Please use this identifier to cite or link to this item:
|Title:||Prediction of membrane fouling by data-driven technology during desalination and water reuse||Authors:||Tan, Kion||Keywords:||Engineering||Issue Date:||2022||Publisher:||Nanyang Technological University||Source:||Tan, K. (2022). Prediction of membrane fouling by data-driven technology during desalination and water reuse. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/157319||Project:||EN35||Abstract:||Today, one of the top priorities of seawater reverse osmosis (SWRO) desalination plants remains to achieve improved energy efficiency and cost-effectiveness of desalination and water reuse, in view of promoting sustainable development and the circular use of resources. The performance of SWRO desalination operations is inextricably linked to the effectiveness of fouling mitigation. The objective of this work is to develop data-driven machine learning (ML) models by employing the MATLAB® software to perform analytical model fitting of literature data related to RO membrane scaling and subsequently generate new predictions of the temporal evolution of the permeate flux based on unseen input data. The input parameters studied can be broadly classified into three main categories: feedwater characteristics, intrinsic membrane properties, and operating conditions. Comparing the fitting results of the models available in the three categories of linear regression, ensemble trees, and artificial neural networks, the results show that the ensemble tree, and more specifically the Bagged Trees model, presented the strongest agreement between the predicted and experimental data with a coefficient of determination (R2) of 0.94 and root mean squared error (RMSE) of 0.068055 while consuming the shortest training time of 7.0378 seconds. The three best-performing models from each of the categories of predictive models were subsequently used for the prediction of flux decline based on simulating a newly introduced set of unseen input parameters. The results demonstrated that the Bagged Trees model also performed the best in predicting the time-variability of the permeate flux decline, as reflected in the lowest RMSE of 0.0849 among the competing models. The influence of selected groups of input parameters (feedwater characteristics, intrinsic membrane properties, and operating conditions) on the fitting accuracy of the Bagged Trees model was evaluated by excluding them from the training and validation processes. The simulation process concluded that the feedwater characteristics were deemed to be indispensable in the accurate prediction of fouling for the Bagged Trees model. Lastly, the key advantages and limitations of the proposed ML models were discussed by comparing them with classical fouling predictive models. Overall, the Bagged Trees model showed great promise in predicting membrane scaling, which could support its potential ability to aid in optimizing RO seawater desalination operations.||URI:||https://hdl.handle.net/10356/157319||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||CEE Student Reports (FYP/IA/PA/PI)|
Updated on Jun 26, 2022
Updated on Jun 26, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.