Research On The Detection Technology Of The Abnormal State Of The Bolt On The Axle End Of Metro Train

With the rapid development of urban rail transit,the subway is an indispensable means of transportation for most people.In order to ensure reliable and safe train operation,train inspections require regular manual inspections.However,there are the following problems in the manual detection security mode: there may be missed or false detections,and the inspection efficiency is low,which brings hidden trouble to the driving safety.Among them,the shaft end bolts of subway train bogies are key components,and their small size and large number have brought certain difficulties to the detection.Therefore,this subject is designed a fast detection algorithm for the shaft end bolts of subway trains with guaranteed accuracy.Firstly,due to the unevenness of the speed of the images returned by the side train acquisition equipment of the subway train,there are distortion problems such as compression or stretching in the horizontal direction.The total number of images taken at different times for the same train deviates,which makes it difficult for subsequent positioning and recognition.Therefore,a train image registration method based on block type is studied.By setting a set of standard images,registering images into fixed-length blocks,stretching or compressing one of them,and then calculating the corresponding block similarity for the width of the same standard image block with the highest similarity,and so on all the parts are stitched together,and fast registration is achieved through the vehicle image.After registration,use the coordinate positioning method to cut the shaft end area.Secondly,a two-stage model cascade of detection is studied.In the first stage,the existing axle box bolts lacked negative samples,and the robustness is studied.The background modeling method is studied.Use positive Gaussian model to train positive samples to generate rough detection model,which is stable and self-learning.It has certain robustness to positive samples under different working conditions,and can detect abnormal bolts different from the positive sample model.Entering the end cover area of each axle box into the Gaussian model can quickly remove a large number of normal bolt samples,and then detect the suspected abnormal samples before entering the next stage of the detection model for verification.In the second stage,aiming at the problem of complex working conditions,uneven brightness and single feature of most detection methods in the running part of subway train,based on the previous rough detection,image processing technology was used to extract the edge and texture features of bolts,and a multi-feature fusion based bolt detection algorithm was studied.Enter the bolt area of the abnormal sample detected in the previous step into the model to achieve accurate detection of the abnormal bolt.The detection algorithm ensures a very low false alarm rate,improves the detection accuracy,and reduces the false alarm rate.The analysis of experimental results verifies the feasibility and accuracy of the algorithm.