Print Email Facebook Twitter Database driven forecasting of spare parts demand at the Royal Netherlands Airforce Title Database driven forecasting of spare parts demand at the Royal Netherlands Airforce Author Schraven, M. Contributor Verhagen, W.J.C. (mentor) Faculty Aerospace Engineering Department Control & Operations Programme Air Transport & Operations Date 2015-02-17 Abstract The Royal Netherlands air force (RNLAF) has been coping with a low availability of F-16 aircraft. It has been acknowledged that increasing the availability of avionics components is most beneficial to increasing fleet availability. Especially for F-16, rich database information is available including past operations and maintenance dating back to 1996. Improving the methods which are used to forecast the demand of avionics components is an important precondition to decrease multiple lead times in the F-16 supply chain. The readily available database information should be used to this goal. The research objective of this research project is hence as follows: Develop a novel spare parts demand forecasting method in order to increase F-16 avionics component availability by (1) evaluating state-of-the-art spare parts demand forecasting methods with the RNLAF time series demand data and (2) applying RNLAF data of component technical characteristics and component operational exposure in conjunction with the original historical demand data to improve selected method(s) Academic literature clearly stipulates that the methods of Croston, Syntetos, Babai and Teunter should be used to forecast intermittent demand. However, no literature has been found on how to improve these demand history based methods by using other data inputs like maintenance and operational history. Also, only limited literature was found about the extraction of demand drivers from databases. In short: A research opportunity is present. The developed conceptual model consists of five parts: Input (of demand, maintenance and operations data), steering variable (SV) generator, forecast method applier, stock level simulator and error minimizer. It is expected that stock level performance for avionics parts can be improved by (1) selecting the optimal forecast method, (2) selecting the optimal smoothing constants, (3) applying an SV deducted from installed base and/or FHRS per month variables. First, an optimal method and settings are established “in sample”. This ‘best fit’ is subsequently tested “out of sample” and compared to RNLAF stock levels to validate its performance. The installed base and FHRS per month variables are computed by combining two datasets and establishing install-removal intervals for a set of 110 unique components. The variables are translated into normalized and smoothed variables which are used to momentarily damp or gain the next estimate of the selected forecast method. Both installed base and FHRS per month variables are related to the utilization of components. An increase/decrease of the installed base (the total amount of installed components of a specific type at a specific time) or FHRS in a specific month for a component is expected to result in more/less defects of and hence demand for that component. A forecast optimization tool is built to implement the conceptual model. Enabling ease of experimentation by including many adjustable parameters and visualizing outputs to gain insights was of high priority throughout the development. The tool is capable of optimizing forecast methods and settings for all component demand inputs (110) in batch mode. The main output of the tool is exported in spreadsheet format. Analysis of the output of the forecast optimization tool resulted in the general conclusion that forecasting performance of the included forecast methods can be increased by using maintenance and operational data of the components, but the overall performance gain is small. In specific: SV information is used to generate better “in sample” forecast fits for approximately 50% of the components demand forecasts for which installed base information was available. In those cases, the mean performance gain in relation to the conventional forecast mode is 17.7% for installed base SV mode and 12.4% for FHRS SV mode. For aperiod of six months out of sample, a percentage of 50.0% of the optimized installed base SV forecasts and 47.9% of the optimized FHRS SV forecasts still perform best and are therefore robust for the demand pattern of the specific components. The out of sample robust SV powered forecasts lead to performance gain for 20.9% of the components. The mean performance gains are 47.7% and 21.3% for installed base SV and FHRS SV modes respectively. The research project succeeded in the development of a model and method for forecast optimization using multiple sources of information in parallel to demand history. The research project addressed two novel aspects not yet covered in academic literature: (1) Batch optimization of the choice for a forecasting method and parameter settings in relation to the specific demand patterns. (2) Operationalize multiple predictive variables from maintenance and operations databases and integrate them in the forecasting method. The provided platform is not perfect but will confidently serve as a starting point for further innovative research. Subject forecastingspare partsdata miningdemand drivers To reference this document use: http://resolver.tudelft.nl/uuid:1fc3ce20-29c4-4895-bf7d-fd66a19b25e1 Part of collection Student theses Document type master thesis Rights (c) 2015 Schraven, M. Files PDF 2015-01-31_Schraven_MSc_t ... asting.pdf 5.97 MB Close viewer /islandora/object/uuid:1fc3ce20-29c4-4895-bf7d-fd66a19b25e1/datastream/OBJ/view