Modeling unit replacement associated with preventive maintenance

Sammanfattning: The work presented in this paper is a master thesis performed at the division of Optimization and Systems Theory at the Royal Institute of Technology in Stockholm. The problem is formulated by Systecon AB, a consultancy firm focusing on life support cost analysis and optimization. In the software that Systecon uses for optimization (OPUS10) there are certain simplistic assumptions. Namely, that preventive maintenance is always fully regulated and in the case where several bases are requesting spare units for preventive maintenance from the same depot there is full coordination between bases, thus assuming full regularity also at the depot. These assumptions work sufficiently at base level but not necessarily in a support system with multiple levels where spare parts for preventive maintenance is requested from a central depot. To investigate the effects of these assumptions a user survey has been conducted and based on the result a simulation tool has been constructed in MATLAB. The results suggest that due to the simplistic assumptions made in OPUS10 there is a risk of underestimating the number of systems demanding spare parts for preventive maintenance, thereby affecting the accuracy of OPUS10. Furthermore, to investigate the possibility of refining the model two approaches have been made. Firstly, the built in OPUS10 factor PMCF has been analyzed on how it may be utilized to compensate for underestimating demand from preventive maintenance. Secondly, the Binomial distribution has been suggested and discussed as a potential replacement for the mathematical modeling of preventive maintenance. A combination of the two distributions has also been included as a final investigation. In conclusion, PMCF may be used quite effectively, but setting the value for PMCF is complicated and no distinct guidelines can , from the conclusions of this thesis, be presented on how to determine a value for PMCF. Even though there are possibilities of constructing such guidelines it is not within the scope of this thesis. The Binomial distribution turns out to be very accurate in modeling of preventive maintenance. But it does require knowledge of the variance from preventive maintenance, something that is not required by the current models in OPUS10. However, the last approach, assuming Bernoulli distribution at base level and Binomial distribution at central level of the support system, gives accurate results while knowledge of the variance is not required.