Researchers use machine learning to help reduce the emissions of heavy vehicle engines.
As the world moves toward electric vehicles, large vehicles such as long-haul trucks are lagging behind. The reason: it’s a real challenge to convert their engines to run on electricity.
Researchers at the Illinois Institute of Technology may have found a solution. Led by Carrie Hall, Associate Professor of Mechanical and Aerospace Engineering, the team is using machine learning and computer modeling to help transition large diesel engines to run on alternative fuels.
Alternative fuels could be an option for traditional diesel-burning engines—but the conversion isn’t a straightforward procedure. Engines are designed with a specific fuel in mind, and simply swapping one fuel for another won’t work.
That’s where machine learning and modeling can lend a hand. Hall created a computer model that can help convert diesel engines to run on a different fuel with a simple software update.
The key factor is how fuel is ignited, which isn’t easy to measure inexpensively inside an engine cylinder. Hall’s model instead pulls data from simpler, cheaper sensors placed outside the cylinder where combustion takes place to diagnose the ignition process.
Due to the nature of the reaction, a model that simply performs multidimensional fluid dynamics calculations isn’t enough. That’s because, for the model to be useful for engine operation, it needs to run in real time and anticipate ignition reactions that can take place thousands of times a minute.
“Our models are used to provide some system feedback,” said Hall. “Understanding the timing of [fuel ignition] gives us an idea of how it was tied to something like fuel injection, which we then might want to adjust based on that feedback.”
The models chosen by some engine control designers use machine learning techniques, or store large data tables to reduce calculations, to work faster. Some models use neural networks to model combustion. But according to Hall, “the problem is that then it’s just a black box, and you don’t really understand what’s happening underneath it, which is challenging for control, because if you’re wrong, you can have something that goes very wrong.”
Hall has taken a different approach, though. Her team’s models are based on the underlying physics and chemistry of the process, even if it gets very complex. The team’s model used the complicated version of the calculations as a starting point and looked for ways to generate equations that could be solved faster while maintaining industry standards for accuracy in control models.
“We’ve tried to capture all the underlying effects, even if it’s in a more detailed way than we know we’re going to really be able to use for real-time control, and let that be our reference point,” said Hall. “Then we simplify it down by using things like neural networks strategically, but we keep this overall structure so that we understand what each piece means and what it’s actually doing inside there.”
As a result, Hall’s model is more adaptable than a pure machine learning approach. When modeling the use of a new fuel, Hall’s model can simply update some parameters that correspond to measurable fuel properties. In contrast, a model based on a pure machine learning approach would need to be retrained from scratch.
And since it’s basically a software upgrade, vehicle users can input the update into their engines’ software at a much lower cost without having to change the vehicle’s hardware.
Hall has already applied her model as a potential intermediate solution for moving diesel engine trucks to gasoline. “There’s an anticipation that with electric vehicles being more common for passenger cars in the United States that there’ll be a lot of extra gasoline that’s not getting used,” said Hall. “That gasoline can be used on heavier-duty vehicles.” Eventually, Hall’s goal is to make engines smart enough to use a variety of fuels—including ones that are carbon-neutral or carbon-negative.
Professor Hall discusses her work on the ASME Dynamic Systems and Control Division podcast.
Hall’s model, developed in collaboration with Argonne National Laboratory, Navistar, and Caterpillar, aims to help companies understand the underlying combustion process—and assist them in designing engines to use fuels other than diesel.
Heavy-duty trucks account for about a quarter of on-road vehicle energy consumption in the U.S., even though they represent only about 1 percent of the vehicles on the road. Enabling these vehicles to transition away from diesel promises to have a significant impact on emissions.
“Everything that we’re doing is looking at trying to get to cleaner and more efficient vehicles,” said Hall.
Read more about technological developments in diesel use at Hybrid Transport Truck Engine Could Replace Diesel.