Liquid–Liquid Dispersion Performance Prediction and Uncertainty Quantification Using Recurrent Neural Networks
Abstract
We demonstrate the application of a Recurrent Neural Network to perform multistep
and multivariate time-series performance predictions for stirred and static mixers
as exemplars of complex multiphase systems. We employ two Long-Short-Term Memory
(LSTM) frameworks in this study which are trained on high-fidelity, three-dimensional,
computational fluid dynamics simulations of the mixer performance, in the presence
and absence of surfactants, in terms of drop size distributions and interfacial areas
as a function of system parameters; these include physico-chemical properties, mixer
geometry, and operating conditions. Our results demonstrate that whilst it is possible to
train a LSTM with a single fully-connected layer more efficiently than a LSTM Encoderdecoder,
the latter is shown to be more capable of learning the dynamics underlying
dispersion metrics. Details of the methodology are presented, which include data
pre-processing, LSTM model exploration, methods for model performance visualisation;
an ensemble-based procedure is also introduced to provide a measure of the model
uncertainty. The workflow is designed to be generic and can be deployed to make
predictions in other industrial applications with similar time-series data.
Domains
Nonlinear Sciences [physics]| Origin | Files produced by the author(s) |
|---|