Reduced-order model for efficient generation of a subsonic missile’s aerodynamic database

Abstract

AbstractThis article reports on the development of a fast method for generating the aerodynamic database for subsonic flow over a missile. At present, this is typically achieved using RANS-based CFD, which is expensive for the complex missile geometries and the multiplicity of operating conditions to be evaluated. The presented reduced-order model (ROM) provides a reasonably accurate prediction of the aerodynamic coefficients of the missile (and, in fact, the full flow field around it) within half a minute. In particular, in the interpolative regime, prediction errors for all coefficients are typically less than 1% of their respective maximum values encountered in the database, with extrapolation incurring more error. The empirical approach ‘learns’ from the CFD solutions calculated for a few operating conditions, and then is able to make predictions for any other condition within a feasible set. The learning employs proper orthogonal decomposition (POD) to characterise the most important features of the flow. The prediction is posed as an optimisation problem that aims to find the flow solution as a linear combination of the above POD modes that minimises the residual of the governing equations. Innovations on the prevailing POD-ROM approach include novel implementation of boundary conditions, simplified computation of the aerodynamic coefficients, and a procedure for choosing modelling parameters based on extensive cross-validation. Challenges overcome in application to a problem of industrial relevance are discussed.