Study On The Fault Prediction Methods Of Mechanical Driven Components Based On Grey Theory

Taking the system with incomplete and uncertainty information as the research object,Grey system theory was established by Professor Deng in1982. It has been widely appliedin many fields such as economics, weather forecasting, agricultures, industries and etc. Asthe key components of mechanical driven system, bearings and gearings have always rununder adverse working conditions of high rotating speed and load, which may result inhigh failure rates. If not repaired timely, the faults may cause substantial damage.Therefore, the research on faults prognosis is of great theoretical significance and practicalvalue. Mechanical system usually has the properties of uncertainty, dynamictime-variability and small sample data. Taking this into account, the Grey system theory isutilized in this paper to predict the health state of the mechanical driven components. Themain research content and conclusion include:(1) After studying the typical gray prediction module GM(1,1) deeply, we find that itwill lead to a relatively higher prediction error when handling the data with impulsedisturbances. Aiming at this problem, a data pretreatment method based on weakeningbuffer operators is presented for better prediction effect;(2) With weakening buffer operators, the original discrete data sequence can besmoothed very well and changed into a more stable sequence. An improved GM(1,1)module based on the weakening buffer operators is correspondingly established. It isapplied to the lifetime prediction of gears and the experimental results show animprovement both on the accuracy and precision.(3) The typical multi-parameters prediction module MGM(1,m) has the defectivenessthat the replacement by the approximated value will cause deviation. Consequently amulti-parameters prediction module MGM(1,m) based on the optimization of backgroundvalue is proposed. It is applied to the fault prognosis of bearings and the experimentalresults indicate that the presented module outperforms the typical module MGM (1,m) interms of the prediction precision.