CN-121998896-A - Method and system for identifying oil throwing fault of coupling of motor train unit

CN121998896ACN 121998896 ACN121998896 ACN 121998896ACN-121998896-A

Abstract

The invention discloses a method and a system for identifying a coupling oil throwing fault of a motor train unit, and belongs to the technical field of fault detection of the running state of a railway motor train unit. The method combines the deep learning technology to detect and analyze, firstly, a target detection network is used for positioning the coupling part so as to accurately mark the target area, then, an example segmentation network is used for segmenting the oil throwing area of the positioned coupling image, the position and area information of the oil throwing area are extracted, the information not only provides a basis for whether the fault needs to be reported, but also provides convenient support for a subsequent inspector to process the fault, and the detection efficiency and accuracy are improved.

Inventors

LIU JIALING
YU DONGYANG
ZHANG MINDONG
WANG PANPAN
ZHU SHANWEI
LIU JIANHENG
LIU PENGFEI

Assignees

北京京天威科技发展有限公司

Dates

Publication Date: 20260508
Application Date: 20251204

Claims (10)

1. The method for identifying the oil throwing fault of the coupling of the motor train unit is characterized by comprising the following steps of: 1) Inputting an image to be detected of the motor train unit into a trained improved YOLO11 model, and outputting a positioning result of a coupling region when the coupling region exists in the image to be detected; The improved YOLO11 model is characterized in that a layer SimAM of modules is added after the original structure of the YOLO11 model, and the original CIoU Loss loss function is replaced by a Shape-IoU loss function in training of the improved YOLO11 model; 2) Cutting out a coupling region from an original image to be detected according to the positioning result to obtain a coupling region image; 3) Inputting the coupling region image into a trained example segmentation model YOLACT ++ so as to segment the coupling region image to obtain an oil slinging region; 4) And determining whether the oil throwing fault of the coupling of the motor train unit exists according to the proportion of the oil throwing area in the whole coupling area image.
2. The method for identifying the oil slinging fault of the motor train unit coupling according to claim 1, wherein the SimAM module is used for calculating the square of the difference between each pixel of the feature map extracted by the forward neural network and all pixels in the neighborhood, summing and dividing the sum result by the number of pixels in the pixel domain to obtain a final result, and normalizing the final result and the final result to obtain the attention weight.
3. The method for identifying the oil slinging fault of the motor train unit coupling according to claim 2, wherein the calculation formula for obtaining the attention weight is as follows: ; Wherein the method comprises the steps of Representing the final result of the pixel at (i, j), Representation of Is used for the normalization of (a), Is a constant for preventing divide by zero errors.
4. The method for identifying the oil slinging fault of the motor train unit coupling according to claim 1, wherein the training process for improving the YOLO11 model comprises training the improved YOLO11 model by using a pre-prepared motor train unit image dataset as a training set, a verification set and a test set, and performing iterative training on the improved YOLO11 model by adopting a random gradient descent method in the training process.
5. The method for identifying a coupling oil slinging fault of a motor train unit according to claim 4, wherein the pre-prepared motor train unit image dataset is: cutting the motor train unit image acquired by the rail side industrial camera by using a sliding window cutting mode with an overlapping area to obtain a cut image; And expanding the cut image by utilizing an image enhancement technology, forming a complete data set by the original cut image and the expanded image, and then marking the data set by adopting a VOC (volatile organic compound) format to obtain the image data set of the motor train unit.
6. The method for identifying the oil slinging fault of the motor train unit coupling according to claim 1, wherein in the step 2), the coupling region is cut out from an original image to be detected through the coordinate information of the coupling region position in the positioning result.
7. The method for identifying the oil slinging fault of the motor train unit coupling according to claim 1, wherein the training process of the example segmentation model YOLACT ++ comprises training an example segmentation model YOLACT ++ by using a pre-obtained coupling part image dataset as a training set, a verification set and a test set, and performing iterative training on the example segmentation model YOLACT ++ by adopting a momentum gradient descent optimization algorithm in the training process.
8. The method for identifying the oil slinging fault of the motor train unit coupling according to claim 7, wherein the pre-obtained coupling part image data set is: Applying training to improve an image of a coupling region positioning result output in a YOLO11 model process; cutting out a coupling region from an image of a positioning result of the coupling region to form an original data set; And expanding the original data set, wherein the expanded data and the original data together form a complete data set, and marking the complete data set by using a COCO data set format to obtain the image data set of the coupling part.
9. The method for identifying the oil slinging fault of the motor train unit coupling according to claim 1, wherein in the step 4), the number of oil slinging pixels in a mask of an oil slinging region is obtained through statistical segmentation, the proportion of the oil slinging region in the whole coupling region image is calculated, and when the proportion of the oil slinging region in the whole coupling region image is larger than a preset value, the oil slinging fault exists.
10. A motor train unit coupling oil slinging fault identification system comprising a processor and a memory, the memory storing computer program instructions, the processor being operable to execute the computer program instructions stored in the memory to implement the motor train unit coupling oil slinging fault identification method of any one of claims 1-9.

Description

Method and system for identifying oil throwing fault of coupling of motor train unit Technical Field The invention relates to the technical field of fault detection of the running state of a railway motor train unit, in particular to a method and a system for identifying the oil throwing fault of a coupling of a motor train unit. Background With the rapid development of high-speed railways in China, when a motor train unit (EMU) runs at a high speed, any minor faults on a train body can cause serious safety problems. Therefore, detecting the states of all parts of the EMU in high-speed operation, giving out fault alarms in time, improving the maintenance quality of the EMU and enhancing the monitoring of maintenance operation is of great importance. In order to ensure the running safety of the railway motor train unit, each large railway station in China is provided with a TEDS (motor train unit running fault image detection system), and the system can collect images of visible parts such as the EMU bottom, skirtboards on two sides of the train body, vehicle connecting parts, bogies and the like through a high-speed area array camera and a high-speed linear array camera which are arranged beside a track. The system uses a mode recognition technology to recognize the faults of the vehicle body and provides a grading alarm function. Meanwhile, the image is digitally processed and then fault information is displayed on an information terminal of the detection center, so that a worker can confirm abnormal alarm and submit a fault report, and the operation quality and efficiency of maintenance of the motor train unit are improved. The system has the advantages that at present, only a few parts realize the automatic judging function of faults, a large number of parts of the motor train unit still rely on a manual image and confirm whether faults exist or not, and the modules realizing the automatic judging of faults still have the characteristics of low fault detection rate and low alarm accuracy. For example, for the fault of coupling oil slinging, each train is usually provided with 8 to 16 carriages, and the railway train is heavy in transportation and dense in passing, and the problem of error detection is easily caused by visual fatigue simply by judging whether the fault exists or not by naked eyes of train detection personnel. Therefore, an automatic detection method for the oil throwing fault of the coupling is urgently needed at present to reduce the working intensity of train inspection personnel, improve the detection effect of the fault and ensure the operation safety of a motor train unit. Disclosure of Invention Therefore, the invention aims to provide a method and a system for identifying the oil slinging fault of a coupling of a motor train unit, which are used for solving the problems that the existing manual oil slinging fault detection mode is low in efficiency and is easy to cause false leakage detection. The invention provides a motor train unit coupling oil throwing fault identification method, which combines a deep learning technology to detect and analyze, firstly, a target detection network is used for locating a coupling part so as to accurately mark a target area, then, an example segmentation network is used for segmenting an oil throwing area of a located coupling image, and the position and area information of the oil throwing area are extracted, so that the information not only provides basis for reporting whether faults are needed, but also provides convenient support for a subsequent inspector to process the faults, and the detection efficiency and accuracy are improved. Specifically, the method comprises the following steps: 1) Inputting an image to be detected of the motor train unit into a trained improved YOLO11 model, and outputting a positioning result of a coupling region when the coupling region exists in the image to be detected; The improved YOLO11 model is characterized in that a layer SimAM of modules is added after the original structure of the YOLO11 model, and the original CIoU Loss loss function is replaced by a Shape-IoU loss function in training of the improved YOLO11 model; 2) Cutting out a coupling region from an original image to be detected according to the positioning result to obtain a coupling region image; 3) Inputting the coupling region image into a trained example segmentation model YOLACT ++ so as to segment the coupling region image to obtain an oil slinging region; 4) And determining whether the oil throwing fault of the coupling of the motor train unit exists according to the proportion of the oil throwing area in the whole coupling area image. The beneficial effects are as follows: According to the invention, by combining the target detection network and the example segmentation network, the effective monitoring of the coupling state in the running of the train of the railway motor train unit is realized, and the detection efficiency and accuracy