This thesis initially presents the work carried out on the research hypothesis –agent-based simulation is better than traditional discrete-event modelling. To test this assertion, a comparison of these two modelling approaches is made by way of a case study. The scenario, a global repair operation of a fleet of civil jet engines, is a real lifecycle costing example which involves logistics and is typical of problems commonly modelled using either of these paradigms.To carry out the comparison, the method involved building a discrete-event model which matched the functions of an existing agent-based model as closely aspossible. Rigorous control was applied during its implementation phase by way of formal code walkthroughs and model dynamic testing. Among the internal metrics, lines of code provided an estimate for model size while the McCabe CyclomaticNumber measured structural complexity. The external software quality of maintainability was derived from these metrics and estimated by modelling experts through Delphi sessions. The dynamic performance of each model was determinedby the execution times of successfully completed simulation runs over a range of engine fleet sizes.This research went on to develop a hybrid approach (which is currently the subject of a Rolls-Royce patent application) which draws on the strengths of both agent and discrete-event paradigms. In order to combine agent roles and discrete event processes, a new model was implemented using a three-layered architecture. A full fleet simulation was developed using this hybrid approach. Although the code size is slightly larger and run times slightly longer than the conventional model, the thesis argues that, crucially, it is more maintainable as it reduces the conceptual gap between problem and model.
展开▼
