Andrew Meade
Department of Mechanical Engineering and Materials Science,
Rice University
Thursday, January 12, 2006
1131 Kemper Hall
3 :10-4:00 p.m.
A Method of Weighted Residuals Approach to Machine Learning With Aerospace Applications
The successful analysis of physical system rests upon the proper balance of theory, physical experimentation, and numerical modeling. This is even more critical in the design and analysis of future aerospace vehicles. As it presently stands, computational fluid dynamics (CFD) with its problems of gridding and flow modeling has limited utility. For example, CFD in aircraft design is used only in a very small region of the flight envelope. Also, it is common knowledge to the experimentalist that one can never really get enough data, but the time an cost of designing and running experiments is only increasing. This, combined with the mothballing of windtunnel facilities and reduction of flight tests due to today's budget constraints, will not support business as usual.
The objective of this work has been to develop tools to help fuse all information available in the emulation of aerospace systems. The approach chosen is a Method of Weighted Residuals interpretation of biologically inspired machine learning and includes such tools as artificial neural networks, Tikhonov regularization, and meshless finite elements. Results will be presented on efforts to improve windtunnel and flight testing of high performance fixed-wing aircraft and rotorcraft.