Military systems typically operate within closed workforce structures and rely on long-term fleet planning, creating complex challenges in joint workforce planning and fleet management. These challenges are further compounded by deep uncertainty and competing objectives. Simulation models provide a powerful means of capturing such complexity and the interdependencies among system components and resources, and simulation--optimisation techniques have therefore been widely used to support strategic decision-making. However, many existing simulation--optimisation studies rely on best-estimate assumptions and do not adequately account for deep uncertainty. To address this limitation, this study explicitly considers deep uncertainty and derives robust strategies that perform satisfactorily across a wide range of plausible future conditions. Specifically, we analyse and compare two robustness frameworks developed for Decision-Making under Deep Uncertainty (DMDU): Multi-Objective Robust Decision-Making (MORDM) and Multi-Objective Robust Optimisation (MORO). The proposed approach integrates a simulation model with Exploratory Modeling and Analysis (EMA) to identify robust strategies governing recruitment, promotion, and fleet transition. Robustness is assessed using metrics derived from key objective functions, including workforce surplus cost and operational fleet availability. The methodology is demonstrated through a realistic case study motivated by the Royal Australian Navy’s fleet modernisation program, which poses a joint workforce planning and fleet management problem under deep uncertainty. The results indicate that strategies identified using the MORO framework exhibit greater robustness across plausible future conditions than those obtained using MORDM, albeit at approximately an order-of-magnitude higher computational cost. Overall, the findings provide military decision-makers with insights into robustness–cost trade-offs and guidance on selecting an appropriate robustness framework for complex defence planning problems.
Turan et al. (Mon,) studied this question.