Research On Brake Disc Crack Detection Method Based On Full Convolution Network

The brake disc plays an important role in the braking system of urban rail transit vehicles,and its performance is related to the safety of train,guarantee for the safe operation of the railway.In recent years,with the improvement of train speed,the kinetic energy increases rapidly,and a large amount of heat energy will be generated in the process of emergency braking,if lack of monitoring,the length and number of cracks exceed a certain value,it will cause a significant safety accident.Therefore,in order to ensure the safety of train operation,the effective detection of brake disc shows great significance.Recently,the methods of crack detection generally include detection and traditional image recognition,but the human factors in manual detection have a greater influence,such as the professional level of technical personnel and the experience,result of low efficiency and high labor intensity,However,traditional image recognition method(Sobel,Canny et.al.)under the light disturbance or the noise situation,shows weak performance in crack segmentation,hardly satisfy the requirements of dynamic monitoring of detailed cracks.Therefore,the design of efficient and accurate crack detection system is the urgent trend of future development.Based on this,the automatic crack detection system comes into being.The goal of this paper is to use deep learning technology to extract crack features and parameterization,which can benefit for further research.In this paper,crack detection algorithm of and feature quantitative of brake disc is studied.Firstly,the sematic segmentation technology is applied to brake disc crack to achieve precious segmentation on pixel level.Secondly,the detection results are optimized by morphological processing,hole filling,small patch removal and so on.Finally,the feature information of the crack is quantified based on the detection results.