A Practical Example of Applying Machine Learning to a Real Turbofan Engine Issue: NEOP

Abstract

There are high expectations for the use of Machine Learning algorithms in Engine Health Management, but the practical application for use with turbofan engines is often hindered by small sample sizes and noisy data.  This paper discusses a case in which Machine Learning techniques were combined with domain expertise to develop a classifier called Non-seal Erratic Oil Pressure (NEOP).  This classifier is used as an engineering tool to support manual review of engines flagged with Honeywell’s OPX (Oil Pressure Transducer) algorithm.  The purpose of the classifier is to assist a human in analyzing engine trend data from the HTF7000 turbofan engine, when the OPX algorithm identifies an engine with erratic oil pressure.  The NEOP history provides an additional data source when deciding if aft sump maintenance is needed to replace a worn carbon seal, or if the erratic signal is associated with some other cause.  The OPX algorithm has enabled the prevention and avoidance of costly unscheduled engine failures resulting in millions of dollars in documented savings, and the NEOP algorithm helps to ensure that the conclusions from the OPX process continue to result in the appropriate engines being identified for maintenance inspection and corrective action.