UCSC-SOE-12-16: Predicting Variable-Length Functional Outputs for Emulation of a Flight Simulator

Yuning He, Herbert K. H. Lee, and Misty Davies
09/20/2012 07:45 PM
Applied Mathematics & Statistics
Adaptive flight control systems using online-learning neural networks have great promise for improving the safety of aircraft. However, these systems are difficult or perhaps even impossible to certify using current methodology, yet it is required that the systems be demonstrated to be safe. A recent approach considers the use of statistical emulation to help understand the behavior of the system. We present a statistical framework to model and predict the output of a function of multiple real variables in which the output is itself a function of a real variable using statistical emulation. This approach has the bene fit that it carries statistical uncertainties with its predictions. Through our model, we can assist NASA with development of their flight control simulator.