A rapid multidisciplinary uncertainty analysis method for loitering munitions based on Bayesian KAN

Low cost and miniaturization represent critical developmental directions for loitering munitions, thereby imposing greater demands on integrated design methodologies. Under cost limitations, the restricted availability of experimental and high-fidelity simulation data renders exclusive reliance on analytical models impractical for system-level design. Furthermore, the multidisciplinary nature of loitering munition design introduces a wide range of uncertainty factors, necessitating the implementation of Multidisciplinary Uncertainty Analysis (MUA) to accurately quantify their influence on system performance. This work introduces a closed-form MUA strategy built upon Bayesian KAN (Kolmogorov–Arnold Networks) to tackle the above challenges. A statistical inference framework combining maximum likelihood estimation with goodness-of-fit testing is developed to determine the distribution characteristics of uncertain variables. Bayesian KAN is employed to quantify epistemic uncertainty in modeling, while the first-order approximation of the second-moment technique is used to analytically derive the expected values and standard deviations of system responses. The methodology is applied to the MUA of a loitering munition, and results indicate that it effectively captures and analyzes uncertainty in coupled multidisciplinary systems, while substantially reducing computational demands in comparison to conventional Monte Carlo simulation techniques.