Spectral fusion-based machine learning classifiers for discriminating membrane breakage in multiple scenarios

Yang Yu; Hui Jia; Fei Gao; Haifeng Zhu; Lei Zhang; Jie Wang

doi:10.1016/j.watres.2024.121714

Spectral fusion-based machine learning classifiers for discriminating membrane breakage in multiple scenarios

Water Res. 2024 May 2:257:121714. doi: 10.1016/j.watres.2024.121714. Online ahead of print.

Authors

Yang Yu¹, Hui Jia¹, Fei Gao¹, Haifeng Zhu¹, Lei Zhang², Jie Wang³

Affiliations

¹ State Key Laboratory of Separation Membranes and Membrane Processes, Tiangong University, Tianjin 300387, China; School of Environmental Science and Engineering, Tiangong University, Tianjin 300387, China.
² Shenyang Academy of Environmental Sciences, Shenyang 110167, China.
³ State Key Laboratory of Separation Membranes and Membrane Processes, Tiangong University, Tianjin 300387, China; School of Environmental Science and Engineering, Tiangong University, Tianjin 300387, China; Hebei Industrial Technology Research Institute of Membranes, Cangzhou Institute of Tiangong University, Cangzhou 061000, China. Electronic address: wangjiemailbox@163.com.

PMID: 38723357
DOI: 10.1016/j.watres.2024.121714

Abstract

Membrane breakage can lead to filtration failure, which allows harmful substances to enter the effluent, posing potential hazards to human health and the environment. This study is an innovative combination of fluorescence and ultraviolet-visible (UV-Vis) spectroscopy to identify membrane breakage. It aims to unravel more comprehensive information, improve detection sensitivity and selectivity, and enable real-time monitoring capabilities. Fluorescence and UV-Vis data are extracted through variance partitioning analysis (VPA) and integrated through a decision tree algorithm to form a superior system with enhanced discrimination capabilities. VPA improves discrimination efficiency by extracting key information from spectral data and eliminating redundancy. The decision tree algorithm, on the other hand, can process large amounts of data simultaneously. In addition, the method has a wide range of applications and can be used in various scenarios accurately. The scenarios include domestic sewage, micropollutant water, aquaculture wastewater, and secondary treated sewage. The experimental results validate the application of machine learning classifiers in membrane breakage detection with an accuracy rate of 96.8 % to 97.4 %.

Keywords: Decision tree algorithm; Fluorescence spectroscopy; Membrane breakage; Spectral fusion; Ultraviolet-visible spectroscopy.