Naval Supply Systems Command
Manuel D. Rossetti, Justin R. Chimka, Heather L. Nachtmann, Vijith Varghese
University of Arkansas
The objectives of this project were to compare the relevant demand forecasting techniques with the forecasting technique employed by the US Navy, attempt to investigate the effect of temporal aggregation on the forecast error, develop an intermittent demand generator based on supplied Navy data whose intermittency can be controlled by the user-specified parameters, and verify and validate the intermittent demand generator as well as the Java implementation of the forecasting techniques.
The Naval Aviation Maintenance Program divides the Navy’s supply chain into three levels: O-Level or organizational level, I-Level or intermediate level, and D-Level or depot level. The O-Level activities are preformed on site and involve line item repair/replacement of modular components. Parts that have failed at the O-Level may be passed up to the I-Level for repair and testing. I-Level maintenance occurs at Aircraft Intermediate Maintenance Departments (AIMD) on shore or at times afloat on aircraft carriers. Naval Aviation Depots (NADEPs) perform maintenance on equipment that cannot be repaired at the lower levels especially the overhaul and refurbishing of end-items and assemblies. After flying a sortie, aircraft are inspected and failed parts or parts requiring scheduled maintenance are removed for repair or maintenance. If a spare is available, then the spare is placed on the aircraft. If a spare is not available, then an order for a replacement is made. The original part is examined to determine whether it is repairable. If so, the repair may occur. If the item cannot be repaired at that level the part is passed up to the next level of repair. The repair cycle may include shipping time, processing time, repair time, waiting time, and delivery time. These times and the amount of spares available directly affect the availability of the weapon system to complete its mission. As indicated, the demand for parts depends upon the flying schedule, the severity of the failure, and the amount of inventory in the system. The primary demand on this system is the failure rate of parts. In conjunction with this, the attrition rate (those parts that cannot be repaired) is also a critical component to the planning of sparing levels over time. Because of the nature of the repair/fail cycle the demand for repair and spare parts is often intermittent in nature.
Intermittent demand is characterized by demand data that has many time periods with zeros and small demands in other periods. Additionally, this type of time series will have intermittent spikes of demand. Intermittent demand is very common in spare parts systems where we have a low usage rate of items, long life-cycles, large number of stock keeping units, and many stocking sources and locations. A variety of methods have been developed for forecasting intermittent demand most notably Croston’s method and its variants. Recently, new methods have appeared for forecasting intermittent demand based on bootstrapping. See for example, the work by Smart and Willemain (www.smartcorp.com/News/wp0212_pfv.stm). The basic premise of these methods is to resample the historical data and directly develop the lead time demand distribution.
Recently the Navy, as well as other branches of the military, has made progress integrating commercial practices within the area of enterprise requirements planning (ERP) systems to improve the information systems that serve as the backbone for military inventory and logistics planning. The Navy’s SMART program exemplifies progress made in this area. The basic underlying premise behind the use of ERP packages for inventory planning involves the forecasting of demand requirements over time. Unfortunately, within systems that involve spare parts, accurate demand forecasting is extremely difficult due to the intermittent nature of the demand. Forecasting errors can have a serious effect on the planning process and negate any potential improvement in the control of the inventory system. In fact, the control of spare parts systems via a time-phased planning approach predicated on distribution requirements planning (DRP) logic has not been fully proven. One of the keys to implementing a distribution requirements planning system based on forecasted demand will be the accuracy of the forecast used in the planning process. The purpose of this research effort will be to examine methods by which intermittent demand may be forecasted a
Created a demand categorization scheme out of an extensive range of demand scenario and identified forecasting technique most appropriate to it;
Intermittent demand generator – correlated or non correlated demands, intermittency, lumpiness