Comparison of Deterministic and Bayesian Calibration of MFiX-PIC, Part 1: Settling Bed

TL;DR

This paper compares deterministic and Bayesian calibration methods for MFiX-PIC modeling of dense granular flows, demonstrating that Bayesian calibration significantly improves prediction accuracy in settling bed simulations.

Contribution

It introduces a systematic Bayesian calibration approach for MFiX-PIC parameters and compares its effectiveness against traditional deterministic calibration.

Findings

Bayesian calibration improves prediction accuracy up to 6.5 times.

Five model parameters were systematically calibrated for settling bed flows.

Bayesian approach offers better sensitivity analysis and parameter optimization.

Abstract

Particle-in-Cell (PIC) approach for modeling dense granular flows has gained popularity in recent years due to its time to solution efficiency. The methodology is useful for modeling large-scale systems with a relatively lower computational cost. However, the method requires the definition of several empirical parameters whose effects are not well understood. A systematic approach to identify sensitivities and optimal settings of these parameters is required. Already, it is known that the choice of these parameters depends on a problem's flow regime. For instance, parameter values would be chosen differently for a settling bed or a fluidized bed. In this study, five different PIC model parameters were selected for calibration when applied to the case of particles settling in a dense medium. PIC implementation from the open-source software MFiX (MFiX-PIC) was used. This study extends the earlier work to assess the five model parameter settings using deterministic calibration by employing a statistical calibration methodology commonly referred as Bayesian calibration. Results from deterministic calibration are compared with Bayesian calibration, and up to 6.5 fold improvement in prediction accuracy is observed with the latter approach.