Advanced inventory planning and forecasting solutions: a case study of the UKTLCS Chinook maintenance programme

Abstract

This article which advances on earlier published work by the authors, evaluates the forecasting performance of the inventory planning and forecasting support system used by Boeing’s United Kingdom Through Life Customer Support to manage inventory for the maintenance of the Royal Air force fleet of Chinook helicopters. Focussing mainly on a sample of the spare-part components considered most influential on maintenance costs, limitations in the current forecasting system are identified. Approaches for improving forecasting performance are explored using Visual Basic for Applications (VBA)-based tools. It is found that generic proprietary inventory and forecasting systems can be enhanced by using VBA tools that allow detailed examination of each component’s demand time series. For propriety reasons, all data have been sanitised.

Keywords:

• forecasting
• inventory planning
• support systems

Notes

^a Items 69, 79 and 83 were excluded as we were not able to establish with exactitude the forecasting method that been used

1. There is a confusion in Wallstrom (2009) in that he uses the same acronym, i.e. MACs to define both Mean Average Change scaled and Mean Absolute Change Rescaled. The MACs calculation we use is the Mean Absolute change Rescaled, Equation 2.29 on page 27 in Wallstrom (2009)

2. Based on MCBF and flying hours

3. Maintenance experts believe temperature changes affect the service needs of the Chinook

4. This calculated the square of the Euclidean distance between any two component demand values as the as the measure of proximity between them. The reason we squared the Euclidean distance was to magnify the dissimilarities between components, giving us the best chance to extract clear clusters

5. For the purpose of this study the two parameters for demand size and demand interval will be treated as the same.

6. The Holt Winters Multiplicative Method will only work with non-zero data. The system will recognise this fact and produce no forecast statistic results.

7. Due to the inevitable decrease in time series size, seasonality methods were not compared as the best fitting modelling methodologies.