Bayesian prediction of the load response of composite structures to streamline certification approaches

Abstract

The ‘pyramid of tests’ or ‘building block’ approach, consisting of analysis and experimental validation, is currently used to demonstrate the structural integrity of composite aerostructures. The premise is that experiemntally obtained allowable strength limits of small elements (mainly coupons) of a composite structure are combined with knock-down factors to predict performance at higher length scales. The approach is suboptimal, as composite material behaviour is scale-dependent. Therefore, coupon testing does not adequately capture structural scale effects. In the best case, this leads to an overly conservative design, and in the worst to catastrophic failure. Composite structure certification programs of this kind are expensive, time consuming, and therefore inhibitive to innovation. A new methodology that defines the necessary tools for a holistic approach that streamlines and optimises the fusion of experimental and simulation data would therefore be desirable. Efficient iteration between physical and numerical tests is required to maximize experimental impact on modelling confidence and to minimize experimental effort, overall reducing cost and time of the certification effort. Therefore, a statistical modelling framework is proposed based on Gaussian processes (GPs) and Bayesian inference using experimental and simulation data to define new efficient ‘pyramid of tests’ approaches