Dynamic Discrepancy
One of the key issues in scale-up is the design of bench-scale experiments for the calibration of chemical models. The models inevitably contain empirical terms whose meaning is mathematical as opposed to scientific.
The danger of extrapolation: models (green) fit to an experiment (black) may diverge
significantly from reality when upscaled.
Such quasi-empirical models work best when they are not required to extrapolate. Our focus in this work is to prevent extrapolation by using the process model as a tool in the design of bench-scale experiments.
The dynamic discrepancy method starts with a Bayesian approach for calibration of the model: this means choosing many plausible fits to bench-scale data, instead of just one. These multiple chemical models are then incorporated into a process model, the results of which are used to generate more experiments. The model is recalibrated to the new data and the resulting models incorporated at the process scale again; the new results will show less uncertainty than before. This iterative procedure is effectively moving the bench scale experiments toward those conditions that the process models predicts will be found at the process scale.
Hypothetical model results (temperature, partial CO
2
pressure, capture rate) for the first and second iteration (1st and 2nd row, respectively)
of upscaling in the context of a bubbling fluidized bed. Compared to the first
iteration, the second is well converged.