For drinking water treatment applications, it is possible to predict the external porosity of an expanded bed of granular activated carbon in fluidized conditions. A new model has been developed with a 2% relative prediction error.

In order to supply sufficient and safe drinking water, water utilities use a treatment train consisting of several unit processes. One of these treatment unit processes is granular activated carbon (GAC) filtration, a crucial unit process widely used for its filtration and adsorption capabilities as a barrier for undesired macro and micro-pollutants. The point of interest for this research is one of the critical steps in the filtration part of the unit process, the backwashing procedure.
Backwashing is a cleaning procedure, which consists on stopping the normal operation of the filter and reversing the normal downward water flow. This upward flow leads to the expansion of the filter’s media and washes away any undesired particles caught in between and on the surface of the media. Inadequate backwashing can lead to unwanted operational outcomes, for instance, solids accumulation and mud balls or media washout, resulting in costly operational expenses. In addition, currently there is a tendency for water utilities to explore new sustainable GAC filter media, which have different expansion tendencies. These operational requirements create the need for the development of prediction models to estimate the expansion degree of the filter bed during backwash procedures. Additionally, a deep understanding of the phenomena that governs this unit process is required to increase its resiliency.

The main goal of this research was to predict the expansion degree of GAC in the water phase. In order to achieve this goal, two innovative approaches combining advanced laboratory techniques and prediction models was the course of action. The first approach was the development of an input model known as the AquaGAC model, to describe the different characteristics of porous media and perform checks using calculated and measured hydraulic parameters. With the combination of the AquaGAC input model and an existing fluidization model (FBI) that computed the expressions of five classical models, the prediction of porosity of the performed experiments was achieved. The second approach consisted on using a data driven model, which consisted on the combination of the outputs the FBI and AquaGAC models with several morphological parameters to derive empirical expressions that accurately estimated the external porosity.

Obtained results suggest that using the 10th percentile of the particle diameter in classical models, delivers porosity prediction errors of 10% in comparison to the 50th percentile used in practice with errors up to 25%. Based on symbolic regression, data driven models produced expressions with accurate correlations with porosity errors ranging from 2-5%.

The results of this research are encouraging as the AquaGAC input model can serve as a basis for other fluidization models that use porous media with different shapes. Recommendations are made to improve the experimental set-up, the accuracy in estimating the particle envelope and wet densities, and the quantitative evaluation of the orientation. Future research of the expansion behavior of mixtures is also recommended.