Research On Vehicle-mounted Detection Method Of Rail Defect Stage Based On Convolutional Neural Network Algorithm

The traditional non-destruction technique of high-speed railway occupies the rail line and they are ineffective.In order to ensure the safe operation of train and obtain rail safety status in time,a new,fast and convenient high-speed rail defect detection method is very necessary.Acoustic emission(AE)technology is derived from seismology,meaning the phenomenon that the strain energy is released in the form of solid elastic wave in the course of fracture and plastic deformation.It has been proved to be applicable in real-time monitoring of railway materials.Based on the research of elastic wave propagation prosperities for AE signals in wheel-rail coupling structures,this paper proposed a new method of vehicle-mounted detection of high-speed rail defect.As the aspect of theoretical modeling,the solid elastic wave model of single wheel rail static contact coupling is established by the finite element analysis software ABAQUS and Hertz contact theory.Trough comparison of signals received in the wheel and rail,it can be concluded that the amplitude of signal will decline under the influence of reflection,but not completely blocked,which means we can extract rail defect features only through the signals received in the wheel.Based on the time domain characteristic parameters,spectrum and modal acoustic emission evaluation analysis of acoustic emission signals characteristics in different rail damage stage,it is shown that compared with the safe signal,parameters of unsafe signal such as kurtosis,representing the impact degree,increases;while effective value,representing the vibration energy will also increase.In addition,the high-frequency components of unsafe signals are more abundant in frequency domain.These features can be used to construct multi-dimensional features of defect to distinguish the life stage of rail.Eventually,an improved algorithm of rail health monitoring based on convolutional neural network and probability analysis of multiple acoustic emission events is proposed to establish classification rules of rail life stage.In contrast,the method proposed is compared with the other several algorithms and proved to be valid and perform better than the previous algorithms.