Skip to main content

Neural Networks And Dynamite: AI Engineer Peter Kirk Talks About His Fascination With Coal Power Plants

Tomas Kellner
September 20, 2016
In April 2016, GE Power acquired the Boston-based machine learning and data analytics startup NeuCo Inc., which uses software and artificial intelligence to improve the efficiency of coal-fired power plants. They are still the most common source of electricity in the world.
NeuCo’s systems are now a key part of the world’s first “Digital Power Plant for Steam,” a set of technologies that can dramatically reduce greenhouse gas emissions of coal-fired power plants by improving their operations and efficiency. GE Reports editor Tomas Kellner sat down with NeuCo’s former CEO Peter Kirk to talk about the technology. Here’s an edited version of the conversation.

Tomas Kellner: Why did you decide to join a software startup that can optimize coal-fired power plants? In many ways, coal seems like an old-fashioned technology that evokes Andrew Carnegie more than Elon Musk.

Peter Kirk: First of all, I was very interested in energy and the environment. I guess you can either pick up a sign and go protest coal or you can try to do something to make it cleaner, and more efficient, and more productive. It is going to be a part of the energy landscape for the foreseeable future. Before I started NeuCo, I had been in energy for a while. I had recently done a company that was a fuel cell startup company. That was so far out into the future that the stakes were just very high, and I looked at NeuCo as a chance to really take a 19th century technology and bring it into the modern age.

I think everyone should go and see one of these plants because they've provided the population of the U.S. with about half of the electricity they’ve used in their lives. They're an odd mix of advanced technology and 19th century engineering. For example, when the operators need to remove slag from the boiler, which is the result of bad combustion, they sometimes take a piece of explosive, tape it to the end of broom, put it where they think it's going to do the most good and blow it up. I love the idea that the same plant that is using a neural net-based artificial intelligence optimization technology is using a broomstick with a stick of dynamite attached to it.

RDK8BoilerMills Above: The boiler mills of the RDK8 the wold's most efficient coal-fired power plant. Top image: The boiler (orange) is by far the largest part of a coal-fired power plant. Images credit: GE Power

GE Reports: Tell us about the AI.

PK: We made a collection of software that leverages artificial intelligence to improve the operations of a coal-fired power plant. This is not as easy as it sounds. There's lots of data. It's very difficult to apply a more traditional analysis. Plus the machines are so complex. They degrade. They move around. They are sensitive to fuel quality. There's a lot of chaos.

GER: What kind of data are you talking about?

PK: There are thousands of points that we can measure in every power plant. What we're doing is applying neural learning, human learning, human knowledge and advanced controls – whatever it takes. They're all in our collection of software, and we’re using the data to improve operations. There’s neural, but also a piece of traditional AI software that can capture existing human knowledge, what's in an engineer's mind.

GER: Can you give a concrete example of the chaos you mentioned?

PK: One very chaotic but key process mixes air in the coal boiler. Imagine the piston in your car. That's where the fuel and air mix, and the fuel efficiency of your car is going to depend on how well it’s done. The difference is that the coal boiler is a 10- to 15-story box that’s also about 50 feet wide and filled with burning rock. It’s a very chaotic environment, and there’s a lot of room for the mixing to be done poorly. The consequences of bad fuel-air mixing leaves you with unburned fuel, but also higher nitrogen oxide (NOx) and carbon dioxide emissions.

We're also trying to keep the inside of that boiler clean – but not clean too much, because most of the cleaning is done with water or steam, and if you over-clean, you are going to be putting corrosive media on a burning tube. That’s a recipe for rust times a thousand. In fact, the No. 1 cause of outages in the coal-fired fleet is tube ruptures and tube leaks. We optimize the soot-cleaning process so that plants can clean less and run more efficiently. CO2 is also very corrosive and something we manage by improving combustion. Our software will save you money on fuel, it will keep smog and carbon from getting into the air, and it will make the machines break down less.

GER: Can you quantify the savings?

PK: The average 600-megawatt power plant that's able to save 1 percent on fuel puts about $1 million a year back into its pocket. And the fuel savings come on top of the carbon and nitrogen oxide reductions.

TK: Does your software work with any kind of technology?

PK: Yes. NeuCo was an independent company. We had to be hardware-agnostic. Over the years, we worked with a number of partners, most recently with Alstom, which had a very strong presence in coal. GE acquired Alstom’s energy business last fall, and NeuCo’s technology is now helping them to optimize Alstom’s fleet.

RDK83DviewpictureGermany A drawing of the RDK8 power plant. The boiler is brown and the steam turbine lilac on the right. Image credit: GE Power