Improved Inventory Control by Better Demand Modeling
The maintenance of capital goods is a major industry in itself (Aberdeen Group, 2005). Inventory
control of spare parts, supporting the maintenance of capital goods, is essential to many
companies, from capital-intensive manufacturers to service organizations, such as medical
equipment producers, car manufacturers, train companies and airlines. Spare parts play an
important role in both preventive and corrective maintenance. In preventive maintenance, spare
parts are needed according to the condition of the components. In corrective maintenance, failed
part is replaced by a new one in order to avoid long and costly downtime. Failed part is either
repaired locally or returned to a repair shop. The demand for spare parts is not known in advance
in the case of corrective maintenance while the demand for spare parts could be known to some
extent in the case of preventive maintenance. The need for supporting maintenance operations
and protecting against equipment failures leads to the inventory of spare parts.
However, spare parts management is difficult because of the high associated holding and
shortage cost. Spare parts demand is lumpy due to their specific nature. Moreover, spare parts
may run high obsolescence risk as they perform specific functions. In fact, typical industrial data
sets have limited demand history with a large portion of zero demand and a few positive ones.
Lead time is another factor which should be considered. Comparing to transportation lead time,
manufacturing lead time can be long (even up to two years) as new molds or special set-ups
might be required by the manufacturing of a spare part to start production. As a result, companies
or organizations face the challenges in striking the right balance between inventory holding,
shortage and obsolescence cost while offering competitive service levels.
Many could be solved if demand of spare parts could be better forecasted. Delphi study with
senior service part managers shows that demand forecasting is the key challenge in service parts
management (Boone et al. 2008). Better forecasting technique could reduce safety stocks and
further reduce cost without reducing service level. The forecasting problem can be solved from
two aspects, pure time series methods and methods incorporating advanced information such as
monitoring information obtained by electronic sensor and communication equipment, system
usage information, and maintenance information (Dekker et al. 2013). Pure time series methods
is easy to implement. They are practical and widely used. (Croston, 1972; Syntetos and Boylan,
2005; Teunter et al. 2011; Willemain et al. 2004; Porras and Dekker, 2008; Wingerden et al.
In order to improve forecasting accuracy and reliability, different kinds of available information
are considered in spare parts demand forecasting, especially for the forecasting of critical
components. Using installed base information, we can take product lifecycle pattern and demand
generated from specific context into consideration. Incorporating condition monitoring system, it
is possible to obtain information about the quantity and time point of the demand for spare parts
in advance and yield substantial savings (Deshpande et al. 2006; Louit et al. 2011; Li and Ryan
2011; Lin et al. 2013; Topan et al. 2013). Applying information on planned maintenance and
repair operations, the forecasting errors can be reduced significantly (Romeijnders et al. 2012).
In project 1, we delve into the analysis of historical data and propose an empirical-EVT demand
forecasting method based on the empirical method proposed by Porras and Dekker (2008). We
aim to build the lead time demand (LTD) distribution by constructing a histogram of demand
over the lead time without sampling and apply extreme value theory (EVT) to model the tail
behavior of LTD distribution. This method not only captures autocorrelations and fixed demand
intervals due to preventive maintenance like what the empirical method achieves, but also can
reach high service levels while the empirical method fails to.
In project 2, we will incorporate the realized and planned maintenance tasks for each type of
component into the forecasting problem. Historical demand data of spare parts for each
component is assumed to be available. We also have the realized number of repairs for each
component and the planned number of repairs in planning horizon. We aim to build an analytical
model which incorporates demand data and maintenance tasks so that we can forecast the
number of spare parts that may needed in future repairs.
In project 3, we work on the management problem of coordinating preventive maintenance and
inventory policy by combined ordering. We consider a system with N identical components
subject to increasing failure rate. Both regular lead time and emergency lead time are considered.
We aim to develop an analytical model for such a system.
- service logistics; pooling; inventory control
- Time frame
- 2013 -
Work in Progress
S. Zhu, R. Dekker, W.L. van Jaarsveld, R. Wang & A.J. Koning (2015). An Improved Method for Forecasting Spare Parts Demand using Extreme Value Theory. (Preprints). :
S. Zhu, R. Dekker, W.L. van Jaarsveld, R. Wang & A.J. Koning (2017). An Improved method for Forecasting Spare Parts Demand using Extreme Value Theory. European Journal of Operational Research, 261 (1), 169-181.
Office: Tinbergen Building H07-30
Burgemeester Oudlaan 50
3062 PA Rotterdam
3000 DR Rotterdam