Gearbox Fault Diagnosis Based On Sequential Probability Ratio Test And Wavelet Neural Network

Gearbox is a machinery commonly used for transfer power and change the speed,it is widely used in mechanical transmission because of its compact structure,high efficiency,long life and reliable work.But it is always operate for long-term with load,use in bad conditions,so it is easily lead to malfunction.So the condition monitoring and fault diagnosis for gearbox is of vital importance to guarantee the machinery run properly.The fault diagnosis of gearbox mainly includes the obtain of fault information,fault feature extraction and fault mode recognition.This article use sequential probability ratio test to extract statistical characteristic,then finish the fault classification combined with neural network,and realize the intelligent diagnosis for gearbox.Sequential probability ratio test(SPRT)is a statistical decision-making procedure based on the sequence of observations,and it is widely used in fault diagnosis.First,set up the statistical mode based on sample observation data,and then put forward the basic assumption for the tested problems,finally,check out the specific fault and make judgement.SPRT is a simple algorithm and has high efficiency.To verify the reliability of experiment result,the three-level SPRT combined with the root mean square error and sequential probability ratio test is used to estimate the malfunction of the gearbox.It shows that this method is effective and reliable.The vibration signal of the gearbox contains much no-linear and non-gauss noises.First,we should filter the noise signal,reduce the noise distraction for normal vibration,use wavelet packet transform method which has well de-noising effect to make pretreatment for the vibration signal of the gearbox,and then use time domain analysis to extract characteristic values,select the kurtosis which is sensitive to impact vibration as characteristic values,and apply SPRT algorithm into gearbox fault mode inspection and recognition,then achieve the purpose of fault diagnosis.When making classification of the malfunction,we should set up a wavelet neural network model,It is a new neural network model combined with time-frequency localization properties of wavelet transform and neural network with self-learning function,it has better fault tolerance and approximation ability.The final result indicated that use SPRT to extract the statistical characteristic,and then use wavelet neural network to make classification is a better intelligent diagnosis.The diagnosis effect has an obvious improvement.