Study On On-line Prognostic For Rotary Machine

With the development of modern economy,science and technology,machinery equipment is gradually complex,high-speed,automated and intelligent.The degree of mutual coupling among components is also getting higher and higher.Once one of the components damages,it often triggers a chain reaction,resulting in the damage of entire equipment or production line.Rotary machinery is not only a source of mechanical power,but also the key part of transmission.Because the rotary machinery often needs to bear the impact of high speed and heavy load during operation,it is very easy to cause machine faults such as damage,breakage,abrasion and so on.Therefore,the research on fault prognostic of rotating machinery is very necessary.As a kind of deep learning algorithm,LSTM(Long Short-Term Memory)network is becoming more and more important in the field of time series prediction.This paper briefly introduces the basic principle of LSTM network and applies it to the field of state prediction and estimation of remaining useful life for rotary machinery.At the first,CEEMDAN(complete ensemble empirical mode decomposition with adaptive noise)soft threshold de-noising algorithm combined CEEMDAN with soft threshold de-noising algorithm is proposed in this paper because the original vibration signal acquired by accelerometer contains a lot of noise.The original signal is decomposed into a number of relatively stable and clean IMFs(Intrinsic Mode Functions).In this paper,the IMFs entropy of energy after CEEMDAN soft threshold de-noising is selected as the state feature for the machine state prediction.To estimate the remaining useful life for rotary machinery,we choose the relative features of the clean signal after reconstructing as characteristic information.Finally,the theory proposed in this paper is proved by the rolling bearing lifetime public datase from the University of Cincinnati IMS(Intelligent Maintenance Systems).The result shows that LSTM network model is of great industrial application value in the fields of state prediction or estimation of remaining useful life for rotary machinery.