Deep learning model for simulating influence of natural organic matter in nanofiltration. (1st June 2021)

• Record Type: Journal Article
• Authors: Shim, Jaegyu; Park, Sanghun; Cho, Kyung Hwa
• Is Part Of: Water research. Volume 197(2021)
• Publication Date: 2021-06-01
• Subjects: Membrane filtration -- Natural organic matter -- Deep learning -- Long short-term memory; LSTM long short-term memory -- NOM natural organic matter -- OCT optical coherence tomography -- ANN artificial neural network -- CNN convolution neural network -- RNN recurrent neural network -- DI deionized water -- HA humic acid -- BSA bovine-serum-albumin -- SA sodium alginate -- TA tannic acid -- DOC dissolved organic carbon -- FRI fluorescence regional integration -- SD-OCT spectral domain optical coherence tomography -- ReLU rectified linear unit -- RMSE root mean square error -- MWCO molecular weight cut-off; Water -- Pollution -- Research -- Periodicals; 363.7394
• DOI: 10.1016/j.watres.2021.117070
• Languages: English
• ISSNs: 0043-1354

Abstract

Highlights: Natural organic matters (NOMs) affected membrane filtration performance. Permeability depended on applied pressure, initial flux, and the type of NOMs. A cake layer was observed due to combination of humic acid and calcium ions. Long short-term memory model was suitable for simulating water treatment processes. Abstract: Controlling membrane fouling in a membrane filtration system is critical to ensure high filtration performance. A forecast of membrane fouling could enable preliminary actions to relieve the development of membrane fouling. Therefore, we established a long short-term memory (LSTM) model to investigate the variations in filtration performance and fouling growth. For data acquisition, we first conducted lab-scale membrane fouling experiments to identify the diverse fouling mechanisms of natural organic matter (NOM) in nanofiltration (NF) systems. Four types of NOMs were considered as model foulants: humic acid, bovine-serum-albumin, sodium alginate, and tannic acid. In addition, real-time 2D images were acquired via optical coherence tomography (OCT) to quantify the cake layer formed on the membrane. Subsequently, experimental data were used to train the LSTM model to predict permeate flux and fouling layer thickness as output variables. The model performance exhibited root mean square errors of <1 L/m 2 /h for permeate flux and <10 µm for fouling layer thickness in both the training and validation steps. In this study, we demonstrated that deep learning can be used to simulate the influence of NOMs on the NF system and also be applied to simulate other membrane processes.