Session: 15-03-02: Machine Learning and Deep Learning in Safety, Reliability, and Maintenance II

Paper Number: 164034

Ai-Driven Safety and Predictive Maintenance in Rotating Packed Bed Systems for Co₂ Capture 

Carbon capture and storage (CCS) has emerged as a critical technology in mitigating climate change, with Rotating Packed Beds (RPBs) gaining significant attention for their high mass transfer efficiency, compact design, and intensified CO₂ absorption capabilities. Making them to be easily retrofitted to existing offshore and industrial infrastructures for solvent-based Post Combustion Capture. However, the widespread deployment of RPB-based CO₂ capture systems in industrial applications will faces operational challenges, particularly in safety, fault detection, and predictive maintenance. These systems are exposed to mechanical stresses, solvent degradation, structural wear, foreign object obstruction and process instabilities, which can lead to efficiency losses and potential hazardous failures. Ensuring the safe and reliable operation of RPBs is, therefore, paramount for the long-term sustainability and commercial viability of this technology.

This research introduces a computer vision-based safety monitoring system utilizing YOLOv8 (You Only Look Once, Version 8), a deep learning model for real-time object detection and anomaly recognition. The objective is to automate safety inspections, detect system irregularities, and predict failures before they escalate into critical issues. Unlike traditional manual inspections and rule-based monitoring, which are often inefficient and prone to human error, computer vision provides a robust, scalable, and continuous monitoring solution that enhances reliability in CO₂ capture operations.

The primary contribution of this work lies in applying advanced deep learning-based computer vision techniques to safety-critical carbon capture systems, bridging the gap between artificial intelligence, process engineering, and industrial safety. This study introduces an AI-powered framework that not only detects visual anomalies in RPB operations but also integrates predictive analytics for fault diagnosis and preventive maintenance scheduling. The approach ensures; 1) automated detection of structural wear and operational anomalies, 2) predictive maintenance & fault prevention and 3) enhanced safety & risk mitigation.

High-resolution images and video footage of operating RPB units are collected from laboratory-scale reactors. Datasets are annotated using bounding boxes to label different failure modes (e.g., structural defects, excessive vibrations, uneven solvent distribution. YOLOv8 is trained on the annotated dataset, learning to detect multiple classes of defects and anomalies with high accuracy. The model is optimized for real-time inference, ensuring that safety inspections are performed continuously and without delays.

By integrating computer vision and deep learning, this research provides an automated, scalable, and highly reliable safety solution for CO₂ capture using Rotating Packed Beds. The proposed approach aligns with global sustainability goals by ensuring that carbon capture technologies are not only effective but also safe, resilient, and optimized for large-scale industrial deployment. This work lays the foundation for future AI-driven safety standardization in CCUS systems, demonstrating the potential of deep learning-based anomaly detection in advancing industrial safety and predictive maintenance strategies.

Presenting Author: Mohammadu Bello Danbatta Sultan Qaboos University

Presenting Author Biography: PhD Candidate at Sultan Qaboos University, studying the applicability of rotating packed bed for carbon capture storage and utilization for industries.

Authors:

Mohammadu Bello Danbatta Sultan Qaboos University
Nasser Ahmed Al-Azri Sultan Qaboos University
Muhammad Abdul Qyyum Sultan Qaboos University
Nabeel Al-Rawahi Sultan Qaboos University