Research On Improvement Of Trouble Of Moving Freight Car Detection System(TFDS)

With the deepening reform of railway transportation,the train evolves and develops in the direction of high-speed and heavy load.In order to ensure the railway capacity and train operation safety,train fault detection has become an essential link in the railway safety guarantee system.Although the tfds-3 system currently used has good performance,it has not yet achieved full automatic detection in the real sense,and the system has hardly been improved since it passed the technical review of the Ministry of Railways in December 2010.In practical application,there are problems of uneven illumination of collected images and low efficiency of manual train inspection.This paper aims to improve the design of tfds-3,mainly including the design of light compensation system and the study of typical faults of freight cars.The specific contents are as follows:The design of light compensation system is to solve the problem of uneven illumination in the image acquisition stage.Considering the uneven illumination of infrared laser lamp and no adjustment module,high brightness infrared LED linear light source is selected to replace it;The mean value of image gray level is selected as the judgment basis of the light compensation system,the brightness of LED lamp is adjusted by PWM through microcontroller technology,and the light compensation scheme of the light compensation system is designed in combination with the use of TFDs system.However,due to the lack of relevant data and the inability to test on site,the simulation experiment is selected;The simulation experiment selects the automatic light filling device based on STM32.The simulation experiment results meet the expectations and the real-time performance also meets the use requirements,which verifies the feasibility of the light filling system.The research on typical faults of freight cars is to solve the problem of low efficiency of manual train inspection.In this paper,a variety of faults(deflection,defect and deformation)of locking plate are selected as the research object.Considering the shortcomings of traditional image processing technology,the deep learning method is selected in the locking plate area location and locking plate segmentation stage,and three target algorithms of Faster R-CNN,SSD and YOLOv5 are selected for comparison in the locking plate area location stage.The experiments show that YOLOv5 has outstanding advantages in accuracy and speed;In the locking plate segmentation stage,the Unet model with VGG16 feature extraction layer and Res Net50 feature extraction layer are selected for comparative analysis.The experiment shows that the Unet model with VGG16 feature extraction layer has better effect.The contour characteristics of the clamping plate area after binarization are extracted,and the corresponding fault identification algorithm is designed according to the fault characteristics,and the designed algorithm model is verified and analyzed in the locking plate fault detection system written by MFC.It is proved that the algorithm can automatically identify the abnormal situation of the locking plate,achieve the purpose of replacing people’s work,and improve the efficiency and quality of typical fault detection of freight cars.