# Iron-Based Adsorbents Derived from Groundwater Deferrization Sludge for Spent Oil Removal from Aqueous Media

## TL;DR

This paper explores using iron oxide nanoparticles made from groundwater sludge to effectively remove oil from water, with promising results and machine learning models for optimization.

## Contribution

The study introduces a novel method for synthesizing iron-based adsorbents from sludge and demonstrates their high oil sorption capacity and stability.

## Key findings

- Iron oxide sorbents with up to 99% magnetite content were synthesized using citric acid and urea at specific temperatures.
- The oil sorption capacity reached 6.1 g/g, outperforming many existing sorbents.
- Machine learning models accurately predicted sorption capacity with R² = 1.0.

## Abstract

The paper presents the results of the synthesis and study
of iron
oxide-based sorbents (Fe
x
O
y
-NPs) obtained from iron removal station sludge by
exothermic combustion in solution using glycine, urea, citric acid,
and urotropine as reducing agents. X-ray phase analysis revealed that
the phase composition depends on the nature of the reducing agent
and temperature: at 300–500 °C, the magnetite content
reached 97–99% for citric acid and urea, whereas when using
glycine, the Fe3O4 fraction did not exceed 30%.
The point of zero charge values shifted to the alkaline region with
increasing synthesis temperature, reaching 8.8 at 700 °C. The
specific surface area for methylene blue was up to 186 m2/g, but the calculated values exceeded the BET data by 3.5–4
times due to multilayer sorption on the functionalized surface, which
is consistent with the FTIR spectra. The oil sorption capacity (OSC)
of the synthesized sorbents reached 6.1 g/g (glycine, 500 °C),
which is comparable to or exceeds the indicators of a number of natural
and commercial sorbents. After five sorption-regeneration cycles at
800 °C, the OSC decreased by only 15.7%, confirming the stability
of the material. The constructed polynomial and machine learning models
(CatBoost, XGBoost) provided high accuracy of OSC prediction (R
2 = 1.0), which demonstrates the promise of
machine learning for optimizing synthesis conditions.