Statistical analysis of the Naval inventory control point repair turn-around time forecase model

Abstract

Accurate forecasting of repair turn-around time (RTAT) of United States Navy depot level repairable items is critical to achieving optimal service levels while minimizing procurement and repair costs. The Navy's Inventory Control Point has developed a forecast model that uses sophisticated Statistical Process Control techniques and non-parametric algorithms to forecast RTAT. This thesis attempts to validate the Navy's RTAT forecast model by comparing its performance to those of simple time series forecasting methods. It was found that the assumptions implicit in the UICP RTAT forecast model have a significant impact on forecast accuracy. In addition to documenting these model properties, a goal of this thesis is to identify variables that the UICP model does not use in RTAT forecasting which may improve its accuracy. The research focuses on data for repairable items that have high dollar value and the greatest number of repair transactions per quarter. Results show that the Navy's model is not consistently more accurate than any of the alternative techniques examined, and that it tends to ignore many large RTAT observations, causing it to under-forecast RTAT. Thesis research also reveals that accounting for differences in disparate designated overhaul points may significantly improve the prediction of RTAT. Finally it is shown that additional variables, derived from a NAVICP Philadelphia database and designed to capture the queueing aspect of the repair process, may significantly improve the prediction of RTAT. These findings point to the use of queueing information to obtain more accurate RTAT forecasts.