Equipment fault forecasting based on a two-level hierarchical model

The analysis of historical time series data that reflects equipment failures is becoming increasingly important in maintenance policies in manufacturing plant. In this paper, we propose a two-level hierarchical modeling framework whose higher level is a model for trend prediction, while whose lower level is a model for residual prediction. Solving the lower level problem is the main focus of this paper. Auto-regressive moving average (ARMA) model is used for residual prediction. One data transformation method is adopted to obtain mean stationary time series by using a defined historical data, which is calculated by an algorithm. The ARMA model which is extensively used in trend and future behavior prediction is used to provide a rigorous prediction of the residual series extracted in the data transformation method. By combining trend prediction and residual prediction approaches, the proposed method can effectively handle the non-linear situation with equipment of highly complicated and non-stationary nature. Its effectiveness has been illustrated by an analysis of real-world data. The proposed method is helpful to reflect the equipment condition and thereby can aid predictive maintenance in manufacturing and reduce the downtime.