Deep learning engine enhances X-ray image resolution and hydrogen fuel cell efficiency

Researchers from UNSW Sydney have developed an algorithm that generates high-resolution model images from lower-resolution computerized tomography (CT) scans.

The new process, detailed in a paper published on natural communicationhave been tested on individual hydrogen fuel cells to accurately model their internals and potentially improve their efficiency.

But the researchers say it could also be used in the future on human X-rays to provide medical professionals A better understanding of the small cell structures inside the body, which could enable better and faster diagnosis of a wide variety of diseases.

The team, which includes Professor Ryan Armstrong, Professor Peyman Mostaghimi, Dr Ying Da Wang and Kunning Tang from the School of Energy and Mineral Resources Engineering and Professor Chuan Zhao and Dr Quentin Meyer from the School of Chemistry, developed algorithms to improve understanding of what is happening inside Proton Exchange Membrane Fuel Cells (PEMFCs).

PEMFC uses hydrogen fuel to generate electricity and be a quiet and clean energy source that can power homes, vehicles and industries.

These fuel cells convert hydrogen, through electrochemistry, into electricity with the only byproduct of the reaction being purified water.

However, PEMFC can become less effective if water cannot properly drain out of the cell and then “flood” the system. Until now, it has been difficult for engineers to understand exactly how water escapes, or actually collects, inside fuel cells due to their very small size and very complex structure.

Improved resolution

Solution created by UNSW researchers study carefully to create a detailed 3D model using lower resolution X-ray images of the cell, while extrapolating the data from the accompanying high-resolution scan of a small portion of the cell.

In more basic terms, it’s equivalent to taking a blurry aerial shot of the entire town from an airplane, along with a very detailed photo of just a few streets, then having can accurately predict the layout of every road in it. the whole area.

Professor Armstrong said: “One of the reasons this study is so novel is that we are pushing the limits of what can be made of images.

“It’s very typical that when you use a piece of hardware, whether it’s a microscope or a CT scanner, the resolution of the image gets worse the further you zoom out.

“Our machine learning technique solves that problem, and the method is widely applicable wherever imaging is taking place, such as medical appor the petroleum industry, or chemical engineering.

“We have done preliminary super-resolution work with radiologists before, and we can surmise that by obtaining higher resolution images from the larger field of view, it is possible to diagnose diseases earlier, such as tumor cells, when they are smaller.”

DualEDSR . Algorithm

Dr Wang said in their published research their super-resolution algorithm, called DualEDSR, improves the field of view by about 100 times compared to high-resolution images.

And he agrees that the implementation in medical images This is an exciting future development.

“If you look at what we’re doing now and apply it to the medical field, it would be interesting to be able to visualize blood vessel and the flow of red blood cells through the capillary network is even more detailed,” he said.

“These out-of-hardware imaging and modeling methods extend beyond fuel cell imaging to enable higher resolution imaging with larger fields of view than was previously possible.”

One limitation to the modeling process as detailed in the study is the fact that low-resolution images at larger scales and high-resolution images at smaller scales require must be taken in the same location, using the same camera.

These are called “Area of ​​Interest” scanners and are specialized devices that may not currently be available at many facilities.

However, the team hopes that further research will allow deep learning techniques to produce similar results in the future when displaying images that were not captured in the same location and potentially even not used. using exactly the same tool or material.

PEMFC provides fast fuel

Now, researchers can provide detailed 3D models of the inside of the PEMFC for manufacturers to improve the management of produced water and make fuel cells more efficient.

During the training and testing of DualEDSR, the algorithm achieved 97.3% accuracy when generating high-resolution models from low-resolution images. It also created a high-resolution model in just 1 hour, compared to the 1,188 hours (equivalent to 50 non-stop days) that it would have taken to obtain a high-resolution image of the entire part. of fuel cells using a computer. CT scanner.

“From our model, we can quickly and accurately see where water tends to accumulate, and so we can help address those issues in designs,” said Dr. Meyer. Future”.

“In the industry, it is known that huge untapped performance improvements can be achieved using these cells, just by improving water management, and it is estimated to increase overall by 60%.

“For the past 20 years, until now, it has been very difficult to have an accurate model of these fuel cells because of the complexity of both the materials and the ways in which gases and liquids are transported, as well as the reactions. where electrochemical reaction occurs.

“Our interdisciplinary team has helped us do that, bringing in a lot of different expertise. That’s what the research is all about.”

More efficient PEMFCs could be an important factor in future clean and eco-friendly electricity supply, due to the fact that they only emit water and heat. In addition, they are compact and lightweight making them suitable for use in vehicles.

They can also be refueled as quickly, as quickly as a car, giving them a key advantage over battery-powered electric vehicles, which can take hours even with a quick charger.