An evaluation of forecasting methods for aircraft non-routine maintenance material demand

Abstract

Aircraft maintenance can be divided into routine and non-routine activities. Material demand associated with non-routine maintenance is typically intermittent or lumpy: it has a large variance in frequency and quantity. Consequently, this type of demand is hard to predict. This paper introduces a method to collect time series datasets for aircraft non-routine maintenance material demand. Non-routine material consumption is linked to scheduled maintenance tasks to gain insight in demand patterns. A structural part selection of the Boeing 737NG fleet of an aviation partner has been sampled to generate various test cases. Subsequently, various forecasting methods are applied to these test cases and evaluated using various accuracy metrics. For the small time series datasets associated with non-routine maintenance, exponentially weighted moving average (EMA) outperformed smoothing methods such as Croston's method (CR) and the Syntetos-Boylan approximation (SBA). To validate the practical applicability of EMA for non-routine maintenance material demand, the method has been applied and verified in the prediction of actual demand for a separate maintenance C-check.