In-situ monitoring of the membrane fouling using image sensing: Experimental and modeling studies

Abstract

Fouling is a major problem in membrane filtration processes that deteriorate filtration performance by reducing water productivity, permeate water quality, and increasing operating pressure. To effectively evaluate the influence of fouling, this study employed novel image sensing techniques and a deep learning model. Initially, the use of real-time sensing techniques provided growth information of the fouling layer during the filtration. The monitoring information was coupled with analytical results to figure out the harmful influence originated from various fouling resources. In particular, the effects of natural organic matters (NOMs) were carefully investigated to explore the organic- and bio-fouling formation on the membrane surface. Then, this study utilized image sensing techniques to develop a fouling resistive feed channel spacer based on its monitoring capability. Furthermore, the fouling prediction model was developed to estimate the filtration performance utilizing the image-based deep learning model. First, optical coherence tomography (OCT) was applied to quantify the growth of the fouling layer, representing high resolution fouling images in real-time (Chapter 3). The organic matter characterization proved that the fouling layer growth was highly affected by the degradation of the dissolved organic matter in the feed water. In addition to the organic matter degradation, the bacterial activity was investigated via confocal laser scanning microscopy (CLSM) at the end of filtration. The quantified volume at OCT and CLSM described the superior monitoring capability of the OCT for evaluating the fouling development on the membrane. Second, this research explored the influence of organic matter bioavailability on the fouling layer development by comparing the filtration of two feed waters (wetland water and graywater) (Chapter 4). The wetland water mostly consisted of humic acid- and fulvic acid-like matter with poor bioavailability, whereas the graywater contained aromatic proteins and microbial byproduct-like matter with high bioavailability. The bioavailability difference yielded a significant difference in the bacterial volume (obtained by CLSM) as well as the total fouling volume (obtained by OCT), proving the importance of organic matter characteristic on the membrane fouling. Third, the practical application for developing a novel feed channel spacer was addressed using image sensing techniques. The honeycomb-shaped spacer was developed by 3D printing technology (Chapter 5). The modified hydrodynamic effect resulted in superior filtration performance generating a higher permeate flux than the standard diamond-spacer. As well, OCT demonstrated the fouling mitigation effect originated from the honeycomb-shaped spacer by comparing the thickness of the fouling layer (119.0 µm) with that of the standard spacer (175.5 µm). Lastly, an image-based deep neural network (DNN) model was developed for estimating variation in permeability and fouling layer thickness (Chapter 6). Convolutional neural network (CNN), which is one of DNN, was applied to handle the fouling images obtained by OCT during filtration. The proposed DNN model showed a higher prediction accuracy than conventional mathematical models (e.g., Faridirad model and pore & cake filtration model). Especially, the DNN model visualized the growth of the fouling layer in the 3D images, thereby it identified the superiority in predicting the fouling layer growth. I anticipated that these approaches would increase the applicability of image sensing and deep learning techniques for evaluating membrane fouling to extend the knowledge of the membrane fouling and mitigation strategy.