Location: Environmental Microbial & Food Safety Laboratory
Title: Deep learning model for membrane biofouling prediction using optical coherence tomography imagingAuthor
PARK, SANGHUN - University Of Ulsan College Of Medicine | |
BAEK, SANGSOO - University Of Ulsan College Of Medicine | |
CHUN, JONG AHN - University Of Ulsan College Of Medicine | |
PARK, YOUNGEUN - University Of Ulsan College Of Medicine | |
Pachepsky, Yakov | |
PARK, JONKWAN - University Of Ulsan College Of Medicine | |
CHO, KYUNGHWA - University Of Ulsan College Of Medicine |
Submitted to: Journal Membrane Science
Publication Type: Peer Reviewed Journal Publication Acceptance Date: 6/3/2019 Publication Date: 6/8/2019 Citation: Park, S., Baek, S., Chun, J., Park, Y., Pachepsky, Y.A., Park, J., Cho, K. 2019. Deep learning model for membrane biofouling prediction using optical coherence tomography imaging. Journal Membrane Science. 587:117164. https://doi.org/10.1016/j.memsci.2019.06.004. DOI: https://doi.org/10.1016/j.memsci.2019.06.004 Interpretive Summary: Biofouling is common in water distributions and filtration systems where biofilms provide habitats for various microorganisms, including pathogens. The optical coherence tomography provides an excellent source of data on biofilm development. The volume of his data is very large. The objective of this work was to research the applicability of the artificial intelligence algorithms to package these data into a predictive model to relate growth of biofilms to the dissolved organic matter in water. The deep learning artificial intelligence algorithms appeared to be very efficient and reproduced observed growth with high accuracy. Results of this work will be of interest for researchers and practitioners working with imaging for characterization and evaluation of microorganism habitats. Technical Abstract: A new approach to modeling of biofouling on nanofiltration membrane surfaces was developed using an artificial intelligence (AI) algorithm with in situ fouling image data obtained through optical coherence tomography (OCT). Fouling experiments were conducted on different bioavailable organic matter concentrations. The OCT device produced a total of 8,000 fouling layer images. A fluorescence regional integration (FRI) method was used to analyze three-dimensional (3D) excitation-emission matrices (EEMs) quantitatively and determine organic matter composition. The AI-based fouling growth model used high-resolution fouling layer images and dissolved organic matter composition as input data to train a network. The developed model considers two types of biofouling growth depending on the bioavailability of organic matter based on FRI analysis. Based on two-dimensional (2D) and 3D spatiotemporal results of continuous simulation, the performance of the model was satisfactory with R2 values ranging from 0.97 to 0.98 and RMSE values ranging from 2.35 to 2.30 for correlation of observed and simulated biofouling areas in the fouling images. These results demonstrated the development of a new tool for reliably and rapidly simulating biofouling processes using high-resolution fouling images and a deep learning technique. It is expected that a deep learning model can be readily constructed and queried by engineers and decision-makers working with nanofiltration membranes. |