Experimental Data Analysis Of Axle Acoustic Emission Signal Based On Optimization Algorithm And DBN

With the high-speed construction and development of China’s modern railway,the reliability requirements for the operation of rail vehicles are getting higher and higher.The axle is an important part of the rail vehicle,which has to bear almost all the load of the train and the huge impact caused by vibration in the process of running.Moreover,the environment of the axle when it is working is quite bad.As a result,the axle is one of the most vulnerable parts,which often suffers from wear,crack and fracture.Therefore,fault diagnosis of fatigue crack fault and residual life prediction of axle are both important means to avoid accidents and have certain practical significance.The deep confidence network is used to realize the classification and identification of ae signals from axle fatigue cracks and the prediction of axle life.It is not only simple and feasible,but also improves the recognition accuracy and prediction accuracy compared with traditional methods.The number of hidden layer nodes in depth confidence network is a very important parameter.When the number of hidden nodes is large,the operation speed is slow,which affects the calculation time.When the number of hidden nodes is small,the running speed will be accelerated,but the algorithm will run unsteadily,and the recognition accuracy and prediction accuracy will be reduced.Besides,the traditional method to determine the number of hidden layer nodes is to find the appropriate number of hidden layer nodes according to the empirical formula or through repeated attempts according to the specific situation.The traditional method not only has a huge workload,but also may not be able to find the most appropriate value.Therefore,genetic algorithm is used to optimize the number of hidden nodes in the deep confidence network recognition model and differential evolution algorithm is used to optimize the number of hidden nodes in the deep confidence network prediction model.In the axle crack signal recognition part,the deep confidence network recognition model after genetic algorithm optimization is used to classify and identify acoustic emission signal data and data extracted by EEMD-SVD features,and the results are compared with DBN network and other traditional algorithms that determine the number of nodes in the hidden layer by empirical formula.In the life prediction part,differential evolution algorithm is used to optimize the number of nodes in the hidden layer of DBN prediction network,and the life prediction is carried out with the DBN prediction network after the optimization.The prediction results obtained by DBN prediction network after searching for optimization are compared with those obtained by DBN prediction network without searching for optimization.The experimental results show that the deep confidence network optimized bythe optimization algorithm performs well in both recognition and prediction,and greatly saves the time cost.