Motor train unit vehicle fault dynamic image detection method based on deep learning

Abstract

The invention discloses a motor train unit vehicle fault dynamic image detection method based on deep learning. Firstly, a motor train unit body fault is directly and preliminarily detected through adeep learning target detection algorithm; and then the motor train unit vehicle fault dynamic images are comprehensively analyzed in cooperation with logic judgment of the vehicle body in the normal state and the fault state, and finally whether the vehicle body breaks down or not is judged according to the analysis result, and early warning is conducted. The invention discloses a motor train unitvehicle fault dynamic image detection method based on deep learning, which relates to the technical field of traffic vehicle overhauling. Vehicle body abnormity is detected through a deep learning target detection algorithm, and a motor train unit vehicle fault dynamic image is comprehensively analyzed in cooperation with logic judgment under the normal state and the fault state of a vehicle body. A fault judgment result is finally obtained. The method has the advantages of being high in speed, high in accuracy and high in stability, meanwhile, the operation quality and the operation efficiency of the motor train are improved, and the capacity of discovering hidden faults in motor train overhauling is enhanced.

Description

technical field [0001] The invention relates to the technical field of traffic vehicle maintenance, in particular to a deep learning-based dynamic image detection method for vehicle faults in EMUs. Background technique [0002] As my country's railways enter a new period of leapfrog development, high-speed EMUs are flying across the vast land of China. Under high-speed operation, any small and subtle faults may cause major accidents. Traditional human-based inspection operations This method is easy to cause missed inspections, and it is difficult to guarantee the quality and efficiency of inspection operations, which increases the probability of potential safety hazards in the operation of EMUs. Therefore, the ability to detect the status of components and abnormal warnings for EMUs in long-distance and high-speed operation is improved, and the For the quality and efficiency of EMU maintenance, it is very important to strengthen the monitoring of the quality of EMU maintenance...

AI Technical Summary

Problems solved by technology

[0002] As my country's railways enter a new period of leapfrog development, high-speed EMUs are flying across the vast land of China. Under high-speed operation, any small and subtle faults may cause major accidents. Traditional human-based inspection operations This method is easy to cause missed inspections, and it is difficult to guarantee the quality and efficiency of inspection operations, which increases the probability of potential safety hazards in the operation of EMUs. Therefore, the ability to detect the status of components and abnormal warnings for EMUs in long-distance and high-speed operation is improved, and the For the quality and efficiency of EMU maintenance, it is very important to strengthen the monitoring of the quality of EMU maintenance operations. At present, EMU inspection operations are facing three major difficulties:

[0003] First, the new operating mode of EMUs leads to insufficient detection capabilities. When the EMUs turn back and run, there is a lack of detection methods for the state of car body components, and the ability to early warning of faults and the ability to detect hidden faults is insufficient;

[0004] Second, the high complexity of the EMU highlights the lack of detection capabilities. The structure of the EMU components is highly complex, the scope of maintenance is large, the number of small parts is large, the amount of detection is large, and the pressure on the guarantee of the EMU out of the warehouse is high;

[0005] Third, it is difficult to monitor the quality of overhaul work in different places. For long-distance EMUs, they often need to be overhauled in different places. This requires the sub-operating stations to grasp and record the component information in a timely manner, which involves the relationship between different operating stations of different railway bureaus The large number of personnel indirectly increased the difficulty of monitoring the quality of maintenance work

[0007] At present, the fault dynamic image detection system of EMU vehicles mainly uses high-speed area array cameras and high-speed line array cameras installed on the side of the track to collect images of visible parts such as the bottom of the EMU car body and the sides of the car body, and uses automatic recognition technology to identify the car body Faults, to achieve hierarchical alarms of faults, and at the same time, the images are transmitted to the indoor monitoring terminal through the network in real time, and the abnormal alarms are confirmed manually and the faults are submitted. However, the existing technology has the following problems: First, the automatic identification technology is divided into image registration In the two stages of image feature analysis and comparison, image registration needs to use historical images of the car body as a reference, so it is necessary to establish a database of the corresponding car body, and the workload in the early stage is huge; second, the image feature analysis and comparison uses prior similarity By comparing the features of the image, the image change area and the degree of change can be obtained. It is sensitive to the brightness, angle, distance, etc. of the car body, and is prone to false alarms. Third, manual confirmation of abnormal alarms requires a lot of manpower. The workload is huge, and the professional quality and attention of people are extremely high, which is prone to false positives

Images