报告人简介：Professor John Ekaterinaris, a renowned scholar of aerospace engineering and current faculty member at Embry-Riddle Aeronautical University (ERAU). He earned his B.S. from Aristotle University of Thessaloniki in 1977, and his M.Sc. and Ph.D. from GaTech. His career includes positions at NASA-Ames Research Center, Naval Postgraduate School, RISOE DTU in Denmark, Nielsen Engineering and Research, and leadership as Research Director at FORTH. His research interests are computational mechanics (including aerodynamics, magnetogasdynamics, flow control, flow transition, turbulence research, and flow structure interaction), multi-scale phenomena, stochastic PDE's, and biomechanics. He is author of over 80 journal papers and many conference publications. The impact of his work in the fields he worked can be found in a study that has been conducted recently https://elsevier.digitalcommonsdata.com/datasets/btchxktzyw/4. In this study, an international list was made of the world's top 2% of scientists recognized for their career scholarly impact (excluding self-citations). There are 1,065 names in the top 2% Aerospace Engineering discipline, and Dr. Ekaterinaris was ranked number 200 in this 2021 yearly impact list. He is associate editor for the Journal Progress in Aerospace Sciences and editor-in-chief for the Journal Aerospace Science and Technology.

内容摘要：The first talk- Construction of mode based reduced order models (ROM) for moving bodies

This study focuses on creating reduced order models (ROMs) for predicting loads on moving bodies like wings and projectiles, and store separation in aerospace applications. To enhance efficiency in separation predictions, the construction of ROMs enables quick and precise modeling. Various techniques, including proper orthogonal decomposition (POD), dynamic mode decomposition (DMD), and convolutional neural network (CNN), were compared for reconstructing surface pressures. The study found POD most efficient for pitch-up store motion. Further, two interpolation schemes were compared for oscillating store cases. A surrogate ROM-based model was developed using 12 combinations of reduced frequencies and amplitudes, demonstrating accurate and swift load predictions. Additionally, the ROM-based model was able to predict store trajectories at intermediate Mach numbers, using data from Mach 0.5, 0.8, and 1.2. The results showed close agreement between the computed CFD and ROM surface loads and trajectory predictions.

The second talk -Machine learning-based surrogate modeling approaches for fixed-wing and rotary wing store separation

This study explores two data-driven surrogate modeling methods, proper orthogonal decomposition (POD) and convolutional neural network (CNN), for predicting surface pressure and shear stress in store separation under supersonic conditions. Utilizing computational fluid dynamics (CFD) at Mach numbers 1.2, 1.4, and 1.6 for fixed-wing separation, the pressure and stress data were employed to create the surrogate models. These models allowed for the prediction of store load distributions and corresponding trajectories at intermediate Mach numbers 1.3 and 1.5. Both CNN and POD-based models yielded accurate load and trajectory predictions with substantially reduced computational costs. The same methodology was effectively applied to the more complex case of helicopter store separation, achieving significant time reduction. The resulting reduced-order models are applicable for multiparameter trajectory optimization studies, especially in helicopter hover scenarios.