| The primary goal of this research was to develop appropriate models for preventive maintenance and opportunistic maintenance planning for a multiunit system under transient state with economic dependency among units. To this end, a methodology was developed that integrated renewal process concepts, numerical analysis, binary integer programming and linear programming techniques, and simulation to provide a unified approach to maintenance planning for transient multiunit systems.; A stochastic model was first developed based on renewal process concepts to study the age replacement policy for individual units under transient response, i.e., with finite planning horizon. The resulting recursive integral equation with delay loss function was then solved numerically to determine the optimal cost-based intervention intervals. The resulting loss function exhibited a shape similar to the reflected lognormal distribution function.; A binary integer programming model was then developed to rearrange the optimal preventive maintenance times for a combination of two or more units to take advantage of opportunistic maintenance. The binary integer programming model for the opportunistic maintenance arrangement possessed a special property of total unimodularity and was solved by the simplex method to determine the combination of components that could be maintained together based on the idea of opportunistic maintenance.; A real life system--a Hot Roll Line was used to demonstrate the application of the methodology developed in this research.; Simulation based on ARENA{dollar}sp{lcub}copyright{rcub}{dollar} was then implemented to verify the analytical approach. The simulation models were developed with the same system configuration and assumptions as the analytical models. The simulation results showed remarkable similarity to the analytical results.; Rather than model the maintenance problem in one step, which so far has proven to be extremely difficult and often impossible in the archival literature, the research partitioned the modeling into several phases. This partitioning approach removed some of the modeling difficulties, reduced the computational burden, and rendered the problem a bit more manageable in today's computing environments. |
