Fluid flows containing complex particles that interact with each other and with vessel walls are a central feature of an enormous range of biological, chemical, and physical processes, and the potential scientific and technological impact of having access to predictive computer models is difficult to overstate. Consequently, improvements in computer simulations for aggregating particulate flows have been actively sought for many years, and to date have been driven largely by increased availability of computer power coupled with advances in mathematical algorithms and techniques. As this trend continues, computational modeling is increasingly blessed (and cursed) by the "big data" streams generated by high resolution experimental measurements and/or by detailed computational simulations. In particular, the meaningful comparison of computational outputs and experimental measurements, both of which are large, complex, and statistically noisy, has emerged as a key challenge. As a result, models often capture many qualitative phenomena correctly but their predictive ability, and hence their usefulness to industry and manufacturing, becomes increasingly hard to establish and exploit. The proposed work seeks to close this gap by implementing, extending and exploiting a broad (and evolving) set of novel data mining techniques that enable new ways of linking tailored experiments to smartly designed simulations and back to model building.
A multifaceted approach will be pursued to interrogate and use data jointly from a multiscale/multi-element model and two particulate-flow experimental systems. The experimental systems include a "target" system (platelets in blood), whose predictive description is ultimately sought, and a "model" system (DNA-functionalized colloids in water), which will be used to develop methods and help interpret the more complicated target. Both systems are defined by "complex" particles that exhibit time-dependent adhesivity leading to transiently evolving aggregates at a specified location on the vessel surface. Modern data mining techniques will be exploited and extended to process the native, high-dimensional data generated by these three sources to discover low-dimensional statistical measures that enable meaningful merging/comparisons of data streams from different sources and runs. Ultimately, the project deliverables are (i) a better understanding of the physical, chemical and biological mechanisms operating in these complex systems, (ii) data-enhanced and data-validated engineering models, and (iii) experimental design rules for complex, multi-parameter systems.
