midpoint, and end point of each rail segment.

[0017] Calculating the absolute value of the curvature difference between each adjacent rail section, and storing the absolute values ​​of all curvature differences in a first preset data set;

[0018] The average value of the absolute values ​​of all curvature differences in the preset data set is calculated to obtain the average curvature change value of the rail.

[0019] In a possible implementation of the first aspect, the preset method includes:

[0020] Obtaining the starting point and the end point of the rail, and obtaining a plurality of data points at equal distances on the rail;

[0021] Calculate the tangent point slopes of all data points, and process the tangent point slopes of all data points using the first method;

[0022] The first method is specifically: starting from the starting point, calculating the absolute value of the slope difference between the adjacent data points corresponding to the tangent points, and storing the calculated absolute value of the slope difference between the adjacent data points corresponding to the tangent points in a second preset data set;

[0023] Calculating an average of the absolute values ​​of the slope differences of the corresponding tangent points of all adjacent data points stored in the second preset data set until the corresponding average value exceeds a first threshold, obtaining a data point in the second preset data set that is farthest from the starting point as a first data point, and recording adjacent data points of the first data point close to the starting point as second data points;

[0024] The second data point is used as a new starting point, and the remaining data points in the rail are traversed using the first method until the end point of the rail is reached, thereby completing the segmentation of the rail. After segmentation, the average value of the difference in slopes of adjacent tangent points of each segment of the rail is less than the first threshold.

[0025] In a possible implementation of the first aspect, adjusting the angles of all physical data acquisition devices and all visual data acquisition devices using the processed three-dimensional model includes:

[0026] Based on the processed three-dimensional model, the starting point of each rail segment after segmentation is obtained and recorded as the first starting point, and the end point is recorded as the first end point;

[0027] Constructing an axis of each rail segment with the first starting point and the first end point, which is recorded as a first axis;

[0028] Obtaining the axis of the physical data acquisition device, recorded as the second axis;

[0029] Adjust the second axis to be parallel to the first axis to complete the adjustment of the angles of all physical data acquisition devices;

[0030] Obtaining the axis of the visual data acquisition device, recorded as the third axis;

[0031] The angle between the third axis and the first axis is adjusted to always remain at a preset angle, thereby completing the adjustment of the angles of all visual data acquisition devices.

[0032] In a possible implementation of the first aspect, correcting the corresponding returned data and image data includes:

[0033] Based on the processed 3D model, the starting point of each rail segment is obtained;

[0034] Constructing a data collection area, with the starting point of each rail section as the length of the data collection area and the width of the data collection area as a second set value;

[0035] Based on the data acquisition area, the return data and image data of each segmented rail are corrected, and the corrected return data and image data are located in the data acquisition area corresponding to each segmented rail.

[0036] In a possible implementation of the first aspect, outputting the rail loss data based on the corrected return data includes:

[0037] Obtain the standard profiles of the rail head, rail waist and rail bottom of each rail segment after segmentation;

[0038] The corrected returned data is spliced ​​to obtain the actual profiles of the rail head, rail waist and rail bottom of each section of rail;

[0039] Based on the standard and actual profiles of the rail head of each section, the loss value of the rail head is output; based on the standard and actual profiles of the rail waist of each section, the loss value of the rail waist is output; based on the standard and actual profiles of the rail foot of each section, the loss value of the rail foot is output.

[0040] In a possible implementation of the first aspect, outputting the rail damage data based on the corrected image data includes:

[0041] The image processed by the visual data acquisition device on one side of the inspection vehicle after correction is recorded as a first type of image, and the image processed by the visual data acquisition device on the other side after correction is recorded as a second type of image;

[0042] Segmenting the image of the type to obtain a rail head image, which is recorded as a first rail head image, a side rail waist image, which is recorded as a first rail waist image, and a side rail bottom image, which is recorded as a first rail bottom image;

[0043] Segmenting the two types of images to obtain a rail head image recorded as a second rail head image, a side rail waist image recorded as a second rail waist image, and a side rail bottom image recorded as a second rail bottom image;

[0044] selecting a plurality of identical feature points in both the first rail head image and the second rail head image, and matching the first rail head image with the second rail head image based on feature point matching to obtain a matching result;

[0045] Based on the matching result, determine whether the first rail head image and the second rail head image completely overlap. If so, use the first rail head image or the second rail head image as the rail head image set of the rail. If not, obtain the non-overlapping area between the first rail head image and the second rail head image, and set a weight coefficient for all pixels in the non-overlapping area according to the distance from the boundary of the first rail head image and the boundary of the second rail head image to perform weighted fusion. Splice the overlapping area between the first rail head image and the second rail head image and the non-overlapping area after weighted fusion to form the rail head image set of the rail.

[0046] The first rail waist image and the second rail waist image are used as a rail waist image set of the rail; the first rail bottom image and the second rail bottom image are used as a rail bottom image set of the rail.

[0047] In a possible implementation of the first aspect, training the damage detection model using the rail loss data and damage data includes:

[0048] Obtain the rail head loss value, rail waist loss value and rail bottom loss value of each rail segment after segmentation;

[0049] A second threshold is set for the loss value of the rail head, a third threshold is set for the loss value of the rail waist, and a fourth threshold is set for the loss value of the rail foot;

[0050] Determine whether the loss value of the rail head exceeds the second threshold value, and obtain a first determination result; determine whether the loss value of the rail waist exceeds the third threshold value, and obtain a second determination result; determine whether the loss value of the rail base exceeds the fourth threshold value, and obtain a third determination result;

[0051] Based on the first judgment result, the second judgment result, and the third judgment result, permutations and combinations are performed, and predicted damage data of the rail is output in seven situations;

[0052] The rail head image set, rail waist image set and rail bottom image set of the rail are all marked for damage, and the marked rail head image set, the marked rail waist image set and the marked rail bottom image set are obtained respectively;

[0053] Based on the labeled rail head image set, output the damage data of the rail head; based on the labeled rail waist image set, output the damage data of the rail waist; based on the labeled rail bottom image set, output the damage data of the rail bottom;

[0054] Integrating the damage data of the rail head, the damage data of the rail waist, and the damage data of the rail bottom into actual damage data of the rail;

[0055] Determine whether the predicted damage data of the rail is consistent with the actual damage data of the rail. If so, add a risk coefficient to the actual damage data of the rail, the risk coefficient is greater than 1, and the risk level of the rail is increased. If not, directly output the actual damage data of the rail.

[0056] In a possible implementation of the first aspect, performing damage detection on the rail to be tested based on the trained damage detection model includes:

[0057] The loss data and damage data of the rail to be tested are obtained and input into the trained damage detection model, and the damage detection result of the rail to be tested is output.

[0058] A second aspect of the present application provides a rail loss measurement and rail surface damage detection system, comprising:

[0059] A three-dimensional model unit, used to obtain BIM data of the rail and establish a three-dimensional model of the rail;

[0060] a model processing unit, configured to process the three-dimensional model, specifically: obtaining an average curvature change value of the rail and determining whether it exceeds a first set value; if so, performing segmented processing on the three-dimensional model of the rail using a preset method; if not, performing no action;

[0061] A physical data acquisition unit is configured to be provided with M groups of physical data acquisition devices on the inspection vehicle, each group of physical data acquisition devices being configured to output structurally symmetrical laser light for the rail and to collect return data, M being ≥ 3;

[0062] A visual data acquisition unit is configured to provide at least one visual data acquisition device on each side of the rail on the inspection vehicle, wherein the visual data acquisition device is configured to acquire image data of the rail;

[0063] A correction unit is used to adjust the angles of all physical data acquisition devices and all visual data acquisition devices based on the processed three-dimensional model, and to correct the corresponding return data and image data;

[0064] A data processing unit is configured to output the loss data of the rail based on the corrected return data; and output the damage data of the rail based on the corrected image data;

[0065] The damage detection unit is used to construct a damage detection model, and use the loss data and damage data of the rail to train the damage detection model, and perform damage detection on the rail to be tested based on the trained damage detection model.

[0066] Compared with the existing technology, the present application provides a method for measuring rail loss and detecting rail surface damage, including: first, obtaining the BIM data of the rail and establishing a three-dimensional model of the rail. If there is a BIM database about the rail, the BIM data of the rail is directly obtained from the BIM database. If there is no BIM data about the rail, the corresponding BIM data can be obtained through ground three-dimensional laser scanning.

[0067] The purpose of constructing a three-dimensional model is to more intuitively measure the curvature change (including curvature or bend) of the rail from the three-dimensional model. The three-dimensional rail model is then processed to obtain the average curvature change value of the rail. The average curvature change value is obtained by dividing the rail into equal-distance segments and then calculating the difference between the curvatures of adjacent segments. Compared with directly obtaining the curvature value of the rail, the average curvature change value can more accurately reflect the curvature of the rail. If the average curvature change value of the rail exceeds a first set value, it indicates that the curvature of the rail segment is complex and the three-dimensional rail model needs to be segmented using a preset method. The logic of the segmentation processing is to segment the rail with multiple curvature changes into rails that are approximately straight lines. This facilitates the corresponding correction of subsequent physical or visual data collection, thereby improving the accuracy of the data collection source. If the average curvature change value of the rail does not exceed the first set value, no action is taken.

[0068] M groups of physical data acquisition devices are set on the inspection vehicle. Each group of physical data acquisition devices is used to output structurally symmetrical lasers for the rails and collect return data. M ≥ 3. The reason for setting more than 3 groups of physical data acquisition devices is that compared with the previous method of only collecting loss data for the rail surface, the present invention collects loss data for the rail head, rail waist and rail bottom.

[0069] At least one visual data acquisition device is set on both sides of the rail on the inspection vehicle. When collecting rail images, due to the influence of blind spots, brightness or shadows, the images collected by relying solely on a single visual data acquisition device may be biased. Therefore, the present invention sets at least one visual data acquisition device on both sides of the rail, and uses the images on both sides for fusion processing to reduce errors caused by blind spots, brightness or shadows.

[0070] Using the processed 3D model, the angles of the physical data acquisition equipment and the visual data acquisition equipment are first adjusted. That is, according to the axis of each segmented rail (approximately a straight line), the physical data acquisition equipment is adjusted to be parallel to the rail axis, and the angle of the visual data acquisition equipment and the axis is kept consistent. This ensures that errors are reduced when collecting data; then the return data and image data are corrected. The correction logic is to set a data collection area corresponding to each segmented rail, and clean the data in the return data and image data that does not belong to the rail data collection area to further ensure the accuracy of data collection.

[0071] Based on the corrected return data, the rail loss data is output. That is, the actual contours of the rail head, rail waist and rail bottom are obtained through the return data, and compared with the corresponding standard contours to obtain the rail loss data; based on the corrected image data, the rail damage data is output. That is, after fusing the images on both sides of the rail, the influence of blind spots, brightness or shadows is reduced, and then the damage type, position and size in the image are manually marked to output the rail damage data.

[0072] A damage detection model is constructed and trained using rail loss and damage data. The training logic is that rail loss data not only reflects the existing rail loss, but also indicates the risk of damage at different locations (including combinations) (e.g., loss of the rail head alone, which can lead to wear / crushing and cracking risks at the head). The rail loss data is used to output predicted damage data, which is then used to output actual damage data. The predicted and actual damage data are then matched. If a successful match indicates a greater potential risk than the actual output damage data, the model is trained by increasing the risk level and adding a risk factor to the actual damage data (e.g., if the crack size on the rail head is 3 mm, adding a risk factor changes it to 4 mm). If the match fails, the actual damage data is directly output. Finally, damage detection is performed on the rail under test based on the trained damage detection model.

[0073] Its beneficial effects are as follows: first, by acquiring the BIM data of the rail, a three-dimensional model of the rail is constructed. Then, based on the three-dimensional model, the average curvature change value of the rail is obtained. This value is used to judge the complexity of the rail. If it exceeds a first set value, the rail is segmented, that is, the rail with complex curvature changes is segmented into multiple sections of approximately straight rails. If it does not exceed the first set value, no action is taken. More than three sets of physical data acquisition devices are installed on the inspection vehicle to collect physical data from the rail head, rail waist and rail bottom, and at least one visual data acquisition device is installed on each side of the rail. By fusing the images on both sides, the accuracy of the collected images is improved. The processed three-dimensional model is used to adjust the angles of the physical data acquisition device and the visual data acquisition device, and then the corresponding return data and image data are corrected to ensure the accuracy of data acquisition. The corrected return data is used to output the rail loss data, and the corrected image data is used to output the rail damage data. Finally, a damage detection model is constructed, using rail loss data to output predicted rail damage data. This data is then compared with the actual damage data output using the rail loss data to ultimately output the damage detection results. This invention effectively combines physical inspection data with machine vision data, improving the accuracy of rail damage detection in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0075] Figure 1 This is a flow chart of a method for measuring rail loss and detecting rail surface damage provided by an embodiment of the present application;

[0076] Figure 2 This is a schematic diagram of the composition of a rail loss measurement and rail surface damage detection system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0077] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0078] In this application, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0079] In order to better understand this application, the following are corresponding explanations of the technical names involved in this application:

[0080] Example 1

[0081] From the above background technology, it can be seen that rails, as core components of rail transportation, bear train loads, impacts and environmental effects for a long time, and may suffer damage such as wear, cracks, and breakage. Measuring the loss and damage of rails can not only ensure the safety of railway transportation, but also extend the service life of rails.

[0082] In existing technologies, rail defect detection mostly adopts a single method (physical inspection or machine vision) or a simple combination of the two methods. However, for rails in complex environments (rails with slopes or curves), their geometric shapes are complex, and changes in curvature may cause the coupling effect between the detection probe (physical inspection probe or camera) and the rail surface to deteriorate, resulting in blind spots in detection and large errors in detection data. It is also impossible to efficiently combine physical inspection data and machine vision data, ultimately affecting the accuracy of rail damage detection.

[0083] Therefore, this application provides a method for measuring rail loss and detecting rail surface damage, such as Figure 1 Shown, including:

[0084] Acquire BIM data of the rail and establish a three-dimensional model of the rail;

[0085] Processing the three-dimensional model, specifically: obtaining the average curvature change value of the rail and determining whether it exceeds a first set value, if so, performing segmented processing on the three-dimensional model of the rail using a preset method; if not, taking no action;

[0086] M groups of physical data acquisition devices are set on the inspection vehicle, each group of physical data acquisition devices is used to output structural symmetrical laser for the rail and collect return data, M ≥ 3;

[0087] At least one visual data acquisition device is provided on each side of the rail on the inspection vehicle, and the visual data acquisition device is used to acquire image data of the rail;

[0088] Based on the processed 3D model, adjust the angles of all physical data acquisition devices and all visual data acquisition devices, and correct the corresponding return data and image data;

[0089] Outputting the rail loss data based on the corrected return data; outputting the rail damage data based on the corrected image data;

[0090] A damage detection model is constructed, and the damage detection model is trained using the loss data and damage data of the rail. Based on the trained damage detection model, damage detection is performed on the rail to be tested.

[0091] The BIM data of the rails is obtained and a 3D model of the rails is created. If a BIM database for the rails is available, the BIM data can be directly obtained from the database. If no BIM data for the rails is available, the corresponding BIM data can be obtained through ground-based 3D laser scanning. Creating a corresponding 3D model based on BIM data is a technique well known to those skilled in the art and will not be further elaborated in this embodiment.

[0092] Among them, the three-dimensional model is processed to obtain the average curvature change value of the rail, specifically: the rail is divided into N sections at equal intervals, and then 3 feature points are selected in each section of the rail. A fitting circle is constructed with the 3 feature points for each section of the rail. It should be noted that the 3 feature points are the starting point, midpoint and end point of the rail respectively; then the radius of the fitting circle is calculated (for example, R), then the curvature of the section of the rail is 1 / R; the absolute value of the difference in curvature of adjacent sections of the rail is calculated. For example, if the rail is divided into 5 sections, the curvature of each section of the rail is 1 / R1, 1 / R2, 1 / R3, 1 / R4 and 1 / R5. The absolute values ​​of the curvature differences between adjacent rail segments are |1 / R1-1 / R2|, |1 / R2-1 / R3|, |1 / R3-1 / R4|, and |1 / R4-1 / R5|. All of these absolute values ​​are stored in a first preset data set. The average value of all absolute values ​​stored in the first preset data set is calculated, such as (|1 / R1-1 / R2|+|1 / R2-1 / R3|+|1 / R3-1 / R4|+|1 / R4-1 / R5|) / 4, to obtain the average curvature change value of the rail. The above data is for illustrative purposes only and can be adjusted according to actual needs. This embodiment does not impose specific limitations.

[0093] Among them, if the average curvature change value of the rail does not exceed the first set value, no action will be taken; if it exceeds the first set value, the preset method will be used to segment the three-dimensional model of the rail, specifically: multiple data points are obtained at equal distances on the rail, and the tangent slope of each data point is calculated, the absolute value of the difference in the tangent slope of adjacent data points starting from the starting point is calculated, and all absolute values ​​are stored in a second preset data set, and then the average value of the absolute values ​​stored in the second preset data set is calculated until the corresponding average value exceeds the first threshold, and the data point farthest from the starting point is obtained in the second preset data set and recorded as the first data point, and the adjacent data points of the first data point close to the starting point are recorded as the second data point, such as A1, A2, A3, ... and A from the starting point to the first data point respectively. n , and the corresponding tangent point slopes are K1, K2, K3, ... and K n , the average value is avg=(|K0-K1|+|K1-K2|+……|K n -K n-1 |) / (n-1), when avg exceeds the first threshold, A n is the first data point, A n-1 As the second data point, A n-1 Using this as the new starting point, the remaining data points are traversed until the end of the rail, thus completing the rail segmentation process. After segmenting the rail using this method, the average difference in the slopes of the corresponding tangent points of all adjacent data points in each rail segment is less than the first threshold, at which point the rail can be approximately considered a straight line. When subsequently correcting the return data collected by the physical data acquisition device and the image data collected by the visual data acquisition device, only the starting point of each rail segment is required to divide the data acquisition area, thereby achieving correction processing for the return data and image data.

[0094] Among them, M groups of physical data acquisition devices are set on the inspection vehicle. Each group of physical data acquisition devices is used to output structural symmetrical lasers for the rails and collect return data. M≥3. The reason for setting more than 3 groups of physical data acquisition devices is to collect the loss data of the rail head, rail waist and rail bottom respectively. Compared with the previous collection of only rail surface loss data, the collected data is more accurate. At the same time, the collected loss data of the rail head, rail waist and rail bottom can also more accurately output the potential damage risks of the rails.

[0095] Among them, at least one visual data acquisition device is set on both sides of the rail on the inspection vehicle. When collecting rail images, due to the influence of blind spots, brightness or shadows, the images collected by relying solely on a single visual data acquisition device may be biased. Therefore, the present invention sets at least one visual data acquisition device on both sides of the rail, and uses the images on both sides for fusion processing to reduce errors caused by blind spots, brightness or shadows.

[0096] Among them, based on the processed three-dimensional model, the physical data acquisition device and the visual data acquisition device are first adjusted in angle, that is, the starting point of each section of the rail after segmentation is obtained, and the axis of each section of the rail is obtained using the starting point, wherein the axis of the physical data acquisition device is adjusted to always remain parallel to the axis of the rail, and the angle between the axis of the visual data acquisition device and the axis of the rail is adjusted to remain unchanged. If the angle between the visual data acquisition devices on both sides and the axis of the rail cannot be kept stable, then the difference in viewing angle will cause the collected image to deviate accordingly. Then, the returned data and image data are corrected and processed. The correction logic is: obtain the starting point of each section of the rail after segmentation, then construct a data acquisition area, and use the starting point of each section of the rail to limit the length of the data acquisition area. The width of the data acquisition area can be set according to the actual size of the rail, and this embodiment does not make specific restrictions.

[0097] Among them, based on the corrected return data, the loss data of the rail is output, specifically: the standard profiles of the rail head, rail waist and rail bottom of each section of rail after segmentation are obtained, the corrected return data are spliced ​​to obtain the actual profiles of the rail head, rail waist and rail bottom of each section of rail, and based on the standard profile and actual profile of the rail, the loss value of the rail head, rail waist and rail bottom are output respectively.

[0098] Among them, after the corrected image data is processed, the damage data of the rail is output, specifically: the image processed by the visual data acquisition device on one side of the inspection vehicle after correction is recorded as a type one image, and the image on the other side is recorded as a type two image; then the type one image and the type two image are segmented, that is, the images of the rail head, rail waist and rail bottom are segmented. Since the type one image and the type two image are separated on both sides of the rail, the image of the rail head in the type one image and the type two image is repeatedly collected, while the images of the rail waist and rail bottom are collected on one side separately. Therefore, when processing the type one image and the type two image, it is only necessary to process the image of the rail head area in the type one image and the type two image. The purpose of the processing is to collect images from different perspectives, which can reduce the blind spots, brightness or shadow effects that may exist in unilateral image collection; and the processing of the rail head area image is to match the first rail head image and the second rail head image through multiple feature point matching. For the area where the first rail head image and the second rail head image overlap, the corresponding area of ​​the first rail head image or the corresponding area of ​​the second rail head can be selected to replace it; for the non-overlapping area of ​​the first rail head image and the second rail head image, all pixel points in the non-overlapping area are weighted fused according to the distance from the boundary of the first rail head image and the boundary of the second rail head image. ,in, is the pixel value of a pixel point in the non-overlapping area, is the pixel value of a pixel in the non-overlapping area of ​​the first track header image, The pixel value of a pixel in the non-overlapping area of ​​the second track header image, is the weight coefficient. The overlapping area and the non-overlapping area after weighted fusion are spliced ​​together to form the rail head image set; the first rail waist image and the second rail waist image are combined as the rail waist image set; and the first rail bottom image and the second rail bottom image are combined as the rail bottom image set.

[0099] Among them, a damage detection model is constructed, and the loss data and damage data of the rails are used to train the damage detection model. Specifically, the loss data of the rail head, rail waist and rail bottom can not only reflect the loss of the rails, but also have differential impacts on the safety of the rails when they are damaged because they are key structures that bear the train load. That is, thresholds are set for the loss values ​​of the rail head, rail waist and rail bottom respectively, so as to judge whether the rail head loss value exceeds the second threshold, whether the rail waist loss value exceeds the third threshold and whether the rail bottom loss value exceeds the fourth threshold. Since there are two situations of exceeding and not exceeding for the rail head, rail waist and rail bottom, there are 8 situations after the judgment results are arranged and combined, but one of them is that the loss values ​​of the rail head, rail waist and rail bottom all exceed the corresponding threshold, indicating that the rail currently has no potential damage risk. Therefore, it is only necessary to output the predicted damage data of the rails in the other 7 situations, such as if only the loss value of the rail head exceeds the corresponding threshold. If the rail head exceeds a threshold, there is a risk of wear / crush and cracking. Cracks originate from the rail surface or interior and extend along the depth of the rail head. If only the waist loss value exceeds the corresponding threshold, there is a risk of corrosion / wear and cracking on the waist. Cracks often originate from internal waist defects or external load fatigue. If only the foot loss value exceeds the corresponding threshold, there is a risk of wear / corrosion and cracking on the foot, and so on. In other words, under these different circumstances, rails potentially face different types of risks. Therefore, this embodiment outputs predicted damage data for rails under different circumstances. This damage data includes damage type and location. Then, damage annotation is performed on the rail head, waist, and foot image sets to obtain an annotated dataset. The annotated dataset is used to output actual rail damage data. This damage data includes damage type, location, and size. Determine whether the predicted damage data of the rail is consistent with the actual damage data (i.e., the type of damage data and the location of occurrence). If so, it means that the potential risk of the rail predicted by the loss value has occurred, and as the loss increases, the damage risk will further increase. At this time, if the actual damage data of the rail is directly output (such as a slight crack in the rail head, the risk level is relatively light), there will be obvious deviation. Therefore, this embodiment adopts the method of increasing the risk level of the rail itself and adding a risk coefficient to the actual damage data, and outputting the damage data of the rail, using predictive thinking to improve the accuracy of rail damage detection; if not, it means that the potential risk of the rail predicted by the loss value has not occurred, and the actual loss data of the rail is directly output.

[0100] Among them, based on the trained damage detection model, damage detection is performed on the rail to be tested, specifically: loss data and damage data of the rail to be tested are obtained, and input into the trained damage detection model, and the damage detection result of the rail to be tested is output.

[0101] In some embodiments, obtaining the average curvature change value of the rail includes:

[0102] Divide the rail into N sections at equal distances, and select three characteristic points for each of the N sections;

[0103] For each of the N rail segments, a fitting circle is constructed using three characteristic points, and the radius of the corresponding fitting circle is calculated, and the inverse of the radius of the fitting circle is used as the curvature of the corresponding rail segment. The three characteristic points are the starting point, midpoint, and end point of each rail segment.

[0104] Calculating the absolute value of the curvature difference between each adjacent rail section, and storing the absolute values ​​of all curvature differences in a first preset data set;

[0105] The average value of the absolute values ​​of all curvature differences in the preset data set is calculated to obtain the average curvature change value of the rail.

[0106] In some embodiments, the preset method includes:

[0107] Obtaining the starting point and the end point of the rail, and obtaining a plurality of data points at equal distances on the rail;

[0108] Calculate the tangent point slopes of all data points, and process the tangent point slopes of all data points using the first method;

[0109] The first method is specifically: starting from the starting point, calculating the absolute value of the slope difference between the adjacent data points corresponding to the tangent points, and storing the calculated absolute value of the slope difference between the adjacent data points corresponding to the tangent points in a second preset data set;

[0110] Calculating an average of the absolute values ​​of the slope differences of the corresponding tangent points of all adjacent data points stored in the second preset data set until the corresponding average value exceeds a first threshold, obtaining a data point in the second preset data set that is farthest from the starting point as a first data point, and recording adjacent data points of the first data point close to the starting point as second data points;

[0111] The second data point is used as a new starting point, and the remaining data points in the rail are traversed using the first method until the end point of the rail is reached, thereby completing the segmentation of the rail. After segmentation, the average value of the difference in slopes of adjacent tangent points of each segment of the rail is less than the first threshold.

[0112] In some embodiments, after processing the three-dimensional model, adjusting the angles of all physical data acquisition devices and all visual data acquisition devices includes:

[0113] Based on the processed three-dimensional model, the starting point of each rail segment after segmentation is obtained and recorded as the first starting point, and the end point is recorded as the first end point;

[0114] Constructing an axis of each rail segment with the first starting point and the first end point, which is recorded as a first axis;

[0115] Obtaining the axis of the physical data acquisition device, recorded as the second axis;

[0116] Adjust the second axis to be parallel to the first axis to complete the adjustment of the angles of all physical data acquisition devices;

[0117] Obtaining the axis of the visual data acquisition device, recorded as the third axis;

[0118] The angle between the third axis and the first axis is adjusted to always remain at a preset angle, thereby completing the adjustment of the angles of all visual data acquisition devices.

[0119] In some embodiments, after processing the three-dimensional model, adjusting the angles of all physical data acquisition devices and all visual data acquisition devices includes:

[0120] Based on the processed three-dimensional model, the starting point of each rail segment after segmentation is obtained and recorded as the first starting point, and the end point is recorded as the first end point;

[0121] Constructing an axis of each rail segment with the first starting point and the first end point, which is recorded as a first axis;

[0122] Obtaining the axis of the physical data acquisition device, recorded as the second axis;

[0123] Adjust the second axis to be parallel to the first axis to complete the adjustment of the angles of all physical data acquisition devices;

[0124] Obtaining the axis of the visual data acquisition device, recorded as the third axis;

[0125] The angle between the third axis and the first axis is adjusted to always remain at a preset angle, thereby completing the adjustment of the angles of all visual data acquisition devices.

[0126] In some embodiments, correcting the corresponding returned data and image data includes:

[0127] Based on the processed 3D model, the starting point of each rail segment is obtained;

[0128] Constructing a data collection area, with the starting point of each rail section as the length of the data collection area and the width of the data collection area as a second set value;

[0129] Based on the data acquisition area, the return data and image data of each segmented rail are corrected, and the corrected return data and image data are located in the data acquisition area corresponding to each segmented rail.

[0130] In some embodiments, outputting the rail loss data based on the corrected return data includes:

[0131] Obtain the standard profiles of the rail head, rail waist and rail bottom of each rail segment after segmentation;

[0132] The corrected returned data is spliced ​​to obtain the actual profiles of the rail head, rail waist and rail bottom of each section of rail;

[0133] Based on the standard and actual profiles of the rail head of each section, the loss value of the rail head is output; based on the standard and actual profiles of the rail waist of each section, the loss value of the rail waist is output; based on the standard and actual profiles of the rail foot of each section, the loss value of the rail foot is output.

[0134] In some embodiments, outputting the rail damage data based on the corrected image data includes:

[0135] The image processed by the visual data acquisition device on one side of the inspection vehicle after correction is recorded as a first type of image, and the image processed by the visual data acquisition device on the other side after correction is recorded as a second type of image;

[0136] Segmenting the image of the type to obtain a rail head image, which is recorded as a first rail head image, a side rail waist image, which is recorded as a first rail waist image, and a side rail bottom image, which is recorded as a first rail bottom image;

[0137] Segmenting the two types of images to obtain a rail head image recorded as a second rail head image, a side rail waist image recorded as a second rail waist image, and a side rail bottom image recorded as a second rail bottom image;

[0138] selecting a plurality of identical feature points in both the first rail head image and the second rail head image, and matching the first rail head image with the second rail head image based on feature point matching to obtain a matching result;

[0139] Based on the matching result, determine whether the first rail head image and the second rail head image completely overlap. If so, use the first rail head image or the second rail head image as the rail head image set of the rail. If not, obtain the non-overlapping area between the first rail head image and the second rail head image, and set a weight coefficient for all pixels in the non-overlapping area according to the distance from the boundary of the first rail head image and the boundary of the second rail head image to perform weighted fusion. Splice the overlapping area between the first rail head image and the second rail head image and the non-overlapping area after weighted fusion to form the rail head image set of the rail.

[0140] The first rail waist image and the second rail waist image are used as a rail waist image set of the rail; the first rail bottom image and the second rail bottom image are used as a rail bottom image set of the rail.

[0141] In some embodiments, training the damage detection model using the rail loss data and damage data includes:

[0142] Obtain the rail head loss value, rail waist loss value and rail bottom loss value of each rail segment after segmentation;

[0143] A second threshold is set for the loss value of the rail head, a third threshold is set for the loss value of the rail waist, and a fourth threshold is set for the loss value of the rail foot;

[0144] Determine whether the loss value of the rail head exceeds the second threshold value, and obtain a first determination result; determine whether the loss value of the rail waist exceeds the third threshold value, and obtain a second determination result; determine whether the loss value of the rail base exceeds the fourth threshold value, and obtain a third determination result;

[0145] Based on the first judgment result, the second judgment result, and the third judgment result, permutations and combinations are performed, and predicted damage data of the rail is output in seven situations;

[0146] The rail head image set, rail waist image set and rail bottom image set of the rail are all marked for damage, and the marked rail head image set, the marked rail waist image set and the marked rail bottom image set are obtained respectively;

[0147] Based on the labeled rail head image set, output the damage data of the rail head; based on the labeled rail waist image set, output the damage data of the rail waist; based on the labeled rail bottom image set, output the damage data of the rail bottom;

[0148] Integrating the damage data of the rail head, the damage data of the rail waist, and the damage data of the rail bottom into actual damage data of the rail;

[0149] Determine whether the predicted damage data of the rail is consistent with the actual damage data of the rail. If so, add a risk coefficient to the actual damage data of the rail, the risk coefficient is greater than 1, and the risk level of the rail is increased. If not, directly output the actual damage data of the rail.

[0150] In some embodiments, performing damage detection on the rail to be tested based on the trained damage detection model includes:

[0151] The loss data and damage data of the rail to be tested are obtained and input into the trained damage detection model, and the damage detection result of the rail to be tested is output.

[0152] Example 2

[0153] Based on the rail loss measurement and rail surface damage detection method provided in the first embodiment of the present application, the second embodiment of the present application also provides a rail loss measurement and rail surface damage detection system, such as Figure 2 Shown, including:

[0154] A three-dimensional model unit, used to obtain BIM data of the rail and establish a three-dimensional model of the rail;

[0155] a model processing unit, configured to process the three-dimensional model, specifically: obtaining an average curvature change value of the rail and determining whether it exceeds a first set value; if so, performing segmented processing on the three-dimensional model of the rail using a preset method; if not, performing no action;

[0156] A physical data acquisition unit is configured to be provided with M groups of physical data acquisition devices on the inspection vehicle, each group of physical data acquisition devices being configured to output structurally symmetrical laser light for the rail and to collect return data, M being ≥ 3;

[0157] A visual data acquisition unit is configured to provide at least one visual data acquisition device on each side of the rail on the inspection vehicle, wherein the visual data acquisition device is configured to acquire image data of the rail;

[0158] A correction unit is used to adjust the angles of all physical data acquisition devices and all visual data acquisition devices based on the processed three-dimensional model, and to correct the corresponding return data and image data;

[0159] A data processing unit is configured to output the loss data of the rail based on the corrected return data; and output the damage data of the rail based on the corrected image data;

[0160] The damage detection unit is used to construct a damage detection model, and use the loss data and damage data of the rail to train the damage detection model, and perform damage detection on the rail to be tested based on the trained damage detection model.

[0161] The specific principles and execution processes of each unit in the rail loss measurement and rail surface damage detection system disclosed in the above-mentioned embodiment 2 of the present application are the same as those of the rail loss measurement and rail surface damage detection method disclosed in the above-mentioned embodiment 1 of the present application. Please refer to the corresponding parts of the rail loss measurement and rail surface damage detection method disclosed in the above-mentioned embodiment 1 of the present application, and no further details will be given here.

[0162] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computing software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0163] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0164] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for measuring rail loss and detecting rail surface damage, characterized in that: include: Acquire BIM data of the rail and establish a three-dimensional model of the rail; Processing the three-dimensional model, specifically: obtaining the average curvature change value of the rail and determining whether it exceeds a first set value, if so, performing segmented processing on the three-dimensional model of the rail using a preset method; if not, taking no action; M groups of physical data acquisition devices are set on the inspection vehicle, each group of physical data acquisition devices is used to output structural symmetrical laser for the rail and collect return data, M ≥ 3; At least one visual data acquisition device is provided on each side of the rail on the inspection vehicle, and the visual data acquisition device is used to acquire image data of the rail; Based on the processed 3D model, adjust the angles of all physical data acquisition devices and all visual data acquisition devices, and correct the corresponding return data and image data; outputting the rail loss data based on the corrected return data; outputting damage data of the rail based on the corrected image data; Constructing a damage detection model, and using the loss data and damage data of the rail to train the damage detection model, and performing damage detection on the rail to be tested based on the trained damage detection model; The preset method includes: Obtaining the starting point and the end point of the rail, and obtaining a plurality of data points at equal distances on the rail; Calculate the tangent point slopes of all data points, and process the tangent point slopes of all data points using the first method; The first method is specifically: starting from the starting point, calculating the absolute value of the slope difference between the adjacent data points corresponding to the tangent points, and storing the calculated absolute value of the slope difference between the adjacent data points corresponding to the tangent points in a second preset data set; Calculating an average of the absolute values ​​of the slope differences of the corresponding tangent points of all adjacent data points stored in the second preset data set until the corresponding average value exceeds a first threshold, obtaining a data point in the second preset data set that is farthest from the starting point as a first data point, and recording adjacent data points of the first data point close to the starting point as second data points; Using the second data point as a new starting point, traversing the remaining data points in the rail using the first method until reaching the end point of the rail, thereby completing segmentation of the rail, wherein the average value of the difference in slopes of adjacent tangent points of each rail segment after segmentation is less than the first threshold; Outputting the rail loss data based on the corrected return data includes: Obtain the standard profiles of the rail head, rail waist and rail bottom of each rail segment after segmentation; The corrected returned data is spliced ​​to obtain the actual profiles of the rail head, rail waist and rail bottom of each section of rail; Based on the standard and actual profiles of the rail head of each section, the loss value of the rail head is output; based on the standard and actual profiles of the rail waist of each section, the loss value of the rail waist is output; based on the standard and actual profiles of the rail foot of each section, the loss value of the rail foot is output; Outputting the rail damage data based on the corrected image data includes: The image processed by the visual data acquisition device on one side of the inspection vehicle after correction is recorded as a first type of image, and the image processed by the visual data acquisition device on the other side after correction is recorded as a second type of image; Segmenting the image of the type to obtain a rail head image, which is recorded as a first rail head image, a side rail waist image, which is recorded as a first rail waist image, and a side rail bottom image, which is recorded as a first rail bottom image; Segmenting the two types of images to obtain a rail head image recorded as a second rail head image, a side rail waist image recorded as a second rail waist image, and a side rail bottom image recorded as a second rail bottom image; selecting a plurality of identical feature points in both the first rail head image and the second rail head image, and matching the first rail head image with the second rail head image based on feature point matching to obtain a matching result; Based on the matching result, determine whether the first rail head image and the second rail head image completely overlap. If so, use the first rail head image or the second rail head image as the rail head image set of the rail. If not, obtain the non-overlapping area between the first rail head image and the second rail head image, and set a weight coefficient for all pixels in the non-overlapping area according to the distance from the boundary of the first rail head image and the boundary of the second rail head image to perform weighted fusion. Splice the overlapping area between the first rail head image and the second rail head image and the non-overlapping area after weighted fusion to form the rail head image set of the rail. The first rail waist image and the second rail waist image are used as a rail waist image set of the rail; the first rail bottom image and the second rail bottom image are used as a rail bottom image set of the rail; Training the damage detection model using the rail loss data and damage data includes: Obtain the rail head loss value, rail waist loss value and rail bottom loss value of each rail segment after segmentation; A second threshold is set for the loss value of the rail head, a third threshold is set for the loss value of the rail waist, and a fourth threshold is set for the loss value of the rail foot; Determine whether the loss value of the rail head exceeds the second threshold value, and obtain a first determination result; determine whether the loss value of the rail waist exceeds the third threshold value, and obtain a second determination result; determine whether the loss value of the rail base exceeds the fourth threshold value, and obtain a third determination result; Based on the first judgment result, the second judgment result, and the third judgment result, permutations and combinations are performed, and predicted damage data of the rail is output in seven situations; The rail head image set, rail waist image set and rail bottom image set of the rail are all marked for damage, and the marked rail head image set, the marked rail waist image set and the marked rail bottom image set are obtained respectively; Based on the labeled rail head image set, output the damage data of the rail head; based on the labeled rail waist image set, output the damage data of the rail waist; based on the labeled rail bottom image set, output the damage data of the rail bottom; Integrating the damage data of the rail head, the damage data of the rail waist, and the damage data of the rail bottom into actual damage data of the rail; Determine whether the predicted damage data of the rail is consistent with the actual damage data of the rail. If so, add a risk coefficient to the actual damage data of the rail, the risk coefficient is greater than 1, and the risk level of the rail is increased. If not, directly output the actual damage data of the rail.

2. A rail loss measurement and rail surface damage detection method according to claim 1, characterized in that: Obtaining the average curvature change value of the rail includes: Divide the rail into N sections at equal distances, and select three characteristic points for each of the N sections; For each of the N rail segments, a fitting circle is constructed using three characteristic points, and the radius of the corresponding fitting circle is calculated, and the inverse of the radius of the fitting circle is used as the curvature of the corresponding rail segment. The three characteristic points are the starting point, midpoint, and end point of each rail segment. Calculating the absolute value of the curvature difference between each adjacent rail section, and storing the absolute values ​​of all curvature differences in a first preset data set; The average value of the absolute values ​​of all curvature differences in the preset data set is calculated to obtain the average curvature change value of the rail.

3. The method for measuring rail loss and detecting rail surface damage according to claim 1, characterized in that: After processing the 3D model, all physical data acquisition devices and all visual data acquisition devices are adjusted in angle, including: Based on the processed three-dimensional model, the starting point of each rail segment after segmentation is obtained and recorded as the first starting point, and the end point is recorded as the first end point; Constructing an axis of each rail segment with the first starting point and the first end point, which is recorded as a first axis; Obtaining the axis of the physical data acquisition device, recorded as the second axis; Adjust the second axis to be parallel to the first axis to complete the adjustment of the angles of all physical data acquisition devices; Obtaining the axis of the visual data acquisition device, recorded as the third axis; The angle between the third axis and the first axis is adjusted to always remain at a preset angle, thereby completing the adjustment of the angles of all visual data acquisition devices.

4. A rail loss measurement and rail surface damage detection method according to claim 1, characterized in that: Correction processing of corresponding returned data and image data includes: Based on the processed 3D model, the starting point of each rail segment is obtained; Constructing a data collection area, with the starting point of each rail section as the length of the data collection area and the width of the data collection area as a second set value; Based on the data acquisition area, the return data and image data of each segmented rail are corrected, and the corrected return data and image data are located in the data acquisition area corresponding to each segmented rail.

5. The method for measuring rail loss and detecting rail surface damage according to claim 1, characterized in that: Based on the trained damage detection model, the damage detection of the rail to be tested includes: The loss data and damage data of the rail to be tested are obtained and input into the trained damage detection model, and the damage detection result of the rail to be tested is output.

6. A rail loss measurement and rail surface damage detection system, implemented by the rail loss measurement and rail surface damage detection method according to claim 1, characterized in that: include: A three-dimensional model unit, used to obtain BIM data of the rail and establish a three-dimensional model of the rail; a model processing unit, configured to process the three-dimensional model, specifically: obtaining an average curvature change value of the rail and determining whether it exceeds a first set value; if so, performing segmented processing on the three-dimensional model of the rail using a preset method; if not, performing no action; A physical data acquisition unit is configured to be provided with M groups of physical data acquisition devices on the inspection vehicle, each group of physical data acquisition devices being configured to output structurally symmetrical laser light for the rail and to collect return data, M being ≥ 3; A visual data acquisition unit is configured to provide at least one visual data acquisition device on each side of the rail on the inspection vehicle, wherein the visual data acquisition device is configured to acquire image data of the rail; A correction unit is used to adjust the angles of all physical data acquisition devices and all visual data acquisition devices based on the processed three-dimensional model, and to correct the corresponding return data and image data; A data processing unit, configured to output the rail loss data based on the corrected return data; outputting damage data of the rail based on the corrected image data; The damage detection unit is used to build a damage detection model, and use the loss data and damage data of the rail to train the damage detection model, and perform damage detection on the rail to be tested based on the trained damage detection model.

Images