Detection method for jamming fault of low-speed rail wheel

Abstract

The invention discloses a method for detecting a jamming fault of a low-speed rail wheel, and belongs to the technical field of rail vehicle fault detection, and the method comprises the following steps: S1, obtaining a video image of the low-speed rail wheel; s2, marking the video image to obtain a wheel part detection data set, and training a target detection model by using the wheel part detection data set; s3, performing target detection on the real-time running video of the wheel by using the target detection model to obtain center coordinates of a nut and a bearing, and performing nearest neighbor matching to obtain a nut movement track; and S4, calculating the motion trail of the nut when the wheel normally runs, comparing the actual motion trail of the nut, calculating a residual value, analyzing the residual value, and constructing a lagging judgment algorithm to carry out fault judgment on wheel lagging. The low-speed rail wheel bearing jamming fault is detected in a non-contact dynamic detection mode, the detection process is simplified, and the detection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of rail vehicle fault detection technology, and in particular to a method for detecting a jamming fault of a low-speed rail wheel. Background Art

[0002] The rapid development of special low-speed rail vehicles (such as sintering machine trolleys) in industrial and other scenarios has significantly improved people's production and lifestyles, while also placing higher demands on the safety of vehicle operation. Wheel bearings are key components that support rotating shafts and are used to reduce friction and wear. Their normal operation is the key to ensuring the smooth operation of vehicles.

[0003] However, in industrial scenarios, low-speed rail vehicles are subject to long-term high temperature, high dust and heavy load operation, which may lead to dust blockage, bearing bending and fracture. When these phenomena occur, bearing failure may occur, resulting in wheel jam. Wheel jam will cause friction between the wheel and the track, which may cause inevitable damage to the track and have a serious impact on the operating safety and quality of rail vehicles. In addition, if the wheel with jam failure is not maintained in time, the wheel may fall off, leading to production accidents.

[0004] Especially in industrial vehicles such as sintering machines, wheel failures account for a large proportion due to the large number of wheels and harsh working environment. If timely maintenance is not carried out, the wheel will wear faster and reduce the service life of the wheel, affecting the operating safety and efficiency of industrial vehicles such as sintering machine trolleys. Therefore, real-time wheel jamming fault detection is required.

[0005] At present, the bearing fault detection of sintering machine trolleys mainly relies on manual inspection. However, the number of wheels of trolleys is always large and the environment they are in is harsh. Relying on manual observation to detect the fault of trolley wheel bearings has the disadvantages of high labor intensity, easy to miss or misdetect, and potential safety hazards.

[0006] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgment or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art. Summary of the invention

[0007] The object of the present invention is to provide a method for detecting a jamming fault of a low-speed rail wheel, which can detect a jamming fault of a low-speed rail wheel bearing by non-contact dynamic detection, simplify the detection process and improve the detection efficiency.

[0008] To achieve the above object, the present invention provides a method for detecting a jamming fault of a low-speed rail wheel, which is applied to a fault detection platform for jamming of a low-speed rail wheel. The method comprises the following steps:

[0009] S1: Acquire video images of low-speed wheels on the track;

[0010] S2: annotating the video image obtained in step S1 to obtain a wheel sub-component detection dataset, and using the wheel sub-component detection dataset to train a target detection model;

[0011] S3: Use the target detection model trained in step S2 to perform target detection on the real-time running video of the wheel to obtain the center coordinates of the nut and the bearing, and perform nearest neighbor matching to obtain the motion trajectory of the nut;

[0012] S4: Calculate the motion trajectory of the nut when the wheel is operating normally, and compare the actual motion trajectory of the nut to calculate the residual value, use the dynamic threshold method to analyze the residual value, and build a jamming judgment algorithm to judge the wheel jamming fault;

[0013] S5: Obtain the failure time of the bearing, locate the faulty wheel, and send the side panel number of the trolley when the jam occurs to the engineer for fault detection and maintenance.

[0014] In one embodiment of the present invention, in step S1, obtaining the video image of the low-speed rail wheel includes: photographing the trolley through a camera on a fault detection platform for the stuck low-speed rail wheel, determining whether the wheel enters the detection area, and when the wheel enters the monitoring area, capturing a valid video clip to obtain a clear image of the front of the wheel.

[0015] In one embodiment of the present invention, in step S3, the trained parameters are used to detect and classify different parts of the wheel, the coordinates of the nut identification target frame and the bearing identification target frame are obtained, the upper right corner coordinates and the lower left corner coordinates of the nut identification target frame are used to calculate the center coordinates of the nut, and the upper right corner coordinates and the lower left corner coordinates of the bearing identification target frame are used to calculate the center coordinates of the bearing.

[0016] In one embodiment of the present invention, in step S3, the process of performing nearest neighbor matching to obtain the nut motion trajectory is as follows: perform nearest neighbor matching on the obtained nut center coordinates, calculate the Euclidean distance between the center coordinate point of each nut in the previous frame image and the center coordinate point in the next frame image, and find the two nut center coordinates with the smallest Euclidean distance; then, perform nearest neighbor matching on the center coordinate sequence of all nuts in the video to finally obtain the nut motion trajectory; and obtain the actual motion trajectory at the nut center coordinate of the n+mth frame.

[0017] In one embodiment of the present invention, in step S4, the coordinate data of the nut in the n+mth frame is calculated based on the real coordinate data of the nut in the nth frame, the actual motion trajectory of the nut is compared with the calculated normal motion trajectory of the nut, and the residual between the trajectory coordinates is calculated.

[0018] In one embodiment of the present invention, the steps for calculating the coordinate data of the nut of the n+mth frame are as follows:

[0019] S401: Obtain the coordinate position of the wheel bearing in the nth frame, establish a rectangular coordinate system with the coordinate position of the bearing in the nth frame as the origin, and calculate the bolt using the inverse tangent function. i Initial angle α to the positive half of the x-axis in the bearing coordinate system i ; where i is the serial number of the nut; initial angle α i The calculation process is as follows:

[0020]

[0021] in, is the coordinate of the i-th wheel nut in the n-th frame, is the coordinate of the wheel bearing in the nth frame;

[0022] S402: When a wheel of a fixed size moves forward at a relatively small angle at a uniform speed, the rotation angle β of the wheel is calculated using an arc length formula; the calculation process of the rotation angle β is as follows:

[0023]

[0024] Where R is the radius of the wheel; is the abscissa of the axis center of the target bearing in the photo coordinate system of the n+mth frame; is the abscissa of the axis center of the target bearing in the nth frame in the photo coordinate system;

[0025] S403: Establish a rectangular coordinate system with the bearing coordinate position of the n+mth frame as the origin, passing through the initial angle α of the target nut i The angle of the target nut in the n+mth frame is calculated by the rotation angle β, and the coordinates of the target nut in the n+mth frame are calculated by using the target nut angle obtained in the n+mth frame and the coordinates of the bearing in the n+mth frame. The calculation process of the target nut coordinate position in the n+mth frame is as follows:

[0026]

[0027] in, is the theoretical coordinate of the i-th nut in the n+m-th frame, is the distance from the nut to the bearing corresponding to the i-th nut in the n+m-th frame.

[0028] In one embodiment of the present invention, in step S4, the method for calculating the residual value by comparing the actual motion trajectory of the nut is as follows:

[0029] S404: Starting from the nth frame, the actual center coordinates of the nut in the nth frame video are obtained. The coordinates of the nut center calculated at the n+mth frame And the actual motion trajectory at the nut center coordinates of the n+mth frame And calculate the residual value:

[0030]

[0031] in, is the central ordinate of the i-th nut in the n+m-th frame of the actual motion trajectory coordinates, is the central ordinate of the i-th nut in the n+m-th frame calculated based on the horizontal and vertical coordinates of the i-th nut in the n-th frame;

[0032] S405: Calculate the residual values ​​and d of all nuts for each frame of the detection video sum , calculate the residual value of the coordinates of the nth frame through the coordinates of the nth frame, and finally form The sequence of residual values ​​is as follows:

[0033]

[0034] Where h is the total number of frames in the video. Represents the residual value of the nth frame.

[0035] In one embodiment of the present invention, in step S4, before analyzing the residual value using the dynamic threshold method, it also includes: setting a threshold for the residual values ​​obtained by different wheels in the detection area, specifically including the following steps:

[0036] Calculate the average of the first n maximum values ​​of the residual value and use the average value as the threshold sum .

[0037] In one embodiment of the present invention, in step S4, constructing a jamming determination algorithm to perform fault determination on wheel jamming includes the following steps:

[0038] In getting the threshold sum Then, find the residual value sequence that exceeds the threshold The calculation is as follows:

[0039]

[0040] If there are consecutive k m or more residual value sequences Greater than threshold sum When the wheel is considered to have a stuck fault, a stuck sign is added. In the present invention, the judgment condition k is generally set m =3; otherwise, it is determined that the wheels in the video are running normally.

[0041] In one embodiment of the present invention, step S5 specifically includes: synchronously capturing images of the trolley side panels through a camera in a rail low-speed wheel jam fault detection platform, training a target detection model for the side panel numbers, and acquiring the number of the trolley side panel currently passing through; after completing the wheel jam fault detection, uploading the wheel jam detection results and the number of the trolley side panel at the current fault time to the engineer at the same time.

[0042] Compared with the prior art, a method for detecting a stuck fault of a low-speed rail wheel according to the present invention adopts a non-contact dynamic detection method, simplifies the detection process, and improves the level of intelligent detection. By real-time analysis and processing of the collected wheel images, it is determined whether the wheel has a stuck fault, thereby realizing real-time detection of wheel stuck faults and early warning of wheel falling off caused by sintering machine wheel bearing failure, thereby ensuring the safety and stability of the operation of special low-speed rail vehicles (such as sintering machine trolleys) in industrial and other scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a workflow diagram of a method for detecting a wheel jam fault on a low-speed track according to an embodiment of the present invention;

[0044] Figure 2 It is a schematic diagram of wheel running video acquisition of a method for detecting wheel jamming fault on a low-speed track according to an embodiment of the present invention;

[0045] Figure 3 A wheel entering a detection area judgment diagram of a method for detecting a stuck fault of a low-speed rail wheel according to an embodiment of the present invention;

[0046] Figure 4 is a schematic diagram of nearest neighbor matching of a method for detecting a jamming fault of a low-speed rail wheel according to an embodiment of the present invention;

[0047] Figure 5 is a schematic diagram of calculating nut trajectory of a method for detecting a jamming fault of a low-speed rail wheel according to an embodiment of the present invention;

[0048] Figure 6 It is a diagram of obtaining wheel position information of a method for detecting a stuck fault of a low-speed rail wheel according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings, but it should be understood that the protection scope of the present invention is not limited by the specific implementation modes.

[0050] Unless explicitly stated otherwise, throughout the specification and claims, the term “comprise” or variations such as “include” or “comprising”, etc., will be understood to include the stated elements or components but not to exclude other elements or components.

[0051] like Figures 1 to 6 As shown, a method for detecting a jamming fault of a low-speed rail wheel according to a preferred embodiment of the present invention is mainly used on a fault detection platform for jamming of a low-speed rail wheel. An industrial camera is arranged on both sides of the trolley on the fault detection platform. The industrial cameras are respectively kept in the same height with the wheel bearings (i.e., the axle center of the wheel), and the wheel running video data on both sides is recorded. It is determined whether the wheel enters the detection area, and then valid video clips are captured. Taking the jamming fault detection of the sintering machine trolley wheel as an example, an industrial camera is arranged on both sides of the sintering machine trolley, and the industrial cameras are respectively kept in the same height with the sintering machine trolley wheel bearings.

[0052] The fault detection platform constructed includes two brackets and four cameras, with one bracket and two cameras arranged on both sides of the trolley respectively; on one side of the trolley, one camera is fixed on the upper part of the bracket, and the other camera is fixed on the lower part of the bracket. The upper camera is maintained at the same height as the center of the trolley side panel to record the side panel video, and the lower camera is maintained at the same height as the center of the trolley wheel to record the wheel operation video.

[0053] Specifically, the fault detection platform for low-speed track wheel jams includes a fixed bracket 7, a bracket base 8, industrial cameras 2 and 3, a video image processor 11, a fill light 10 and a display storage 12 respectively arranged on both sides of the trolley, wherein the industrial cameras 2 and 3 are connected to the video image processor 11 through a cable 9, and the cable 9 includes a power line and a data line. The wheel 1 runs on the rail 5, and a sleeper 6 is arranged under the rail 5.

[0054] like Figure 2As shown, taking one side of the trolley as an example, the fixed bracket 7 and the bracket base 8 are arranged beside the track, the fixed bracket 7 is installed on the bracket base 8, and the bracket base 8 is fixed on the ground. Industrial cameras 2 and 3 are fixed on the fixed bracket 7, the industrial camera 2 on the lower side is used to collect real-time running video data of the wheel 1, and the industrial camera 3 on the upper side is used to collect real-time video data of the side panel 4; the fill light 10 is installed on the fixed bracket 7, and is used to fill light when the industrial cameras 2 and 3 collect images. The video image processor 11 is connected to the industrial cameras 2 and 3 on the fixed bracket 7 through the cable 9, and is used to process the acquired images; the display memory 12 is connected to the video image processor 11 through the cable 9, and is used to display and save the processed image data information. The cable 9 includes a power line and a data line, which can be used for data transmission and power supply. Figure 2 , the arrangement of one side of the wheel 1 is shown, and the wheel and the side plate on the other side connected to the wheel 1 via the axle 13 are arranged in the same manner as described above.

[0055] In this embodiment, a fixed bracket 7 is provided outside the rail 5 in the direction of travel of the wheel 1, and industrial cameras 2 and 3 and a fill light 10 are sequentially installed on the fixed bracket 7; and the height of the industrial camera 2 is consistent with the height of the wheel 1 bearing, and the height of the industrial camera 3 is consistent with the center height of the side plate 4, and the fill light 10 is located above the horizontal plane of the height of the industrial cameras 2 and 3, and the industrial camera 3 is located above the horizontal plane of the industrial camera 2. In this embodiment, the industrial camera 2 is used to collect the video data of the real-time operation of the wheel and detect whether the wheel travels to the measurement area, and the effective video image of the wheel in the measurement area is intercepted in the video; the industrial camera 3 is used to collect the real-time video data of the side plate and determine the position of the wheel.

[0056] like Figure 1 As shown, a method for detecting a jamming fault of a low-speed rail wheel of the present invention specifically comprises the following steps:

[0057] S1: Acquire video images of low-speed wheels on the track.

[0058] In this step S1, obtaining the video image of the low-speed track wheel specifically includes: photographing the trolley through the camera on the fault detection platform of the low-speed track wheel jam, judging whether the wheel enters the detection area, and when the wheel enters the monitoring area, intercepting the effective video clip to obtain a clear image of the front of the wheel.

[0059] The detection area is generally an area covered by 1 / 4 to 3 / 4 of the length and height of the video image (the range of the detection area is set according to actual conditions).

[0060] Specifically, the x-axis coordinate of the center position of the bearing detection bounding box is known to be x axle , the y-axis coordinate of the center position of the bounding box is yaxle , the pixel size of the video x-axis is x video-l , the pixel size of the video y axis is y video-l , the range of the wheel nut running trajectory detection area in the video has the following relationship:

[0061]

[0062] According to the above formula, when the wheel bearing center coordinates are between the 1 / 4 position and the 3 / 4 position of the x-axis pixel size and the 1 / 4 position and the 3 / 4 position of the y-axis pixel size of the video, the wheel running video starts to be recorded.

[0063] S2: Annotate the video image obtained in step S1 to obtain a wheel sub-component detection dataset, and use the wheel sub-component detection dataset to train a target detection model.

[0064] The video image acquired in step S1 is annotated manually, and the annotated content includes the identification frames of the nut and the bearing. In addition, the wheel hub, the outer contour, etc. may also be annotated.

[0065] In this step S2, the nuts and bearings of the wheel are marked in the captured video image, and a wheel sub-component detection dataset is created; then, the object detection model is trained using the created wheel sub-component detection dataset. The object detection model can be a YOLOv9 model, a YOLOv7 model, or another YOLO object detection model. The object detection model that can be used is not fixed, and other object detection models can also be used as long as they are accurate enough.

[0066] S3: Use the trained target detection model to perform target detection on the real-time wheel running video to obtain the center coordinates of the nut and bearing, and perform nearest neighbor matching to obtain the nut motion trajectory.

[0067] After completing the target detection model training, use the trained target detection model to detect and classify different parts of the wheel, obtain the coordinates of the wheel sub-component recognition target frame (nut recognition target frame and bearing recognition target frame), and generate the coordinates of the two diagonal vertices of the rectangular bounding box of the detection target object. Use the upper right corner coordinates and lower left corner coordinates of the nut to calculate the center coordinates of the nut, and use the upper right corner coordinates and lower left corner coordinates of the bearing to calculate the center coordinates of the bearing.

[0068] The coordinates of the wheel sub-component recognition target frame obtained by the target detection model are: the coordinates of the upper right corner and the lower left corner of the nut are The coordinates of the upper right corner and lower left corner of the bearing are (x axle-r ,y axle-r )、(x axle-l ,yaxle-l ); among them, bolt i -r represents the coordinates of the upper right corner of the 6-nut target box, bolt i -l represents the coordinates of the lower left corner of the 6 nut target boxes, where 1≤i≤6, axle-r represents the coordinates of the upper right corner of the bearing target box, and axle-l represents the coordinates of the lower left corner of the bearing target box.

[0069] Therefore, the expression of the center position of the nut target recognition frame is as follows:

[0070]

[0071] in, represents the center position of the nut detection bounding box, represents the position of the upper right corner of the nut detection bounding box, Indicates the position of the lower left corner of the nut detection bounding box.

[0072] The center position of the bearing target recognition frame is as follows:

[0073]

[0074] Among them, (x axle ,y axle ) represents the center position of the nut detection bounding box, (x axle-r ,y axle-r ) represents the position of the upper right corner of the bearing detection bounding box, (x axle-l ,y axle-l ) represents the position of the lower left corner of the bearing detection bounding box.

[0075] like Figure 4 As shown in FIG. 1 , the specific process of performing nearest neighbor matching to obtain the nut motion trajectory is as follows: performing nearest neighbor matching on the obtained nut center coordinates, calculating the Euclidean distance between the center coordinate point of each nut in the previous frame image and the center coordinate point in the next frame image, and finding the two nut center coordinates with the smallest Euclidean distance, and obtaining 6 sets of matching coordinates, namely, the center coordinates of the 6 nuts in the previous and next two frames of images; then, performing nearest neighbor matching on the center coordinate sequences of all six nuts in the video to finally obtain the nut motion trajectory; and obtaining the nut center coordinates of the actual motion trajectory in the n+mth frame

[0076]

[0077] Specifically, in the previous frame image, the center coordinate point of each nut is expressed as The center coordinates of each nut in the next frame are expressed as where p i ∈P,qi ∈Q. By calculating the Euclidean distance between the nut center coordinates in the previous frame image and the next frame image, and finding the two nut center coordinates with the smallest Euclidean distance, the nearest neighbor matching formula using the Euclidean distance is as follows:

[0078]

[0079] in, is the coordinate p to be matched in the previous frame i The nearest neighbor matching result of , P is the set of coordinate points in the next frame, and Q is the set of coordinate points in the next frame; and is the point p in the set P i The horizontal and vertical coordinates of and is the point q in the set Q i The horizontal and vertical coordinates of .

[0080] The 6 sets of matching coordinates that can be obtained through nearest neighbor matching are the center coordinates of the 6 nuts in the two frames before and after. The nearest neighbor matching is performed on the center coordinate sequence of all six nuts in the video. When the number of frames of the detection video is h frames, the center coordinate trajectory sequence of the six nuts in the video is finally obtained as follows:

[0081]

[0082] In the above formula, represents the center coordinate sequence of the nut numbered i, Indicates the horizontal coordinate of the center of the nut numbered i in the nth frame image, Indicates the horizontal coordinate of the center of the nut numbered i in the nth frame image.

[0083] S4: Calculate the motion trajectory of the nut when the wheel is operating normally, and compare the residual value with the actual motion trajectory of the nut. Use the dynamic threshold method to analyze the residual value, and build a jamming judgment algorithm to perform fault judgment on wheel jamming.

[0084] like Figure 5 As shown, the motion trajectory of the nut when the wheel is running normally under the current situation is calculated according to the actual motion trajectory of the nut, that is, the coordinate data of the nut in the n+mth frame is calculated according to the real coordinate data of the nut in the nth frame (n and m are both optional parameters), and the actual motion trajectory of the nut and the calculated normal motion trajectory of the nut are compared to calculate the residual between the trajectory coordinates.

[0085] In the above calculation process, the coordinate data of the nut in the n+mth frame is calculated as follows:

[0086] S401: Obtain the coordinate position of the wheel bearing in the nth frame, establish a rectangular coordinate system with the coordinate position of the bearing in the nth frame as the origin, and calculate the bolt using the inverse tangent function. i Initial angle α to the positive half of the x-axis in the bearing coordinate system i . Where i is the serial number of the nut.

[0087] In step S401, since the range of the inverse tangent function is It is necessary to determine the quadrant to which the nut coordinates of the nth frame belong in the rectangular coordinate system, so as to obtain the initial angle α of the nut to the positive half axis of the x-axis in the bearing coordinate system. i ; Initial angle α i The calculation process is as follows:

[0088]

[0089] in, is the coordinate of the i-th wheel nut in the n-th frame, is the coordinate of the wheel bearing in the nth frame;

[0090] S402: When a wheel of a fixed size moves forward at a constant speed at a relatively small angle, the rotation angle β of the wheel is calculated using an arc length formula.

[0091] Specifically, when a wheel of a fixed size moves forward at a relatively small angle β at a constant speed, the distance the wheel moves forward is approximately equal to the arc length corresponding to the wheel rotation angle, and the rotation angle β can be calculated by the following process:

[0092] (1) Calculate the distances from all nuts to bearings in the nth frame and the n+mth frame. The specific calculation formula is as follows:

[0093]

[0094] Among them, r i is the distance from the i-th wheel nut to the bearing; is the center coordinate of the i-th wheel nut; (x axle ,y axle ) are the coordinates of the wheel bearing.

[0095] (2) Calculate the average value of the distances from all nuts to the bearings, and calculate the radius R of the wheel by the ratio l of the actual wheel radius and the nut-bearing distance; the specific calculation formula is as follows:

[0096]

[0097] Where R is the radius of the wheel, is the distance from the i-th nut of the wheel to the bearing in the n-th frame; is the distance from the i-th nut to the bearing of the wheel in the n+m-th frame; l is the conversion coefficient between the nut-bearing distance and the wheel radius.

[0098] (3) When the rotation angle β is relatively small, the length of the wheel surface arc rotation is approximately equal to the distance the axle bearing moves forward. The length of the wheel surface arc rotation is calculated using the arc length formula, thereby calculating the wheel rotation angle β. The specific calculation formula is as follows:

[0099]

[0100] in, is the abscissa of the axis center of the target bearing in the photo coordinate system of the n+mth frame; It is the horizontal coordinate of the axis center of the target bearing in the nth frame in the photo coordinate system.

[0101] S403: Establish a rectangular coordinate system with the bearing coordinate position of the n+mth frame as the origin, passing through the initial angle α of the target nut i The angle of the target nut in the n+mth frame is calculated by the rotation angle β, and the coordinates of the target nut in the n+mth frame are calculated by using the target nut angle obtained in the n+mth frame and the coordinates of the bearing in the n+mth frame. The calculation process of the target nut coordinate position in the n+mth frame is as follows:

[0102]

[0103] Specifically, since the nut coordinate position calculated by trigonometric function and the actual nut coordinate position belong to different quadrants, it is necessary to determine the quadrant to which the nut coordinates of the n+mth frame belong in the rectangular coordinate system, so as to obtain the target nut coordinate position in the n+mth frame.

[0104]

[0105] in, is the theoretical coordinate of the i-th nut in the n+m-th frame, where Indicates the distance from the nut to the bearing corresponding to the i-th nut in the n+m-th frame.

[0106] In step S4, the residual between the trajectory coordinates is calculated by comparing the actual motion trajectory of the nut with the calculated normal motion trajectory of the nut.

[0107] Specifically, the method for calculating the residual value by comparing the actual motion trajectory of the nut is as follows:

[0108] S404: Starting from the nth frame, the actual center coordinates of the nut in the nth frame video are obtained. The coordinates of the nut center calculated at the n+mth frame And the actual motion trajectory at the nut center coordinates of the n+mth frame And calculate the residual value.

[0109] Considering that there is still horizontal displacement when the wheel is stuck, the horizontal coordinate distance difference is not obvious. The vertical coordinate of the nut center calculated at the n+mth frame The ordinate of the nut center at the n+mth frame of the actual motion trajectory The residual value is calculated as the judgment parameter of the jamming. The residual calculation process is as follows:

[0110]

[0111] in, is the central ordinate of the i-th nut in the n+m-th frame of the actual motion trajectory coordinates, It is the central ordinate of the i-th nut in the n+m-th frame calculated based on the horizontal and vertical coordinates of the i-th nut in the n-th frame.

[0112] S405: Calculate the residual values ​​and d of all nuts for each frame of the detection video sum , calculate the residual value of the coordinates of the nth frame through the coordinates of the nth frame, and finally form The sequence of residual values ​​is as follows:

[0113]

[0114] Where h is the total number of frames in the video. Represents the residual value of the nth frame.

[0115] Through the calculation of the nut motion trajectory described above, a residual value sequence between the actual motion trajectory of the nut detected in the video and the calculated normal motion trajectory is obtained.

[0116] In step S4, before analyzing the residual value using the dynamic threshold method, it is necessary to set a threshold for the residual values ​​obtained by different wheels in the detection area. Specifically, the thresholds for the first n residual values ​​are ′ The average value k of each maximum value is calculated and used as the threshold. When setting the dynamic threshold for the collected data, the parameter n ′ It can be adjusted according to the actual situation. In the present invention, when n ′ When =5, the calculated threshold is the most appropriate threshold.

[0117] The specific steps of using the dynamic threshold method to analyze the residual value are as follows:

[0118] S406: Find all the maximum values ​​of the residual value and use M sum represents the set of these maximum values, the set M sum As follows:

[0119]

[0120] S407: For the set of maximum values ​​M sum Sort in descending order to get an ordered maximum value set Sorted sum As follows:

[0121] Sorted sum = sort(M sum ).

[0122] S408: Take Sorted sum The values ​​in the top n ′ maximum values ​​and calculate their average threshold sum , which is the threshold value, and the calculation process is as follows:

[0123]

[0124] When setting the dynamic threshold value for the collected data, the present invention selects n ′ =5, calculated threshold sum is the most appropriate threshold.

[0125] In step S4, constructing a jamming determination algorithm to perform fault determination on wheel jamming includes the following steps:

[0126] In getting the threshold sum Then, find the residual value sequence that exceeds the threshold The calculation is as follows:

[0127]

[0128] If there are consecutive k m or more residual value sequences Greater than threshold sum When the wheel is considered to have a stuck fault, a stuck sign is added. In the present invention, the judgment condition k is generally set m =3; otherwise, it is determined that the wheels in the video are running normally.

[0129] S5: Obtain the failure time of the bearing, locate the faulty wheel, and send the side panel number of the trolley when the jam occurs to the engineer for fault detection and maintenance.

[0130] After completing the wheel jam fault judgment, in order to help engineers quickly find the faulty wheel, the present invention locates the faulty wheel. The faulty wheel is located by locating the side panel number corresponding to the wheel. The image of the trolley side panel is synchronously captured by the camera on the upper side of the rail low-speed wheel jam fault detection platform, and the target detection model of the side panel number is trained to complete the acquisition of the side panel number of the trolley currently passing. After completing the wheel jam fault detection, the wheel jam detection result and the number of the trolley side panel at the current fault time are uploaded to the engineer at the same time, and the engineer completes the maintenance of the faulty wheel in time.

[0131] The foregoing description of specific exemplary embodiments of the present invention is for the purpose of illustration and demonstration. These descriptions are not intended to limit the present invention to the precise form disclosed, and it is clear that many changes and variations can be made based on the above teachings. The purpose of selecting and describing the exemplary embodiments is to explain the specific principles of the present invention and its practical application, so that those skilled in the art can realize and utilize various different exemplary embodiments of the present invention and various different selections and changes. The scope of the present invention is intended to be limited by the claims and their equivalents.

Claims

1. A method for detecting a jamming fault of a low-speed rail wheel, applied to a fault detection platform for a low-speed rail wheel jamming, characterized in that: The method comprises the following steps: S1: Acquire video images of low-speed wheels on the track; S2: annotating the video image obtained in step S1 to obtain a wheel sub-component detection dataset, and using the wheel sub-component detection dataset to train a target detection model; S3: Use the target detection model trained in step S2 to perform target detection on the real-time running video of the wheel to obtain the center coordinates of the nut and the bearing, and perform nearest neighbor matching to obtain the motion trajectory of the nut; S4: Calculate the motion trajectory of the nut when the wheel is operating normally, and compare the actual motion trajectory of the nut to calculate the residual value, use the dynamic threshold method to analyze the residual value, and build a jamming judgment algorithm to judge the wheel jamming fault; S5: Obtain the failure time of the bearing, locate the faulty wheel, and send the side panel number of the trolley when the jam occurs to the engineer for fault detection and maintenance.

2. The method for detecting a jamming fault of a low-speed rail wheel according to claim 1, characterized in that: In step S1, the video image of the low-speed track wheel is obtained, which includes: photographing the trolley through a camera on a fault detection platform for the low-speed track wheel jam, determining whether the wheel enters the detection area, and when the wheel enters the monitoring area, capturing effective video clips to obtain a clear image of the front of the wheel.

3. The method for detecting a jamming fault of a low-speed rail wheel according to claim 1, characterized in that: In step S3, the trained parameters are used to detect and classify different parts of the wheel, the coordinates of the nut recognition target frame and the bearing recognition target frame are obtained, the upper right corner coordinates and the lower left corner coordinates of the nut recognition target frame are used to calculate the center coordinates of the nut, and the upper right corner coordinates and the lower left corner coordinates of the bearing recognition target frame are used to calculate the center coordinates of the bearing.

4. The method for detecting a jamming fault of a low-speed rail wheel according to claim 3, characterized in that: In step S3, the process of performing nearest neighbor matching to obtain the nut motion trajectory is as follows: perform nearest neighbor matching on the obtained nut center coordinates, calculate the Euclidean distance between the center coordinate point of each nut in the previous frame image and the center coordinate point in the next frame image, and find the two nut center coordinates with the smallest Euclidean distance; then, perform nearest neighbor matching on the center coordinate sequence of all nuts in the video to finally obtain the nut motion trajectory, and obtain the nut center coordinates of the actual motion trajectory in the n+mth frame 5. The method for detecting a jamming fault of a low-speed rail wheel according to claim 1, characterized in that: In step S4, the coordinate data of the nut in the n+mth frame is calculated based on the real coordinate data of the nut in the nth frame, the actual motion trajectory of the nut is compared with the calculated normal motion trajectory of the nut, and the residual between the trajectory coordinates is calculated.

6. The method for detecting a jamming fault of a low-speed rail wheel according to claim 5, characterized in that: The calculation steps of the coordinate data of the nut of the n+mth frame are as follows: S401: Obtain the coordinate position of the wheel bearing in the nth frame, establish a rectangular coordinate system with the coordinate position of the bearing in the nth frame as the origin, and calculate the bolt using the inverse tangent function. i Initial angle α to the positive half of the x-axis in the bearing coordinate system i ; where i is the serial number of the nut; initial angle α i The calculation process is as follows: in, is the coordinate of the i-th wheel nut in the n-th frame, is the coordinate of the wheel bearing in the nth frame; S402: When a wheel of a fixed size moves forward at a relatively small angle at a uniform speed, the rotation angle β of the wheel is calculated using an arc length formula; the calculation process of the rotation angle β is as follows: Where R is the radius of the wheel; is the abscissa of the axis center of the target bearing in the photo coordinate system of the n+mth frame; is the abscissa of the axis center of the target bearing in the nth frame in the photo coordinate system; S403: Establish a rectangular coordinate system with the bearing coordinate position of the n+mth frame as the origin, passing through the initial angle α of the target nut i The angle of the target nut in the n+mth frame is calculated by the rotation angle β, and the coordinates of the target nut in the n+mth frame are calculated by using the target nut angle obtained in the n+mth frame and the coordinates of the bearing in the n+mth frame. The calculation process of the target nut coordinate position in the n+mth frame is as follows: in, is the theoretical coordinate of the i-th nut in the n+m-th frame, is the distance from the nut to the bearing corresponding to the i-th nut in the n+m-th frame.

7. The method for detecting a jamming fault of a low-speed rail wheel according to claim 6, characterized in that: In step S4, the method for calculating the residual value by comparing the actual motion trajectory of the nut is as follows: S404: Starting from the nth frame, the actual center coordinates of the nut in the nth frame video are obtained. The coordinates of the nut center calculated at the n+mth frame And the actual motion trajectory at the nut center coordinates of the n+mth frame And calculate the residual value: in, is the central ordinate of the i-th nut in the n+m-th frame of the actual motion trajectory coordinates, is the central ordinate of the i-th nut in the n+m-th frame calculated based on the horizontal and vertical coordinates of the i-th nut in the n-th frame; S405: Calculate the residual values ​​and d of all nuts for each frame of the detection video sum , calculate the residual value of the coordinates of the nth frame through the coordinates of the nth frame, and finally form The sequence of residual values ​​is as follows: Where h is the total number of frames in the video. Represents the residual value of the nth frame.

8. The method for detecting a jamming fault of a low-speed rail wheel according to claim 7, characterized in that: In step S4, before analyzing the residual value by the dynamic threshold method, the method further includes: setting a threshold for the residual values ​​obtained by different wheels in the detection area, which specifically includes the following steps: Calculate the average of the first n maximum values ​​of the residual value and use the average value as the threshold sum .

9. The method for detecting a jamming fault of a low-speed rail wheel according to claim 8, characterized in that: In step S4, constructing a jamming determination algorithm to perform fault determination on wheel jamming includes the following steps: After obtaining the threshold sum Then, find the residual value sequence that exceeds the threshold The calculation is as follows: If there are consecutive k m or more residual value sequences Greater than threshold sum When the wheel is considered to have a stuck fault, a stuck sign is added. In the present invention, the judgment condition k is generally set m =3; otherwise, it is determined that the wheels in the video are running normally.

10. The method for detecting a jamming fault of a low-speed rail wheel according to claim 1, characterized in that: Step S5 specifically includes: synchronously capturing images of the trolley side panels through the camera in the rail low-speed wheel jam fault detection platform, training a target detection model for the side panel numbers, and acquiring the numbers of the trolley side panels currently passing through; after completing the wheel jam fault detection, uploading the wheel jam detection results and the numbers of the trolley side panels at the current fault time to the engineers at the same time.

Images