Top Down Design Using Bayesian Network Classifiers for Composite Panels

Abstract

Optimization of composite panels has historically been approached using bottom-up design techniques. Such strategies require evaluating a large number of designs and then exploring these solutions to better understand performance impacts. This is often time consuming and is not computationally suitable for more difficult problems. Top-down approaches have been recently introduced as a way of quickly zeroing in on desired performance characteristics and only evaluating designs that exist within that region. In this presentation Bayesian Network Classifiers (BNC) will be used to enable this top-down strategy. We will evaluate the effectiveness of this approach on a problem that has two loading scenarios - Tensile in the 1 direction and then compressive buckling in the 2 direction. 4, 8 and 16 ply laminates will be evaluated to examine the benefits and tradeoffs as problem complexity increases. This top-down approach will be compared to the bottom-up approach to evaluate the effectiveness of the BNC in the design process. Finally we will examine simple problems with tow steering and how this new method can be implemented into this framework.