Transient CFD simulations of liquid mixing in stirred tanks are too expensive for systematic design studies, and their predictions of the mixing time deviate from experiments even when the turbulent scales are resolved. Recurrence CFD (rCFD) based on transport decouples the transport of a scalar from the solution of the flow. A short, precomputed flow database is replayed through maps of the displacement from cell to cell, so that corrections at the level of transport can be applied without repeating the solution of the momentum equations.
The approach is applied to the 95 % mixing time in a stirred tank reactor of 30 litres with two impellers, using a flow database of 4.5 seconds from a lattice Boltzmann large eddy simulation in M-Star as the only input. The first rCFD results inherit the overprediction of the input flow field, with 33.1 ± 1.4 seconds compared with 30.1 ± 3.0 seconds for M-Star, 28.1 ± 2.7 seconds for Ansys Fluent at comparable spatial resolution and 25.0 ± 3.0 seconds measured with conductivity probes in the experiment. Calibrating a single diffusion parameter for the exchange across cell faces against one probe measurement brings the prediction to 24.5 ± 1.0 seconds, which recovers the agreement with the experiment without repeating the flow simulation.
The rCFD simulation requires about 3 seconds of computing time per second of process time, which is 423 times faster than M-Star and 1200 times faster than Ansys Fluent, and it is almost insensitive to a fivefold increase of the extrapolation time step. Ninety injection locations varying in radius, height and angular position were therefore screened from the single database in 2.25 hours, compared with an estimated 40 and 113 days for the reference solvers. The results reveal that the mixing time varies by a factor of 3.6, from 13.6 to 48.8 seconds, dominated by the injection height and the angular position relative to the baffles.
The main objective of the work is a fast description of scalar transport that is resolved in space and time, while the 95 % mixing time serves as the quantitative metric for validating it against sparse experimental data.
The study is a collaboration between the Institute of Solids Process Engineering and Particle Technology and the Institute of Multiphase Flows at Hamburg University of Technology, and the Johannes Kepler University Linz.
Asif Shaik, Swantje Pietsch-Braune, Ryan Rautenbach, Christian Weiland, Michael Schlüter, Stefan Pirker, Stefan Heinrich (2026). Data-driven Transport Modeling for Tracer Injection and Mixing Time Screening in Stirred Tank Reactors. Chemical Engineering Science, 125476.
