Identification Of Acoustic Emission Signals Of Fatigue Cracks Of Axles Based On BP Neural Network And Deep Belief Network

In the component system of high-speed railway vehicle,the axle is undoubtedly one of the key components of the train running part.It is the key component to bear the dynamic load of the vehicle.Because of its own factors or harsh working environment,its running is often accompanied by friction and corrosion.The disturbance of cyclic stress will lead to fatigue crack,surface damage and even fatigue fracture of axle,which will lead to the occurrence of accident.Therefore,it is of great significance to detect and diagnose the axle faults in real time and on line,to ensure the safe and stable operation of the train and to ensure the personal and property safety of passengers.The study on the identification of fatigue cracks caused by the vertical load on the axle during operation is divided into two parts.First of all,in order to simulate the real running state of the axle,in the fatigue crack experiment of the acoustic emission axle,the experimental axle was struck to simulate the wheel-rail impact load,and a strong background noise was introduced to simulate the aerodynamic noise in the high-speed train operation.The time domain analysis was performed on the signal data obtained from the experiment.Twelve kinds of statistical parameters were selected to calculate the signal and the characteristic values of various signals were extracted.Then for the recognition of the axle fatigue crack signal,this paper introduces the BP neural network in artificial neural network and the deep belief network in deep learning theory.The powerful nonlinear mapping ability and self-learning and self-adaptive ability of BP neural network have been widely used in the field of fault diagnosis.As a new type of machine learning intelligent network,deep belief network can realize the multi-hidden layer training calculation.The multi-hidden layer can more autonomously and deeply mine the characteristics of the data,making the accuracy of the operation results higher.This paper uses these two methods to establish a network model for the identification of axle fatigue cracks,and conduct comparison and analysis study to classify and identify axle fatigue crack signals obtained in the experiments.The experimental results show that compared with BP neural network,the depth belief network is more effective than BP neural network in the classification and identification of axle crack signals obtained from acoustic emission experiments,and the intelligent degree of crack detection is higher.