Evaluation of membrane fouling and damage by machine-learning classification of pyrolysis-GC/MS and hyperspectral imaging data

Abstract

To investigate membrane fouling and damage various experiments require a lot of devices, efforts, and time. As the state-of-art technologies, XPS (X-ray photoelectron spectroscopy), FEEM (fluorescent excitation-emission matrix), and FT-IR (Fourier transformed infra-red spectroscopy) have been frequently used to analyze the information on the characterization of foulants, type and distribution of membrane surface damage. However, due to the complexity of organic foulants, types of damage and localized responses, a fast and reliable examination method is still needed. In this thesis, pyrolysis with GC/MS (Pyro-GC/MS) and the short-wave infrared hyperspectral image (SWIR-HSI) have been adopted as novel analytical techniques for the investigation of membrane organic foulants and damage, respectively. SWIR-HSI technique combined with multivariate image analysis was found to be capable to determine physically and chemically damaged membranes as a novel quantitative and qualitative tool compared to existing methods such as autopsy and Fujiwara test. Among the four different classifications of supervised learning models (partial least-squares discriminant analysis, K-nearest neighbors, support vector machine, artificial neural network), SVM provided the most distinct identification images of membrane damage. In addition, preprocessed spectra obtained from HSI were correlated with chlorine contents corresponding to the membrane damage, and showed a good correlation (R$^2$= 0.9015) with XPS data. A novel method for the characterization of organic foulants was proposed by interpreting and categorizing Pyro-GC/MS signals of model organic matters. Total 342 pyrolysis products were obtained from the Pyro-GC/MS of 34 model organic compounds, then categorized into 21 chemical groups according to their chemical or molecular structures. After the application of partial least squares-discriminant analysis (PLS-DA), the representative chemical groups could be categorized into four foulant types (Protein, Saccharide, Lipid, and Natural organic matter), and verified as a tool to provide fingerprints of complex organic foulants. The methods proposed in the thesis will provide more in-depth information on the membrane foulants and damage both in a quantitative and qualitative way.