Document Type

Article

Publication Date

3-1-2023

Publisher

Elsevier

Source Publication

Fuel

Source ISSN

0016-2361

Original Item ID

DOI: 10.1016/j.fuel.2022.127020

Abstract

Combustion of pulverized coal involves millions of particles with a distribution of morphologies. A workflow is presented to characterize, classify, and model distributions of highly porous coal char particles. To understand the impacts of morphology, 3-D pore-resolving combustion simulations for 150 particles imaged with micro-CT are first performed. To facilitate modeling distributions of morphology in reactor-scale CFD, a machine learning algorithm is then trained to classify particles according to their expected combustion behavior and to select an accurate, computationally efficient 1-D particle-scale combustion model. Whereas existing approaches classify particles solely according to morphology and use 2-D measurements, the present approach classifies particles according to combustion behavior, using 3-D morphological data as input and 3-D pore-resolving combustion simulation data for training. To facilitate application of the workflow to other highly porous coal chars, an automated image analysis routine is developed to filter and segment particles and to measure their morphological parameters for use in classification. When using the most appropriate 1-D combustion model for each particle as determined by the 3-D simulations, the average relative error in 1-D effectiveness factors for the 130 particles tested compared to their 3-D effectiveness factors was 9.9%, while the average relative error in 1-D effectiveness factors predicted by the proposed workflow with automated image analysis and a pretrained random forest classifier was 13.7%.

Comments

Accepted version. Fuel, Vol. 335 (March 1, 2023). DOI. © 2023 Elsevier. Used with permission.

Download

Available for download on Tuesday, April 01, 2025

Share

COinS