Real-time performance reliability prediction

Abstract

The purpose of this paper is to describe an approach to real-time reliability prediction, applicable to an individual product unit, operating under dynamic conditions. The concept of conditional reliability estimation is extended to real-time applications using time-series analysis techniques to bridge the gap between physical measurement and reliability prediction. The model is based on empirical measurements, self-generating, and applicable to online applications. This approach has been demonstrated to the prototype level. Physical performance is measured and forecast across time to estimate reliability. Time-series analysis is adapted to forecast performance. Exponential smoothing with a linear level and trend adaptation is applied. This procedure is computationally recursive and provides short-term, real-time performance forecasts which are linked directly to conditional reliability estimates. Failure clues must be present in the physical signals, and failure must be defined in terms of physical measures to accomplish this linkage. On-line, real-time applications of performance reliability prediction are useful in operation control as well as predictive maintenance.