The Application Of One-Dimensional CNN Model In Acoustic Emission Signal Of Axle Fatigue Crack

With the rapid development of the rail transportation industry,the running speed,passenger traffic and cargo load of rail trains are gradually increasing.CR400 AF,the Fuxing EMU independently developed and designed by China,was first on the Beijing-Shanghai highspeed rail on June 2,2017.The starting speed is 350 Km / h,which is the highest commercial speed in the world so far.Although the research and development technology of high-speed railways is developing rapidly,safety problems are always inevitable.The faster the vehicle speed,the greater the safety risks it bears.As a key part in the bogie of high-speed trains,the axles continue to bear huge pressure during work,and cracks will inevitably occur,and the cracks will expand to a certain extent,which will cause the axles to break,which will bring extremely serious consequences.Therefore,there is an urgent need to establish a reliable health monitoring system,monitor and analyze the signals received by sensors placed near the axle,and perform classification identification and remaining life prediction to finally achieve the purpose of remotely real-time acquisition of the failure status of the axle.However,most of the traditional diagnostic models rely on experts to manually extract features,and then realize the diagnosis through shallow model training.This method is time-consuming and labor-intensive,cannot guarantee the universality of the algorithm,and has a long-term dependence on expert experience.In response to this problem,this article combines acoustic emission technology with deep learning,and takes the axles of railroad trains as the research object,proposes an intelligent diagnosis model based on convolutional neural network,which can automatically complete feature extraction,Realize classification recognition and life prediction without manual experience and assistance.First,a recognition model with five convolutional layers is proposed,which is characterized by combining the large convolution kernel of the first layer with a plurality of consecutive small convolution kernels.The model’s recognized rate on the original signal exceeds 99%,and the training time is short.At the same time,the simulation model is verified under the environment of variable load and noise interference,that method is still applicable.Secondly,try to replace the fully connected layers with convolutional layers and pooling layers,which not only improves the prediction accuracy but also reduces the calculation pressure.When proposed model was tested on different data sets,the results show that can predict the axle life more accurately.Compared with the prediction results of traditional models such as support vector regression and classic CNN model,the results show that the error of the CNN prediction model of the fully convolutional layer is lower and the scoring coefficient is higher.