Caglar Oskay delivered an invited talk at Johns Hopkins University – 02/06/2020
Lecture Title: Multiscale Modeling of Failure in Heterogeneous Materials and Structures:Roles of Reduced Order Modeling and Machine Learning
Location: Civil and Systems Engineering Department, Johns Hopkins University, Baltimore, MD.
Multiscale computational modeling has seen tremendous advances recently, and is now often the method of choice in academia for predicting mechanical and multifunctional response of many heterogeneous material/structural systems. However, many technical challenges remain especially in tackling large-sized engineering problems that involve failure and fracture under extreme loading and environments. Some of these technical challenges are well-known but remain unresolved: The online and offline cost of running multiscale simulations remains debilitatingly high; lack of existence of a well-defined representative volume concept reduces fidelity of the simulations, model predictions are often mesh sensitive; the model parameters at the fine scale are presumed known, but are often far from known, etc.
In this talk, we present new developments in multiscale modeling of failure response of heterogeneous materials and structures that address many of the aforementioned technical limitations, and offer pathways toward physics-based prediction of failure based on multiscale principles. We will discuss accelerating multiscale computations based on model order reduction methodologies that retain important physical features and mechanisms at the expense of some computational cost, as well as performing uncertainty quantification and sensitivity analyses using machine learning algorithms that are extremely fast yet are based on (mostly simulation) data. The attributes and capabilities of these computational developments will be demonstrated in the context of a variety of materials (particulate composites, laminated composites, polycrystals) subjected to a range of loading conditions (static, fatigue, high rate impact).