Experimental Data Analysis Of EEMD-SVD And DBN Acoustic Emission Signals

Bogie(also known as running part)is one of the five components of rail vehicles,its working environment is very bad.As a decisive factor for the speed and safety of rail vehicles,the axle,which connects two train wheels on the bogie,has always been regarded as a key moving component in the whole rail vehicles.With the rapid development of China's national economy,railway transportation is developing towards higher speed and heavier haul.During the operation of railway vehicles,the axle is not only affected by its own material and physical design,but also by hit against,corrosion,temperature and cyclic external forces and other external factors.They are often the main causes of wear,cracks and even breakage of train axles.If the axle of a railway vehicle breaks during operation,it will directly threaten the safety of passengers'lives and property.In order to avoid the occurrence of misfortune and protect the safety of life and property of the country and people,it is very important to monitor the fatigue crack of railway vehicle axle in real time.In this paper,a new method based on EEMD-SVD feature extraction and DBN axle crack acoustic emission signal recognition is proposed.Firstly,acoustic emission(AE)technology in nondestructive testing is used to sample the actual motion state of train axles,and three kinds of ae signals are obtained,namely crack ae signal,knock ae signal and noise ae signal.Then,the collected ae signals are decomposed into a number of stable IMF components through EEMD algorithm.In case of axle failure,the energy value of the signals will change,and the optimal IMF component will be selected based on the principle of maximum energy ratio of IMF component.On this basis,a signal feature extraction method based on EEMD algorithm and SVD algorithm is proposed.Singular value decomposition was carried out on the best IMF classifier to obtain singular matrix.Finally,the classification and recognition of axle pair obstacle types were conducted by DBN system,BPNN system,ELM algorithm and SVM algorithm respectively,and the final output results were analyzed and compared.The experimental results show that the recognition rate of the acoustic emission signal after EEMD-SVD feature extraction is significantly higher than that of the acoustic emission signal originally collected.At the same time,the accuracy of crack classification and recognition by DBN system classifier is also 'higher than that by BPNN system,ELM algorithm and SVM algorithm classifier.