Description
Predicting the failure response of complex systems often requires computational models that can capture the nonlinear response of the material and structure across multiple scales. Typically, the output response is a direct result of the complex interactions of different phenomena at different scales of the hierarchical system. Therefore, computed model errors correspond to accumulated model errors that have been propagated across several levels of the system. The objective of the current work is to identify and quantify the errors introduced by computer analytical models at different scales in the ballistic impact response simulation of a composite laminate. To that end, a Bayesian network based framework was implemented to systematically estimate the model contribution of uncertainty to the response prediction at each sub-scale of the composite problem. The developed method can be used for optimal allocation of validation resources by determining the type and number of experimental tests needed to reduce uncertainty at different subsystems levels of large engineering systems.
