On the construction of a mode based reduced order model for a moving store

Abstract

Store separation from aircraft and spacecraft has been a critical issue for the aerospace industry. A potential remedy to reduce resources and to improve turn around time in store separation predictions is the construction of a reduced order model (ROM) that can be utilized for fast and accurate modeling. The objective of this study is to investigate how to construct ROMs for moving bodies and utilize them for both load and subsequently for trajectory predictions of a separating store. To analyze the ROM's capability for load prediction, a preliminary study for modeling distributed pressure loads of pitching and oscillating stores was undertaken. For comparison between modal decomposition and neural networks, the study first used proper orthogonal decomposition (POD), dynamic mode decomposition (DMD), and convolutional neural network (CNN) to reconstruct surface pressures and free stream flow for a pitch up store. Results indicated for pitch up store motion POD would perform most efficiently. The study then used the oscillating store cases to compare two interpolation schemes; namely surface splines and Kriging interpolation. For the oscillating store case, a surrogate ROM-based model was constructed using CFD-simulation data for 12 combinations of reduced frequencies and amplitudes. The resulting ROM-based surrogate model was able to make accurate and fast load predictions for intermediate reduced frequencies and amplitudes. In an attempt to investigate ROM ability to predict store trajectories of intermediate conditions, three cases of store separation at Mach 0.5, 0.8, and 1.2 were considered. It was shown that predictions of trajectories at intermediate Mach numbers can be accomplished using the ROM-based surrogate model constructed from computed store separations at Mach 0.5, 0.8, and 1.2. Results showed a close agreement between CFD and ROM surface loads and as a result trajectory predictions.