Modelling and characterization of membrane fouling in osmotically-driven membrane processes (ODMPs)

• Main Author: Lai, Li
• Other Authors: Chong Tzyy Haur
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Environmental engineering::Water treatment
• Online Access: https://hdl.handle.net/10356/148857
• Institution: Nanyang Technological University

With intensification of water scarcity, forward osmosis has been drawing increasing attention due to its low energy requirement and power of removing pollutants in wastewater treatment and desalination processes. Similar to other membrane processes, membrane fouling is still a big obstacle with respect to filtration effectiveness and membrane longevity in forward osmosis. To address this problem, this study focuses on two main part on investigation of membrane fouling: Characterization and Modelling. In this first part of this study, a non-destructive real-time method, ultrasonic time domain reflectometry (UTDR), was for the first time applied to characterize colloidal fouling in osmotically-driven membrane process. In a dead-end forward osmosis setup, the ultrasonic signal is revealed to be much more sensitive to fouling than traditional fouling indicator water flux, especially when active layer was faced to feed side. From qualitive analysis to quantitative analysis, the location of peak raised by colloidal deposition is used to differentiate the deposition on surface of active layer and inside of support layer after being analyzed by arrival time shift and confirmed by statistical testing. Moreover, the further short-time Fourier transformation (STFT) was applied to transfer the original time domain data into time-frequency domain, which reveal a noticeable magnitude reduction on high frequency components when Rayleigh scattering was triggered by particle inside of membrane pores. By contrast, the frequency components are kept nearly unchanged on AL-FS mode in which no internal fouling occurs. Finally, the total reflected power (TRP) calculations for the arrival-times and principal peaks in the frequency-domain correlated well with the overall fouling and internal fouling, respectively. In the second part of this study, the fouling model derived from mass balance equations in multiple boundary layers in forward osmosis membrane was studied and used to interpret colloidal particle deposition behavior in forward osmosis on AL-DS mode. Due to changing microscopic condition near concentration polarization layer of pollutants, particle deposition factor which characterizes the percentage of particle that attaches on membrane surface out of cross flow cannot be simply assumed as constant. In order to connect this experimentally inaccessible parameter with model, the parameter was extracted from optimization through series layers model and fed into a backpropagation neural networks with various experimentally-accessible real time parameters. Correlating particle deposition factor with real-time experimental parameters through trained neural networks achieves high prediction accuracy with R squared and Pearson’s coefficient as 0.938 and 0.970, respectively. Performance of neural networks bears strong relationship with learning rate, number of hidden layers and number of nodes in hidden layer. Finally, using gradient-based indices is able to identify the positive/ negative contributions of real-time parameters on particle deposition in different fouling stages: the most important factors promoting particle deposition on early stage include time and initial flux; the counterparts on late stage include solute flux and feed concentration. The result is in agreement with ‘two-stage’ fouling theory, which is characterized by transition from particle-membrane interaction energy dominative system to particle-particle interaction energy dominative system. In the third part of this study, the focus was put on how deposited particles with different particle size affect deposition of new particles in fouling process on TFC forward osmosis membrane, from perspective of characterization and modelling. The result shows particle concentration and initial flux plays increasingly important role in early fouling stage with decreasing particle size. By modelling trajectories of particles, it’s revealed that partition of partilce on surface and inside of an ideal cyndrical pore is dominated by particle concentration at early fouling stage, in which nearly all partciles transported from bulk ends up with blocking pore. By contrast, this relationship become less noticeable at late fouling stage due to strong repulsive effect from deposited particles. At late fouling stage, the status of blocked pore plays dominative role in determining the maxinum flux in which no pore blocking occurs