A method for detecting brakes of railway freight cars

Abstract

The present invention discloses a brake detection method for a railway freight car, comprising: acquiring parameter information of the railway freight car; acquiring positioning information of the RGV trolley in real time when the RGV trolley travels along a first direction, generating a first target detection area of ​​multiple train cars in real time based on the parameter information and the positioning information, the RGV trolley collecting a plurality of first detection images within the first target detection area; acquiring a pre-trained neural network model, inputting a plurality of the first detection images into the neural network model to generate a first defect detection result; generating a second target detection area when the RGV trolley travels along a second direction based on the first defect detection result, the RGV trolley collecting a plurality of second detection images within the second target detection area, inputting a plurality of the second detection images into the neural network model to generate a second defect detection result; and generating a brake defect detection report based on the first defect detection result and the second defect detection result.

Description

Technical Field

[0001] The invention relates to the technical field of rail vehicle detection, and in particular to a brake detection method for a railway freight car. Background Art

[0002] To ensure safe operation, fault detection plays a major role in the transportation field. There are many typical examples of using fault detection in railways, aviation, navigation, and highway bridge maintenance. Since the transportation field has a heavy responsibility, once important equipment fails, it will cause huge losses of personnel and property. Therefore, many countries in the world have invested a lot of manpower, material resources and financial resources in the research of fault detection. Fault detection has become one of the research hotspots in the transportation field today.

[0003] For railway freight cars, the operation and maintenance party has set up a large number of train inspection yards to inspect freight cars to ensure safe operation. At present, the train inspection yard mainly relies on manual maintenance, which has problems such as large workload and inability to standardize the inspection results. For this problem, our company's previous research idea is: based on the RGV trolley to continuously collect the bottom pictures of the railway freight cars, then find the pictures containing the brake motor from the pictures, and finally perform fault diagnosis. However, this research idea has the following two problems: all the inspection pictures are screened on the RGV trolley side, but this requires high computing power on the trolley, whether it is heat dissipation, computing power, or size, it is difficult to meet the requirements (the stronger the computing power, the larger the size). According to this idea, the trolley can only be slowed down to adapt to the problem of insufficient computing power of the trolley, but the efficiency will be reduced; the inspection pictures are transmitted to the server side and screened using the server's computing power, but this method requires too high network transmission bandwidth. Although the processing speed becomes faster, the transmission takes too long and cannot meet the real-time requirements.

[0004] In summary, the existing railway freight car brake detection method has the problems of low detection efficiency and poor real-time performance. Summary of the invention

[0005] In view of this, the present invention provides a railway freight car brake detection method, which solves the problems of low detection efficiency and poor real-time performance of traditional railway freight car brake detection methods by improving data collection and data processing methods.

[0006] To solve the above problems, the technical solution of the present invention is to adopt a brake detection method for a railway freight car, including: obtaining parameter information of the railway freight car; when the RGV trolley is traveling along a first direction, obtaining the positioning information of the RGV trolley in real time, and generating a first target detection area for multiple train cars in real time based on the parameter information and the positioning information, and the RGV trolley collects multiple first detection images within the first target detection area; obtaining a pre-trained neural network model, and inputting multiple first detection images into the neural network model to generate a first defect detection result; based on the first defect detection result, generating a second target detection area when the RGV trolley is traveling along a second direction, the RGV trolley collects multiple second detection images within the second target detection area, and inputting multiple second detection images into the neural network model to generate a second defect detection result; generating a brake defect detection report based on the first defect detection result and the second defect detection result.

[0007] Optionally, obtaining parameter information of a railway freight car includes: when the railway freight car enters a train inspection yard, extracting the car number information of the railway freight car and the vehicle wheelbase information and the vehicle axle number information corresponding to the car number information based on a car number recognition unit; when the railway freight car enters a train inspection yard, extracting the number of axles of the part of the railway freight car that has entered the train inspection yard based on an axle counting sensor; and constructing the parameter information based on the number of axles, the car number information, the vehicle wheelbase information and the vehicle axle number information.

[0008] Optionally, the positioning information of the RGV trolley is obtained in real time, including: obtaining the travel distance information based on the encoder of the RGV trolley; generating the information on the number of wheels of the RGV trolley that have passed the railway freight car based on the wheel sensor of the RGV trolley, and using it as auxiliary correction information; generating the positioning information of the RGV trolley compared to the railway freight car based on the travel distance information, the auxiliary correction information and the initial position information of the RGV trolley, wherein the initial position of the RGV trolley is the starting point of the front of the railway freight car.

[0009] Optionally, a first target detection area is generated in real time based on the parameter information and the positioning information, including: when a brake is arranged between the second axle and the third axle of each carriage of the railway freight car, the first target detection area is generated for detecting the area between the second axle and the third axle of each carriage of the railway freight car based on the positioning information, the vehicle wheelbase information and the vehicle axle number information.

[0010] Optionally, pre-training the neural network model includes: constructing an initialized network model, wherein the network model includes a semantic segmentation model; obtaining a training data set and a test data set consisting of brake image samples with multi-category labels, wherein the label categories include brakes and multiple categories of defects of brakes; and training and testing the neural network model based on the training data set and the test data set.

[0011] Optionally, multiple first detection images are input into the neural network model to generate a first defect detection result, including: inputting multiple first detection images of the same car into the neural network model one by one, and when the neural network model generates a brake detection frame for the first time, stopping the image recognition of the remaining first detection images in the same car and inputting multiple first detection images of the next car into the neural network model one by one, and at the same time, if other categories of brake defect detection frames are generated when the neural network model generates the brake detection frame for the first time, then generating a defect detection result for the car based on the brake detection frame and the brake defect detection frame; when the neural network model traverses the first detection images of all cars, generating the first defect detection result based on the defect detection results of all cars.

[0012] Optionally, a second target detection area is generated based on the first defect detection result when the RGV trolley travels along the second direction, including: based on the detection frame area information and coordinate information contained in the brake detection frame of each train car in the first defect detection result, a second target detection area is generated in the first target detection area of ​​each train car, whose area is smaller than the first target detection area and larger than the detection frame area, and the second target detection area includes the area to which the brake detection frame belongs.

[0013] Optionally, multiple second detection images are input into the neural network model to generate a second defect detection result, including: inputting multiple second detection images of the same car into the neural network model one by one, and when the neural network model generates a brake detection frame for the first time, stopping the image recognition of the remaining second detection images in the same car and inputting multiple second detection images of the next car into the neural network model one by one, and at the same time, if other categories of brake defect detection frames are generated when the neural network model generates the brake detection frame for the first time, then generating a defect detection result for the car based on the brake detection frame and the brake defect detection frame; when the neural network model traverses the second detection images of all cars, generating the second defect detection result based on the defect detection results of all cars.

[0014] The primary improvement of the present invention is the brake detection method for railway freight cars. By collecting parameter information of railway freight cars and positioning information of the RGV trolley relative to the railway train, the axle area where the brake is located is accurately located as the first target detection area, which effectively avoids the need for the RGV trolley to continuously collect photos of the bottom of the train in real time during driving, resulting in a large number of photos and thus causing a huge computing load on the data processing unit. The RGV trolley only needs to intermittently detect the first target detection area during driving, reducing the amount of data input into the data processing unit of the RGV trolley, and realizing real-time detection of brake defects by the RGV trolley. At the same time, since the brake cylinder of each carriage may be set on the left or right side of the car body, the RGV trolley can only generate the axle area where the brake is located as the first target detection area based on the axle position of each carriage and the positioning information of the RGV trolley when detecting along the first direction. However, after generating the first defect detection result, the data processing unit can determine whether the brake cylinder of each carriage is set on the left or right side of the car body according to the brake detection frame information, thereby further narrowing the detection area and generating a more accurate second target detection area, thereby further reducing the amount of data input into the RGV trolley data processing unit, realizing real-time and accurate detection of the brake of the railway freight car, and solving the problems of low detection efficiency and poor real-time performance of the traditional brake detection method of the railway freight car. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a simplified flow chart of a brake detection method for a railway freight car according to the present invention. DETAILED DESCRIPTION

[0016] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0017] like Figure 1As shown, a brake detection method for a railway freight car includes: obtaining parameter information of the railway freight car; when the RGV trolley travels along a first direction, obtaining the positioning information of the RGV trolley in real time, generating a first target detection area for multiple train cars in real time based on the parameter information and the positioning information, and the RGV trolley collects multiple first detection images within the first target detection area; obtaining a pre-trained neural network model, and inputting multiple first detection images into the neural network model to generate a first defect detection result; based on the first defect detection result, generating a second target detection area when the RGV trolley travels along a second direction, the RGV trolley collects multiple second detection images within the second target detection area, and inputting multiple second detection images into the neural network model to generate a second defect detection result; generating a brake defect detection report based on the first defect detection result and the second defect detection result. Among them, the first direction can be the direction from the front of the train to the rear of the train, and the second direction can be the direction from the rear of the train to the front of the train; since the first detection image and the second detection image are respectively taken by the RGV trolley at different angles in different driving directions, there are certain differences between the first defect detection result and the second defect detection result output by the neural network model. At the same time, the credibility of the detection result output by the neural network model is affected by environmental interference, camera parameters and other factors. The user can freely choose the method of generating a brake defect detection report based on the first defect detection result and the second defect detection result according to the credibility of the first defect detection result and the second defect detection result output by the neural network model. For example: when the credibility of the first defect detection result and the second defect detection result is high, the brake defect detection report can be generated based on the union of the first defect detection result and the second defect detection result; when the credibility of the first defect detection result and the second defect detection result is low, the brake defect detection report can be generated based on the intersection of the first defect detection result and the second defect detection result.

[0018] Furthermore, obtaining parameter information of railway freight cars includes: when the railway freight car enters the train inspection yard, extracting the train number information of the railway freight car and the vehicle wheelbase information and vehicle axle number information corresponding to the train number information based on the train number recognition unit; when the railway freight car enters the train inspection yard, extracting the number of axles of the railway freight car that has entered the train inspection yard based on the axle counting sensor; and constructing the parameter information based on the number of axles, the train number information, the vehicle wheelbase information and the vehicle axle number information. Among them, the train number recognition unit and the axle counting sensor are both mature general devices in this field, so this application does not specifically limit their setting method and model.

[0019] Furthermore, the real-time acquisition of the positioning information of the RGV trolley includes: acquiring the travel distance information based on the encoder of the RGV trolley; generating the information on the number of wheels of the RGV trolley that have passed the railway freight car based on the wheel sensor of the RGV trolley, and using it as auxiliary correction information; generating the positioning information of the RGV trolley relative to the railway freight car based on the travel distance information, the auxiliary correction information and the initial position information of the RGV trolley, wherein the initial position of the RGV trolley is the starting point of the front of the railway freight car. Among them, the RGV trolley is an automatic trolley based on track guidance, which is a mature general device in this field. Therefore, this application does not specifically limit its setting method and model. At the same time, the RGV trolley can be equipped with encoders, wheel sensors, cameras, data processing units and other conventional detection devices in this field.

[0020] Furthermore, the first target detection area is generated in real time based on the parameter information and the positioning information, including: when the brake is set between the second axle and the third axle of each carriage of the railway freight car, the first target detection area for detecting the area between the second axle and the third axle of each carriage of the railway freight car is generated based on the positioning information, the vehicle wheelbase information and the vehicle axle number information. In this field, the brake is usually set between the second axle and the third axle of each carriage of the railway freight car, so this case is used to illustrate how to generate the first target detection area. When the brake is set between other axles of each carriage of the railway freight car, the area between other axles can still be used as the first target detection area.

[0021] Furthermore, pre-training the neural network model includes: constructing an initialized network model, wherein the network model includes a semantic segmentation model; obtaining a training data set and a test data set consisting of brake image samples containing multi-category labels, wherein the label categories include brakes and multi-category defects of brakes; training and testing the neural network model based on the training data set and the test data set. It should be noted that the neural network model used in this application is a conventional prior art in the field and does not involve further improvement of the model architecture. Therefore, the type and architecture of the neural network model are not specifically limited. The type of the neural network model can be YOLO-V3, FASTER RCNN, etc.

[0022] Further, inputting multiple first detection images into the neural network model to generate a first defect detection result includes: inputting multiple first detection images of the same car into the neural network model one by one, when the neural network model generates a brake detection frame for the first time, stopping the image recognition of the remaining first detection images in the same car and inputting multiple first detection images of the next car into the neural network model one by one, and at the same time, if other types of brake defect detection frames are generated when the neural network model generates the brake detection frame for the first time, then generating the defect detection result of the car based on the brake detection frame and the brake defect detection frame; when the neural network model traverses the first detection images of all cars, generating the first defect detection result based on the defect detection results of all cars. Wherein, the information of multiple types of detection frames output by the neural network model includes (x, y, w, h, i), (x, y) is the coordinate of the upper left corner of the detection frame, w is the width of the detection frame, h is the height of the detection frame, and i is the confidence of the detection frame.

[0023] Furthermore, based on the first defect detection result, a second target detection area is generated when the RGV trolley travels along the second direction, including: based on the detection frame area information and coordinate information contained in the brake detection frame of each train car in the first defect detection result, a second target detection area is generated in the first target detection area of ​​each train car, whose area is smaller than the first target detection area and larger than the detection frame area, and the second target detection area includes the area to which the brake detection frame belongs.

[0024] Furthermore, multiple second detection images are input into the neural network model to generate a second defect detection result, including: inputting multiple second detection images of the same car into the neural network model one by one, and when the neural network model generates a brake detection frame for the first time, stopping the image recognition of the remaining second detection images in the same car and inputting multiple second detection images of the next car into the neural network model one by one, and at the same time, if other categories of brake defect detection frames are generated when the neural network model generates the brake detection frame for the first time, then the defect detection result of the car is generated based on the brake detection frame and the brake defect detection frame; when the neural network model traverses the second detection images of all cars, the second defect detection result is generated based on the defect detection results of all cars.

[0025] The present invention accurately locates the axle area where the brake is located as the first target detection area by collecting parameter information of railway freight cars and the positioning information of the RGV trolley compared to the railway train, thereby effectively avoiding the problem that the RGV trolley needs to continuously collect photos of the bottom of the train in real time during driving, resulting in a large number of photos and thus causing a huge computing load on the data processing unit. The RGV trolley only needs to intermittently detect the first target detection area during driving, reducing the amount of data input into the data processing unit of the RGV trolley and realizing real-time detection of brake defects by the RGV trolley. At the same time, since the brake cylinder of each carriage may be set on the left or right side of the car body, the RGV trolley can only generate the axle area where the brake is located as the first target detection area based on the axle position of each carriage and the positioning information of the RGV trolley when detecting along the first direction. However, after generating the first defect detection result, the data processing unit can determine whether the brake cylinder of each carriage is set on the left or right side of the car body according to the brake detection frame information, thereby further narrowing the detection area and generating a more accurate second target detection area, thereby further reducing the amount of data input into the RGV trolley data processing unit, realizing real-time and accurate detection of the brake of the railway freight car, and solving the problems of low detection efficiency and poor real-time performance of the traditional brake detection method of the railway freight car.

[0026] The above is a detailed introduction to a brake detection method for a railway freight car provided by an embodiment of the present invention. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referenced to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

[0027] 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, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. 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 to be beyond the scope of the present invention.

[0028] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

Claims

1. A method for detecting brakes of railway freight cars, characterized in that: include: Get parameter information of railway freight cars; When the RGV trolley is traveling in a first direction, the positioning information of the RGV trolley is obtained in real time, and a first target detection area of ​​multiple train carriages is generated in real time based on the parameter information and the positioning information, and the RGV trolley collects a plurality of first detection images in the first target detection area; the first target detection area includes the axle area where the brake motor is located; Obtaining a pre-trained neural network model, and inputting a plurality of the first detection images into the neural network model to generate a first defect detection result; Based on the first defect detection result, a second target detection area is generated when the RGV trolley travels in a second direction, the RGV trolley collects a plurality of second detection images within the second target detection area, and inputs the plurality of second detection images into the neural network model to generate a second defect detection result; The second target detection area includes the area to which the brake detection frame in the first defect detection result belongs; A brake defect detection report is generated based on the first defect detection result and the second defect detection result.

2. The brake detection method according to claim 1, characterized in that: Get the parameter information of railway freight cars, including: When a railway freight car enters a train inspection yard, the vehicle number information of the railway freight car and the vehicle wheelbase information and vehicle axle number information corresponding to the vehicle number information are extracted based on the vehicle number recognition unit; When a railway freight car enters a train inspection yard, the number of axles of the railway freight car that has entered the train inspection yard is extracted based on an axle counting sensor; The parameter information is constructed based on the number of axles, the vehicle number information, the vehicle wheelbase information and the vehicle axle number information.

3. The brake detection method according to claim 2, characterized in that: Get the positioning information of RGV in real time, including: The encoder of the RGV car is used to obtain the driving distance information; The wheel sensor of the RGV car generates information about the number of wheels of the RGV car that have passed the railway freight car, and serves as auxiliary correction information; The positioning information of the RGV trolley relative to the railway freight car is generated based on the travel distance information, the auxiliary correction information and the initial position information of the RGV trolley, wherein the initial position of the RGV trolley is the starting point of the front of the railway freight car.

4. The brake detection method according to claim 3, characterized in that: Generating a first target detection area in real time based on the parameter information and the positioning information includes: In the case where the brake is provided between the second axle and the third axle of each carriage of the railway freight car, The first target detection area for detecting an area between a second axle and a third axle of each carriage of the railway freight car is generated based on the positioning information, the vehicle wheelbase information, and the vehicle axle number information.

5. The brake detection method according to claim 1, characterized in that: Pre-training the neural network model includes: Constructing an initialization network model, wherein the network model includes a semantic segmentation model; Obtaining a training data set and a test data set consisting of brake image samples containing multiple categories of labels, wherein the label categories include brakes and multiple categories of defects of brakes; The neural network model is trained and tested based on the training data set and the test data set.

6. The brake detection method according to claim 4, characterized in that: Inputting the plurality of first detection images into the neural network model to generate a first defect detection result includes: Inputting a plurality of the first detection images of the same carriage into the neural network model one by one, and when the neural network model generates a brake detection frame for the first time, stopping the image recognition of the remaining first detection images in the same carriage and inputting a plurality of the first detection images of the next carriage into the neural network model one by one, and at the same time, if other types of brake defect detection frames are generated at the same time when the neural network model generates the brake detection frame for the first time, generating a defect detection result of the carriage based on the brake detection frame and the brake defect detection frame; When the neural network model traverses the first detection images of all carriages, the first defect detection result is generated based on the defect detection results of all carriages.

7. The brake detection method according to claim 5, characterized in that: Generating a second target detection area when the RGV travels in a second direction based on the first defect detection result includes: Based on the detection frame area information and coordinate information contained in the brake detection frame of each train car in the first defect detection result, a second target detection area is generated in the first target detection area of ​​each train car, whose area is smaller than the first target detection area and larger than the detection frame area.

8. The brake detection method according to claim 7, characterized in that: Inputting the plurality of second detection images into the neural network model to generate a second defect detection result includes: Inputting a plurality of the second detection images of the same carriage into the neural network model one by one, and when the neural network model generates a brake detection frame for the first time, stopping the image recognition of the remaining second detection images in the same carriage and inputting a plurality of the second detection images of the next carriage into the neural network model one by one, and at the same time, if other types of brake defect detection frames are generated at the same time when the neural network model generates the brake detection frame for the first time, generating a defect detection result of the carriage based on the brake detection frame and the brake defect detection frame; When the neural network model traverses the second detection images of all carriages, the second defect detection result is generated based on the defect detection results of all carriages.

Images