OPM/IS 910 Prof. Dr. Rommert Dekker, Erasmus University Rotterdam, Niederlande

Rommert Dekker


Im Rahmen des CDSB Research Seminars freuen wir uns Herrn Prof. Dr. Rommert Dekker von der Erasmus University Rotterdam, Niederlande digital bei uns begrüßen zu dürfen.


Am Mittwoch, den 16.11.2022 von 12:30 bis 13:30 Uhr hat Herr Professor Fleischmann ihn eingeladen einen Vortrag zu halten.


Untenstehend finden Sie die Zugangsdaten zum Portal2 in dem Sie den Zoom Raum Link abrufen können



Spare parts demand forecasting and inventory control  


Spare parts demand is characterized by intermittency and erratic. Intermittency implies that non-zero demand occurs scarcely, erratic implies that non-zero demand can vary a lot. Accordingly, standard forecasting procedures, relying on mean, trend and seasonality does not work. Instead special methods were invented. The first method was developed by Croston. Later on, extensions were formulated by Syntetos and Boylan, as well as by Teunter, Syntetos and Babai. All these approaches forecast the mean demand and for spare parts forecasting a distribution needs to be assumed. Non-parametric methods were formulated by Willemain, using a Markov chain to predict a demand and by Porras, by sampling lead time demand occurrences. Recently machine learning methods were formulated by Kourentzes.
Forecasting methods are typically assessed by their Mean Square Error (MSE), but for many items it is better to normalize errors according to mean demand. In inventory control, with base stock or (R,Q) policies, the mean demand is not that important, yet it is the demand tail which determines a service level and this constitutes another performance criterion for forecasting methods.

In this presentation we first give an overview of these methods and then we present the results of testing these methods both on 4 industrial data sets and on own created datasets with special characteristics. The results are varied: methods which perform good on MSE, do not perform best in inventory control! Finally, we present two methods using indirect forecasting: one using so-called installed-base information and the other one using maintenance planning info.