A method and system for detecting the broken side of the bearing saddle of a railway freight car

Abstract

The invention discloses a method and system for detecting a broken fault of a bearing saddle rib of a railway freight car, belonging to the technical field of freight train detection. The present invention aims to solve the problem of low efficiency and low accuracy in the manual image inspection method for detecting the broken side of the saddle, and the problem of low accuracy in detecting the broken fault of the saddle by the existing automatic image detection method. The invention collects images and intercepts sub-images including bearing saddle ribs and rolling bearing front cover components; puts the sub-images into the data set, marks two parts of bearing saddle rib areas and rolling bearing front cover parts near the bearing saddle; trains bearing The deep learning network model for the segmentation of the saddle rib and the rolling bearing front cover; in the detection process, use the deep learning network model for the segmentation of the bearing saddle rib and the rolling bearing front cover to predict the image of the bearing saddle rib to be tested, obtain the segmentation result, and realize the bearing Judgment of broken saddle ribs. It is mainly used for fault detection of saddle rib breakage.

Description

technical field

[0001] The invention relates to a fault detection method for the broken side of a load-carrying saddle of a freight car. It belongs to the technical field of freight train detection. Background technique

[0002] The load-carrying saddle is an important part of the parts of the truck. When the load-carrying saddle is broken, it may cause a dangerous driving safety situation. In the fault detection of the bearing saddle rib breaking, the existing method adopts the method of manually checking the image to detect the fault. Due to the fact that the inspectors are prone to fatigue and omissions during their work, resulting in missed inspections and wrong inspections, affecting driving safety.

[0003] Because the structure of the saddle is irregular and relatively complex, and due to the light problem at the location of the saddle, as well as the influence of factors such as stains, the ribs of the saddle in the image are not easy to distinguish from other part...

Examples

specific Embodiment approach 1

[0068] Specific implementation mode one: refer to figure 1 To describe this embodiment in detail,

[0069] This embodiment is a method for detecting a broken side of a saddle of a railway freight car, which includes the following steps:

[0070] 1. Collect data and label it

[0071] Take high-definition truck images through the railway pillow equipment. According to the wheelbase information of the train and the prior knowledge of the position of the saddle rib, the sub-image including the saddle rib and the front cover of the rolling bearing is intercepted from the whole truck image. Put the sub-images into the dataset, the size of each sub-image needs to be consistent, and the size of the image must be an integer multiple of 32.

[0072] Because the image of the truck is easily affected by natural conditions such as rain, snow, wind and sand, mud and oil, and the shape of the saddle ribs of different bogie types will also be different. Therefore, in the process of collec...

specific Embodiment approach 2

[0102] This implementation mode is a fault detection system for the breakage of the bearing saddle rib of a railway freight car, and the system includes:

[0103] The image collection unit is used to collect the image of the whole truck, and intercept the sub-image including the bearing saddle rib and the front cover of the rolling bearing;

[0104] A segmentation processing unit, the segmentation processing unit uses a deep learning network model for predicting the image of the bearing saddle rib to be detected, and obtains the segmentation result of the bearing saddle rib and the front cover of the rolling bearing;

[0105] The described deep learning network model is as follows:

[0106] Input the image into the model, the size of the image is an integer multiple of 32, and use it as the input layer;

[0107] Use a 32-channel, 3×3-sized convolution kernel to convolve the input layer to obtain the conv1 layer, and use a 32-channel, 3×3-sized convolution kernel to convolve t...

technical field

[0001] The invention relates to a fault detection method for the broken side of a load-carrying saddle of a freight car. It belongs to the technical field of freight train detection. Background technique

[0002] The load-carrying saddle is an important part of the parts of the truck. When the load-carrying saddle is broken, it may cause a dangerous driving safety situation. In the fault detection of the bearing saddle rib breaking, the existing method adopts the method of manually checking the image to detect the fault. Due to the fact that the inspectors are prone to fatigue and omissions during their work, resulting in missed inspections and wrong inspections, affecting driving safety.

[0003] Because the structure of the saddle is irregular and relatively complex, and due to the light problem at the location of the saddle, as well as the influence of factors such as stains, the ribs of the saddle in the image are not easy to distinguish from other part...

Examples

specific Embodiment approach 1

[0068] Specific implementation mode one: refer to figure 1 To describe this embodiment in detail,

[0069] This embodiment is a method for detecting a broken side of a saddle of a railway freight car, which includes the following steps:

[0070] 1. Collect data and label it

[0071] Take high-definition truck images through the railway pillow equipment. According to the wheelbase information of the train and the prior knowledge of the position of the saddle rib, the sub-image including the saddle rib and the front cover of the rolling bearing is intercepted from the whole truck image. Put the sub-images into the dataset, the size of each sub-image needs to be consistent, and the size of the image must be an integer multiple of 32.

[0072] Because the image of the truck is easily affected by natural conditions such as rain, snow, wind and sand, mud and oil, and the shape of the saddle ribs of different bogie types will also be different. Therefore, in the process of collec...

specific Embodiment approach 2

[0102] This implementation mode is a fault detection system for the breakage of the bearing saddle rib of a railway freight car, and the system includes:

[0103] The image collection unit is used to collect the image of the whole truck, and intercept the sub-image including the bearing saddle rib and the front cover of the rolling bearing;

[0104] A segmentation processing unit, the segmentation processing unit uses a deep learning network model for predicting the image of the bearing saddle rib to be detected, and obtains the segmentation result of the bearing saddle rib and the front cover of the rolling bearing;

[0105] The described deep learning network model is as follows:

[0106] Input the image into the model, the size of the image is an integer multiple of 32, and use it as the input layer;

[0107] Use a 32-channel, 3×3-sized convolution kernel to convolve the input layer to obtain the conv1 layer, and use a 32-channel, 3×3-sized convolution kernel to convolve t...

Images