Graduate students Kasun Amarasinghe and Daniel Marino sit in a computer control room watching readouts from a grinder as it devours bale after bale of corn stover.

An Idaho National Laboratory engineer asks Amarasinghe and Marino if they’d like to make an adjustment. They consult the gauges on their computers, then each other, before giving a nod to increase the speed.

The control room overlooks the Process Development Unit (PDU) — a system of grinders, conveyors, dryers and pellet mills that turns biomass into a processed form ready for conversion into biofuels, bioproducts and/or biopower.

Researchers from industry or academia can run tests of new processing techniques at the PDU, part of the Department of Energy’s Biomass Feedstock National User Facility. Maximizing the efficiency of raw biomass processing could help a budding bioenergy industry grow into a thriving domestic energy source.

Biorefineries currently rely on human operators alone to adjust biomass preprocessing equipment. Human operators often perform well for plants that process gases, liquids or refined solids, but biomass variability poses unique operational challenges.

Biomass — anything from corn stover to wood chips to sorted municipal solid waste — doesn’t flow through machinery as easily and is more variable than, say, grain or oil. Consequently, biofuels producers have faced challenges during biomass preprocessing operations, the step right before the biomass gets turned into fuel.

A single clogged screen, hopper or conveyor belt, or a piece of machinery prematurely worn down by excess silica (i.e., sand), can shut down the entire biorefinery until the problem is fixed. As a result, pioneer commercial-scale biorefineries have struggled to achieve anything close to their production capacity, even two years into startup.

So Amarasinghe, Marino and INL engineers are actually working to overcome one of the biggest barriers facing the bioenergy industry.

The computer scientists from Virginia Commonwealth University are pushing the PDU’s limits. The trick is to maximize the number of bales processed per hour without overloading the machinery and shutting down the system.

Moisture is one of the most important variables. The more moisture contained within the bale, the harder it is to process through the grinder (a hammer mill with screens to limit the flow of oversized material).

Amarasinghe and Marino are making their decisions based on PDU data from hundreds of hours of operation. With those data, they developed computer models that relate process throughput (how much material the system processes per hour) and reliability (how long the system is shut down in the event of a problem). That gives them an estimate of PDU performance.

“This is entirely data driven,” Amarasinghe said. “The information that we’ve gained is immense.”

Kasun Amarasinghe (far left) and Daniel Marino (far right) are working with INL engineers to overcome one of the biggest barriers facing the bioenergy industry.

The research team aims to soon turn the decision-making process over to a computer. That computer will use a combination of artificial intelligence, computer models, and a system of sensors that detect moisture, dirt and other types of biomass variability to adjust the PDU machinery.

This takes the critical decision-making out of the hands of the human operators. As a result, decisions can be made much more quickly and reliably, thus helping to minimize bottlenecks and downtime.

Even without the artificial intelligence, their models achieved a substantial increase in the PDU’s reliability during the 320-minute test. The system shut down only once, for roughly 13 minutes. An unexpected wet spot in one of the bales clogged a screen.

In this initial test, the PDU was onstream and working 90 percent of the time (90 percent capability).

If the technology behind the PDU’s intelligent, adaptive control system can be disseminated to biorefineries across the U.S., it could improve reliability by augmenting human operators with the rapid analysis capabilities at which computers excel.

These automated systems could have applications well beyond bioenergy. Many industries — especially those that process bulk solids, such as paper manufacturers and agriculture producers — might benefit from a system capable of responding to variable materials automatically and nearly instantaneously.

For Amarasinghe and Marino, one challenge was how to capture the experience and know-how of the PDU’s engineers. “How do you mix data-driven knowledge with expert-driven knowledge?” Amarasinghe asked.

The answer turned out to be fuzzy logic. Where regular logic can only determine an answer to a question as “true” or “false,” fuzzy logic allows for degrees of truth. This is useful for questions such as, “Is this bale low moisture, medium moisture or high moisture?”

“There is not an abrupt transition from low to medium to high, so you need to create intermediate conditions,” Amarasinghe said. To do that, the team assigned “degrees of belonging” for each element. “Then you can expand to more complex systems with additional ‘rules’ determined by the experts. Eventually, it gets complex.”

This type of logic allows the computer to make good decisions when faced with a system as complex as the PDU that is very rapidly processing a material as variable as biomass.

Though a testing of the fully autonomous control system is not scheduled until late September 2018, some within the bioenergy industry have already taken notice. Researchers say at least one major biofuels producer has asked for a tour of the control system in the coming months.


Posted Sept. 6, 2017