Deep learning model for simulating influence of natural organic matter in nanofiltration

Area of Science:

• Membrane science and engineering
• Water treatment technologies
• Artificial intelligence in environmental applications

Background:

• Membrane fouling significantly impacts filtration performance in nanofiltration (NF) systems.
• Predicting membrane fouling is crucial for proactive control and maintaining system efficiency.
• Natural organic matter (NOM) is a primary cause of fouling in NF processes.

Purpose of the Study:

• To develop a predictive model for membrane fouling in NF systems.
• To investigate the influence of different types of NOM on fouling behavior.
• To forecast filtration performance and fouling layer thickness using deep learning.

Main Methods:

• Conducted lab-scale membrane fouling experiments with four model NOM foulants.
• Utilized optical coherence tomography (OCT) for real-time quantification of cake layer thickness.
• Developed and trained a long short-term memory (LSTM) deep learning model.

Main Results:

• The LSTM model accurately predicted permeate flux and fouling layer thickness.
• Achieved root mean square errors below 1 L/m²/h for permeate flux and 10 µm for fouling layer thickness.
• Demonstrated the model's effectiveness in simulating NOM fouling in NF systems.

Conclusions:

• Deep learning, specifically LSTM, is a viable tool for simulating and predicting membrane fouling.
• The developed model can aid in controlling membrane fouling and optimizing NF system performance.
• The approach shows potential for application in other membrane processes beyond NF.