ARMA&GARCH와 회색이론을 이용한 기계 상태 예지

Abstract

Appropriate strategy of machinery maintenance is one of significant factors to enable companies’ owners to improve their product quality and to reduce manufacturing cost. Vibration is considered to be the best operating parameter to judge dynamic conditions such as imbalance (overall vibration), bearing defects and stress applied to components. However, a real severity of vibration may not be correctly recognized due to mechanical and electrical noises from the measuring equipments. This research presents two prognostic methods for methane compressor and speed reducer. The first is a new method based on Grey model and survival probability for machine degradation prediction, and the second is a novel application of autoregressive moving average (ARMA) and generalized autoregressive conditional heteroscedasticity (GARCH) to evaluate and predict the actual severity of vibration collecting from machines to aid in making more accurate conclusions about their health condition. In this work, ARMA and GARCH models are respectively utilized to specify conditional mean and conditional variance of vibration data. The mutual combination of ARMA and GARCH models will be able to give and accurate or real severity of machinery vibration. The forecasts of combined model, So-called ARMA/ GARCH model, play and important role in making decisions on machine repair or possible improvements in order to reach its maximum run-ability, before any unplanned breakdown. In the Grey model prediction result, a modification of GM(1,1) has been made to improve the accuracy of prediction, since the model is built by using only four input data. It is able to track closely the sudden change in machine degradation condition. Real trending data of low methane compressor and speed reducer acquired from condition monitoring routine are employed for evaluating the proposed method. In the ARMA/GARCH prediction model result, it shows more than 90% accuracy. This provides a systematic study of using the ARMA/GARCH prediction model to identify the occurrence and growth of machine fault based on the signal of acceleration peak recorded from a real system of low methane machine in a petrochemical plant and speed reducer in the wind turbine. The quite accurate results indicate the adequacy of the proposed model used in the machine condition monitoring system as well as its application in the CBM system.