Predicting jet engine component wear to enable proactive fleet maintenance

Abstract

The principle driver of maintenance costs for commercial jet engines is the replacement of components that, upon inspection, are determined to be damaged beyond their repairable limits. In order to better predict the lifetime cost of maintaining engines through its flight hour agreement program, Pratt & Whitney aims to predict the probability of needing to replace these parts using information about how an engine has been used. Using historical repair records, we study a suite of statistical models and evaluate their performance in predicting part replacement rates. Despite a preference for interpretable models, we conclude that a random forest approach provides drastically more accurate predictions. We also consider the wider business implications of improved part replacement predictions, particularly as they pertain to forecasting material requirements and reducing volatility upstream in the supply chain.