Wednesday, August 4 | 2:00 pm EDT

Iterative Reconstruction and AI-Segmentation on In-Situ XRM of Reverse Osmosis Membranes


Yara Suleiman

University of Connecticut

Yara Suleiman is a first-year Ph.D. student at the University of Connecticut. After completing her B.S. degree at Manhattan College including a Research Assistantship position, she moved to UConn for graduate studies where she now works in the REFINE Lab (Reverse Engineering, Fabrication, Inspection, and Non-Destructive Analysis). Her work is centered around biomedical engineering with a primary focus on imaging systems. She is a primary operator of REFINE’s X-ray system (ZEISS Xradia 520 Versa), working with samples ranging from polymers and metals to biological specimens.


The development of reverse osmosis membranes that maintain performance at pressures up to 200 bar can help to replace energy-intensive thermal desalination with a high-pressure reverse osmosis (HPRO) process. This has the potential to reduce the energy consumption of brine concentration by up to 90%. In-situ X-ray microscopy was utilized to quantify changes under pressure in membrane structure of traditional sea-water reverse osmosis (SWRO) membranes and HPRO membranes for comparison. Changes such as reduced porosity and increased tortuosity that lead to decreased membrane permeance were characterized. Optimized parameters were established for imaging to achieve a high resolution of 1.5m, which is necessary for quantitative analysis. Using an iterative reconstruction method, samples could be imaged at 1201 projections rather than 2401 projections, which reduced the scan time by 50%. This was necessary to reduce motion artifacts and produce high fidelity data that can be caused by the high pressure being subjected onto the membranes. An unloaded, control scan was done on both the SWRO and HPRO membranes initially to provide a baseline for comparison. A force of 40N was then applied to the same membranes to observe the compressive effects.

In order to quantify the microstructural properties in each layer of membranes, a high-fidelity segmentation method has been developed using DigiM Solutions (Boston, MA) platform. Conventional methods such as thresholding are unable to produce reliable results due to challenges in separating pores and material in each layer as the greyscale values are close to one another. Also, greyscale values vary from slice to slice due to imaging artifacts inherent in XRM. The mentioned challenges make automated image analysis very difficult. Manual methods are time consuming, human-dependent, and prone to error eliminating the possibility of comparative study intended for the remainder of the project. An AI-based image segmentation process has been developed and has been verified to provide repeatable results over the entirety of the 3D Images. The resulting segmentation was done on the SWRO and HPRO membrane before and after compaction. A thorough 3D pore analysis has been performed which include porosity and pore size distribution in each layer in each sample type. Additionally, analyses such as thickness calculations, tortuosity and permeability were done on for each layer in the membranes. The expected changes such as reduced porosity and increased tortuosity after compaction were observed from this characterization.