Comparison of Data-driven Prognostics Models: A Process Perspective

Abstract

Remaining useful life (RUL) prediction is crucial for the implementation of Prognostics and Health Management (PHM) systems, enabling application of predictive maintenance strategies for critical systems (e.g. in aviation, power, railway). Existing literature addresses aspects of data-driven prognostic approaches, with a predominant focus on introducing and testing various novel prediction techniques which are purposed towards improving prediction accuracy performance. However, a relative lack of research can be identified when considering a comparative evaluation of competing for data-driven approaches. In particular, the contributing process elements and characteristics of data-driven prognostics methods are typically not compared in detail. To overcome these drawbacks, this paper aims to evaluate the underlying technical processes for statistical and artificial neural networks (ANN) methods for prognostics. A case study is conducted to implement both approaches on the PHM08 Challenge Data Set for comparison. This research comprehensively compares the statistical and ANN prognostic methods in a systematic manner, covering and comparing their respective technical processes, and evaluates the results with respect to prediction accuracy