Research On Typical Fault Recognition Method Of TFDS Based On Machine Vision

With rapid development of national railway system,the trains run faster and faster,and the freight train operation is also showing a trend of high speed,heavy load,but the traditional static detection of parking train inspection personnels own uncertain factors which lead to the train inspection low efficiency,which contradicts security requirements of the rapid development of the freight train.In order to solve this problem,we china researchs and develops the Trouble of moving Freight car Detection System(TFDS),which greatly improved in train inspection efficiency and train inspection quality compared to traditional detection system.But the mode of this system still need artificial recognition.This paper adopts the theory of machine vision,promoting the transformation of TFDS identification work to the "machine control" mode.In this paper,the TFDS a block key losing and truncated plug door handle close the two typical faults present as research background,the theory of machine vision as basis to design automatic recognition and implementation of fault.For block key losing fault,an fault recognition method for block key losing was proposed based on Hough transform.According to the inherent geometric structure of the bogie,the mathematical model is established by using the geometric position relation between the axle,the endpoint and the block key,Which solves the problem that the mathematical model is not applicable due to the structure of the bogie.By using discriminant relative gradient histogram feature extraction and classification method solves the problem caused by lighting reason which affects fault recognition.For truncated plug door handle close fault,an fault recognition method for truncated plug door handle close was designed to recognize fault based prior knowledge combined with the geometric model of auxiliary positioning.According to the geometric relation between the plug door and the handle,and the corresponding prior knowledge,a mathematical model was established to solve the handle detection area rapid calibration problem.Using the method of target area ratio with combination calculate the number of connected regions of after exclusion for fault identification,the segmentation problem caused by bad the image segmentation threshold was solved.Finally,the two fault recognition algorithms were tested,and the results were analyzed.The experimental results show the reliability of the algorithm.