| Rocket vehicles have traditionally been designed in an iterative fashion, beginning with system requirements before proceeding sequentially through requisite analytical disciplines until resources are exhausted. A sequentially designed system, while adequate, is not an optimum due to the approximations and loss of fidelity inherent in separating analytical disciplines which are, in fact, coupled. Recently, increased computational power and advances in algorithms have allowed multidisciplinary optimization (MDO) to emerge as a system-level design tool accessible to industry. To date, MDO has primarily been applied to some facets of aircraft systems and, to a lesser extent, rocket vehicles in literature but has not yet met with widespread industry use. To this end, four obstacles have been identified: (1) MDO efforts to date have focused on system-level parameters rather than physical dimensions and hence have not yielded a preliminary design which includes manufacturing, cost, and other constraints, (2) Prohibitive computational performance requirements associated with high-fidelity analyses such as computational fluid mechanics (CFD) and finite element analysis (FEA), (3) Lack of an integrated design environment which incorporates computational tools already widely used in industry while remaining accessible to individual users without high-level expertise in the individual tools, and (4) The widely-varying and tightly-coupled environments to which rocket vehicles are typically exposed, including analyses not required for aircraft applications.;Here, an MDO method for rocket systems has been formulated which simultaneously overcomes the challenges listed above. First, a response surface-based approach to modeling computationally expensive analyses with arbitrary dimensionality and general constraints was developed. This method focused on an evenly-distributed representation of the entire feasible region at any fidelity level, including combinations of discrete and continuous variables. The analytical disciplines required in the design of a general rocket vehicle were then developed, focusing on computational cost and multi-fidelity methods were applicable. Finally, this integrated framework was applied to three diverse case studies. Where possible, the results obtained were compared to traditional design methods demonstrating considerable performance gains while maintaining manageable computational cost. Within this framework, many opportunities for improvement and future directions were noted, both in the analytical disciplines and optimization architecture as a whole. |
