AI + UQ + Physics

Multiphase Fluidized Bed Problem with Tensorflow + Dakota + CFD (MFiX): We carried out a nondeterministic analysis of flow in a fluidized bed. The flow in the fluidized bed is simulated with the open source software: MFiX (National Energy Technology Laboratory), Dakota (Sandia National Labs), & TensorFlow. MFiX, a legacy software mostly written in Fortran, provides multiphase fluid dynamics suite capable of running massively parallel high performance computing (HPC) systems.

It does not possess built-in tools for uncertainty quantification, so we developed a C++wrapper to integrate with Dakota, an uncertainty quantification toolkit developed at Sandia National Laboratories. The wrapper exchanges uncertain input parameters and key output parameters among Dakota and MFiX.

A data-driven & machine learning (ML) framework is also developed to obtain reliable statistics as it is not feasible to get them with Dakota-MFiX. The data generated from Dakota-MFiX simulations, with the Latin Hypercube (LHC) method of sampling size 500, is used to train a machine learning algorithm. The trained and tested deep neural network algorithm is integrated with Dakota via the wrapper to obtain low order statistics of the bed height and pressure drop across the bed. In addition, it can also be used to obtain statistics with various distributions of the uncertain input variables and various uncertainty quantification methods. We carried out sensitivity analysis on 9 input with 2 output parameters.

Schematic of a loosely-coupled or “black-box” interface between Dakota and MFiX – a multiphase CFD solver

Left: Pressure for flow in a fluidized bed with Lagrangian-Eulerian (DEM) approach & uncertain input parameters at two different times.

Right (top): Time histories of instantaneous and averaged bed height and pressure drop across the bed, and (bottom) average bed height comparison DNN-vs-actual or computed values. The network trained with 80% of the data.