Imagine a black box with knobs on the outside that you can turn. If you input some fuel, the box produces electricity. By adjusting the knobs you can change the power output, but there’s a catch—you’re not sure how to turn the knobs to produce the most power.
This black box roughly characterizes research on microbial fuel cells—renewable energy sources that use bacteria to convert
biodegradable materials, like wastewater pollutants, into electricity. And the knobs in this scenario are nanostructures, such as carbon nanotubes, that researchers Frank Chaplen and Hong Liu, of OSU’s Department of Biological and Ecological Engineering, and Jun Jiao, of Portland State University’s Department of Physics, were using to boost the power output of microbial fuel cells.
Although some evidence in the scientific literature suggested that adding nanostructures to the surface of the anodes where the bacteria in fuel cells live could improve the power output, they didn’t know much about how or why. Not only that, it’s difficult to control properties of nanostructures, like width or density. It gave them too many variables to work with to solve their problem in any reasonable amount of time.
This is where Alan and Xiaoli Fern come in. Chaplen asked the couple, who both teach in OSU’s School of Electrical Engineering and Computer Science (EECS), to create a complicated mathematical algorithm that would inform the researchers which of the myriad of variables would be best to tackle first—in other words, which way to turn the knobs.
Alan Fern is an expert in automated planning and decision theory, which uses computing power to make intelligent decisions about sequential problems. And his wife, Xiaoli, specializes in active machine learning, which aides in identifying the most useful data points, which was at the core of the solution.
And so, for the first time, although they have been together since graduate school, the Ferns’ academic interests converged and they began working on the problem together.
“Traditional research has focused mostly on design problems that have clean, analytical solutions, which requires many simplifying assumptions. We come at the problem from a different angle—we start with realistic, messy problems and design algorithms that solve them with raw computing power,” Alan says.
They saw it as an opportunity to make a difference not only for microbial fuel cell research, but for all science—and especially for problems where the experiments are difficult to control. For example, in the microbial fuel cell project, instead of requiring an exact density of the nanomaterial, their algorithm could account for a range.
It was just the kind of math-oriented challenge that graduate student, Javad Azimi, was looking for when he joined the project as a research assistant—helping to design the algorithms and writing the software.
“I really love math, and I like working with real data. So when they told me about it, I said, ‘Yeah, let’s do it!’” Azimi says.
The team set to work on helping Chaplen and Liu find out which nanomaterials, such as gold, iron, or carbon nanotubes, and what properties, such as length, width and density, would most likely produce the best power output.
“These statistical models try to capture the researcher’s uncertainty about regions they haven’t explored and take advantage of regions they have explored fairly thoroughly,” Alan says.
They also performed simulated experiments that can be run hundreds of times by a computer using past data. The Ferns and Azimi used this type of modeling to inform decisions about what would be the best experiment to run next, but also which experiments would be advantageous to run in parallel. It saved Chaplen, Liu and Jiao both cost and time.
”These experiments are very time consuming and the researchers can’t afford to run them sequentially, so they have to be in parallel, and we can help them figure out which experiments would complement each other in terms of the information they provide,” Xiaoli says.
The team was able to successfully identify nanomaterials that enhanced power production by 10 to 20 times during their current four-year grant from the Oregon Nanoscience and Microtechnologies Institute in collaboration with the U.S. Army Research Laboratory, which is due to end this year.
The Ferns and Azimi have also applied their work to data from a project examining hydrogen produced from cyanobacteria as a potential renewable energy source.
In the future they expect to continue working with the team of researchers studying microbial fuel cells. In fact, they have already submitted another grant which proposes also delving into the nanotechnology side of the project to help Jiao advance the understanding of nanostructure properties. Nanotechnology has diverse applications in many areas including medicine, electronics, and green energy production.
Azimi said that after presenting their research in papers and at conferences he has discovered that it could apply to areas that they had never even thought of—such as improving the movement of robots.
“Because we are working to solve real problems with our algorithms, I believe the impact of our work will be really high,” says Azimi, who plans to continue this work for his dissertation.
-Story by Rachel Robertson