Engine Removal Projection Tool

The US Navy has over 3500 gas turbine engines used throughout the surface fleet for propulsion and the generation of electrical power. Past data is used to forecast the number of engine removals for the next ten years and determine engine down times between removals. Currently this is done via a FORTRAN program created in the early 1970s. This paper presents results of R&D associated with creating a new algorithm and software program. We tested over 60 techniques on data spanning 20 years from over 3100 engines and 120 ships. Investigated techniques for the forecast basis including moving averages, empirical negative binomial, generalized linear models, Cox regression, and Kaplan Meier survival curves, most of which are documented in engineering, medical and scientific research literature. We applied those techniques to the data, and chose the best algorithm based on its performance on real-world data. The software uses the best algorithm in combination with user-friendly interfaces and intuitively understandable displays. The user can select a specific engine type, forecast time period, and op-tempo. Graphical displays and numerical tables present forecasts and uncertainty intervals. The technology developed for the project is applicable to other logistic forecasting challenges.