Image-based machine learning applications for flocculation modelling in water treatment: Prospects towards automation

Abstract

• Flocculation modelling using Image-based ML and automation prospect is reviewed • Use of advance vision and computing platform on floc image analysis were elucidated • Computing complexities, groundtruthing and treatment variability limits ML adoption • Sensitivity analysis and weakly supervised techniques could aid automatic modelling • Autonomous and real-time water treatment modelling, and industry 4.0 is attainable Advancements in machine learning (ML) underscore the necessity for real-time monitoring in water treatment processes, particularly the flocculation phase. Digital imaging profoundly enhances the monitoring of precipitation, sedimentation, and especially coagulation and flocculation units. Flocculation is an essential phase that influences subsequent treatment stages like sedimentation and filtration, as it aids the removal of diverse pollutants. Yet, despite significant progresses in ML, and its advanced deep learning (DL) techniques, image-based flocculation modelling remains under-represented due to limited research. This review bridges that gap by critically examining image-based ML and DL techniques for flocculation modelling, charting a pathway towards automation. First, the evolution of flocculation kinetics modelling and best practices in floc image acquisition techniques are outlined. The significance of advanced vision platforms and image processing methodologies in flocculation monitoring is discussed, highlighting the benefits and limitations of various operators. State-of-the-art DL algorithms for semantic segmentation, data augmentation, weakly supervised learning, and real-time segmentation techniques are carefully examined, emphasizing their transformative potential in automated flocculation modelling and water treatment monitoring. Findings showed that by integrating sensitivity analysis with sparse sampling and weakly supervised learning, we propose innovative strategies to accelerate accurate floc mask generation for robust databases. Implementing real-time segmentation algorithms promises to revolutionize pollutant monitoring during treatment, propelling system automation towards Industry 4.0 and 5.0 standards. Our insights offer a roadmap for future research, aiming to promote the adoption of automated systems in water treatment and pollutant tracking.