Prediction of heel build-up on activated carbon using machine learning

Area of Science:

• Environmental Science
• Materials Science
• Chemical Engineering

Background:

• Adsorbent performance in air treatment systems declines over time due to heel buildup, the accumulation of non-desorbed substances.
• Current theoretical models cannot adequately predict heel buildup based on adsorption/desorption parameters.
• Understanding and predicting heel buildup is essential for designing durable and efficient air purification technologies.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting cyclic heel buildup of volatile organic compounds (VOCs) on activated carbons (ACs).
• To identify key factors influencing heel buildup, including adsorbent characteristics, adsorbate properties, and regeneration conditions.
• To provide a predictive tool for optimizing air treatment processes and selecting suitable adsorbents.

Main Methods:

• Applied two machine learning algorithms: XGBoost and a neural network (NN).
• Trained models using data on adsorbent characteristics, adsorbate properties, and regeneration conditions.
• Generated partial dependency plots to analyze feature interactions influencing heel buildup.

Main Results:

• The neural network (NN) model demonstrated superior predictive performance with an R² of 0.94.
• The XGBoost model achieved a predictive accuracy of R² = 0.81.
• Analysis revealed significant interactions between heel buildup and the considered operational and material parameters.

Conclusions:

• Machine learning, particularly neural networks, offers a robust approach to predict cyclic heel buildup in activated carbon systems.
• The developed models can optimize operating conditions to minimize heel buildup and improve adsorbent longevity.
• These predictive tools facilitate rapid screening of adsorbents, enhancing the efficiency of VOC removal in air treatment applications.