Integrating Machine Learning with Parametric Models for Rapid Cost Trades in Early-Stage Missile Design

In classic missile design methodology the development of cost is assessed late using rough estimators based almost exclusively on the overall weight. This approach offers minimal leverage for shaping design decisions with consideration for cost-effectiveness, thereby restricting its uses in the development of low-cost missile systems. To compensate for this gap, this paper presents MCDAT (Missile Cost Data-Driven Analysis Tool), a hybrid cost model platform created to incorporate economic factors into early stage implementation considerations. MCDAT decomposes the missile into critical subsystems such as solid rocket motor, inertial measurement unit or structures and exploits machine learning models, empirical parametric formulas and historical cost analysis for estimating unit production and life-cycle costs. Besides the first-unit production cost, the framework includes learning curve analysis, RDT&E, research and development estimation, and 30 years of maintenance forecasting cost using Monte Carlo simulation. Validation against French VT-1 surface-to-air missile cost showed 5.8% prediction errors and the subsystem allocation was consistent with tactical missile industry benchmarks. By design with cost being implemented as an active driver in the design loop versus a post-hoc constraint, MCDAT enables identifying and prioritizing affordable architectures at concept time, directing design space exploration, providing quantitative economic feedback when the engineering freedom of action is high.