Practical Challenges in Formulating Virtual Tests for Structural Composites

Abstract

Taking advantage of major recent advances in computational methods and the conceptual representation of failure mechanisms, the modeling community is building increasingly realistic models of damage evolution in structural composites. The goal of virtual tests appears to be reachable, in which most (but not all) real experimental tests can be replaced by high fidelity computer simulations. The payoff in reduced cycle time and costs for designing and certifying composite structures is very attractive; and the possibility also arises of considering material configurations that are too complex to certify by purely empirical methods. However, major challenges remain, the foremost being the formal linking of the many disciplines that must be involved in creating a functioning virtual test. Far more than being merely a computational simulation, a virtual test must be a system of hierarchical models, engineering tests, and specialized laboratory experiments, organized to address the assurance of fidelity by applications of information science, model-based statistical analysis, and decision theory. The virtual test must be structured so that it can function usefully at current levels of knowledge, while continually evolving as new theories and experimental methods enable more refined depictions of damage.

To achieve the first generation of a virtual test system, we must pay special attention to unresolved questions relating to the linking of theory and experiment: how can we assure that damage models address all important mechanisms, how can we calibrate the material properties embedded in the models, and what constitutes sufficient validation of model predictions? The virtual test definition must include real tests that are designed in such a way as to be rich in the information needed to inform models; and model-based analyses of the tests are required to mine the information. To date these compelling issues have been greatly underserved by both the modeling and experimental communities. Model-based analysis of tests has been undertaken only in terms of very simple (linear or continuum) engineering concepts; information-rich tests for more complex damage mechanisms have not been defined; and in fact the information in which experiments need to be rich has not been stated. Specific challenges in designing experiments for informing virtual tests and some promising experimental methods are summarized here.