MIT Research News' Journal

Small engine packs a punch

Noise, excessive vibration, and relative inefficiency are drawbacks of the piston-based internal combustion engines (ICE) that power today’s lawn and garden equipment, such as leaf blowers and lawn trimmers.

But now MIT startup LiquidPiston has developed a rotary ICE that it says is significantly smaller, lighter, and quieter, as well as 20 percent more fuel-efficient than the ICEs used in many such small-engine devices.

“If you think of handheld tools — for example, a chain saw or hedge trimmer — after about a half hour you don’t want to use it anymore because your hand feels like it’s going to fall off,” says Alexander Shkolnik PhD ’10, president of LiquidPiston and co-inventor of the engine. “Our engine has no vibration at all and it’s a lot quieter. It should be a much nicer user experience all around.”

LiquidPiston’s 70-cubic-centimeter engine, the X Mini, produces about 3.5 horsepower at 10,000 RPM; at 4 pounds, it’s also about 30 percent smaller than the four-stroke, 50-cubic-centimeter piston ICEs it aims to replace. When fully complete, Shkolnik says, the X Mini could churn out about 5 horsepower at 15,000 revolutions per minute, and weigh 3 pounds.

The engine runs the novel high-efficiency hybrid cycle (HEHC) — developed by Shkolnik and his physicist father, Nikolay — that achieves combustion at constant volume and overexpansion for greater energy extraction. With only two moving parts, a rotor and shaft, and no poppet valves — commonly used in other four-stroke ICEs to control fuel intake — the engine also has reduced noise, vibration, and harshness characteristics, Shkolnik says.

Initial applications will be handheld lawn and garden equipment, Shkolnik says. But the engine can be scaled and modified for other applications, including mopeds, drones, marine power equipment, robotics, range extenders, and auxiliary power units for boats, planes, and other vehicles. The company has also demonstrated proof-of-concept for high-efficiency diesel versions of the engine, including the 70-horsepower X1 and the 40-horsepower X2, for generator and other applications. The company hopes to eventually develop small diesel versions of the X Mini engine for military applications.

“If you look at a 3-kilowatt military generator, it’s a 270-pound gorilla that takes five people to move around,” Shkolnik says. “You can imagine if we can make that into a 15-pound device, it’s pretty revolutionary for them.”

Shkolnik presented a paper on both the X2 and X Mini on Nov. 19 at the Society of Automotive Engineers’ 2014 Small Engine Technology Conference and Exhibition in Italy.

An inverse Wankel

The X Mini is essentially an upgrade in design and efficiency of the compact Wankel rotary engine, invented in the 1950s and used today in sports cars, boats, and some aircraft.

In the Wankel, a rounded triangle rotor spins in an eccentric orbit within an oval chamber, with each rotation producing three power strokes — where the engine generates force. In the X Mini, an oval rotor spins within a modified, rounded triangular housing.

“We’ve inverted everything about the traditional rotary engine, and now we can execute this new thermodynamic cycle [HEHC] and solve all the problems that were plaguing the traditional Wankel engine” for small-engine applications, Shkolnik says.

A Wankel engine, for instance, uses a long combustion chamber (like a thin crescent moon), which contributes to poor fuel economy — as the flame can’t reach trailing edges of the chamber and gets quenched by the chamber’s large surface area. The X Mini’s combustion chamber is rounder and fatter, so the flame burns over less surface area.

Air and fuel intake and gas exhaust in the X Mini occur through two ports in the rotor, opened or closed as the rotor revolves, removing the need for valves. Asymmetrical location of these ports slightly delays the exhaust process during expansion. This allows for HEHC’s overexpansion process — from the Atkinson thermodynamic cycle, used in some hybrid cars — where gas is expanded in the chamber until there’s no pressure, allowing the engine more time to extract energy from fuel. This design also accommodates HEHC’s “constant volume combustion” — from the Otto thermodynamic cycle, used in spark-ignition piston engines — where compressed gas is held in the chamber for an extended period, letting the air and fuel mix and ignite completely before expanding, resulting in increased expansion pressures and higher efficiency.

“It takes a long time to burn fuel in an engine,” Shkolnik says. “In most engines, by the time you’re burning fuel, you’re expanding gases, and you’re losing efficiency from the combustion process. We keep combusting while the rotor is at the top of the chamber and force combustion under those conditions. It’s much more efficient that way.”

Additionally, the X Mini has relocated the apex seals, leading to decreased oil consumption. In Wankels, apex seals join the edges of the triangular rotor, where they slide and move. Lubricating them requires supplying the air-fuel mixture with large amounts of oil that burns and leaks, increasing emissions and oil consumption. In the X Mini, however, these seals are located in the triangular-shaped housing that stays put. “Now we can supply tiny amounts of oil through the stationary housing, exactly how much oil the seal needs, and you’re not burning any oil and you’re not losing any oil to the environment,” Shkolnik says.

LiquidPiston’s “roadmap”

An interest in robotics and artificial intelligence led Shkolnik to MIT as a PhD student in electrical engineering and computer science in 2003. That year, Nikolay Shkolnik filed his first HEHC patent, and his son learned about the MIT $50K Entrepreneurship Competition (now $100K) in a class that focused on tech entrepreneurship. They teamed up with students at the MIT Sloan School of Management to create a business plan and pitch an HEHC engine in the 2004 competition, where they took home the $10,000 runner-up prize to launch LiquidPiston.

The competition itself proved helpful to the father-and-son entrepreneurs — who, at that point, had no startup experience. In building a detailed business plan and learning how to explain their technology to investors, “It really showed us a roadmap for what to do and we were forced to think through a lot about issues we were going to face,” Shkolnik says.

Over the next six years, Shkolnik helped his father develop the LiquidPiston engine out of the family garage, using skills he honed in MIT’s Robot Locomotion Group, led by Russell Tedrake, an associate professor of electrical engineering and computer science. “It was a lot of optimization, and control, and simulations, and modeling,” he says. “All those same techniques are applicable to designing an engine.”

Shkolnik attributes much of LiquidPiston’s development to the extended MIT community. During the $50K, venture capitalist Bill Frezza ’76, SM ’78 mentored the team; his firm then became an early investor. MIT Sloan team members Brian Roughan MBA ’05, Jennifer Andrews Burke MBA ’05, and Vikram Sahney MBA ’05 conducted market research, wrote the business plan, worked on business development, and pitched the company to investors.

Mentors from MIT’s Venture Mentoring Service (VMS) — including the late Dave Staelin, who founded the VMS — also guided LiquidPiston’s growth, offering advice on product development, hiring, and seeking venture capital. (So far, the company has earned more than $15 million in funding.)

In 2006, after analyzing dozens of engine iterations, LiquidPiston earned a $70,000 military grant to produce an initial diesel-engine prototype. (Today, LiquidPiston has analyzed and patented about 60 different engine designs to embody HEHC.)

Due to overwhelming feedback from power equipment manufacturers — calling for lighter, quieter, vibration-free engines — LiquidPiston recently pivoted to the X Mini, which it developed and released in the last six months. The company has now received interest from potential customers, and is speaking to engine manufacturers interested in licensing the X Mini technology.

“In addition to improving existing engine applications,” Shkolnik explains, “the X Mini may enable entirely new applications not currently possible with current engine or battery technology.”

Early next year, the company plans to host a competition to solicit ideas from the public surrounding these new uses for the X Mini. “We want to get the creative juices flowing and open up to the wider community to see if there’s something interesting,” Shkolnik says.

Computers that teach by example

Computers are good at identifying patterns in huge data sets. Humans, by contrast, are good at inferring patterns from just a few examples.

In a paper appearing at the Neural Information Processing Society’s conference next week, MIT researchers present a new system that bridges these two ways of processing information, so that humans and computers can collaborate to make better decisions.

The system learns to make judgments by crunching data but distills what it learns into simple examples. In experiments, human subjects using the system were more than 20 percent better at classification tasks than those using a similar system based on existing algorithms.

“In this work, we were looking at whether we could augment a machine-learning technique so that it supported people in performing recognition-primed decision-making,” says Julie Shah, an assistant professor of aeronautics and astronautics at MIT and a co-author on the new paper. “That’s the type of decision-making people do when they make tactical decisions — like in fire crews or field operations. When they’re presented with a new scenario, they don’t do search the way machines do. They try to match their current scenario with examples from their previous experience, and then they think, ‘OK, that worked in a previous scenario,’ and they adapt it to the new scenario.”

In particular, Shah and her colleagues — her student Been Kim, whose PhD thesis is the basis of the new paper, and Cynthia Rudin, an associate professor of statistics at the MIT Sloan School of Management — were trying to augment a type of machine learning known as “unsupervised.”

In supervised machine learning, a computer is fed a slew of training data that’s been labeled by humans and tries to find correlations — say, those visual features that occur most frequently in images labeled “car.” In unsupervised machine learning, on the other hand, the computer simply looks for commonalities in unstructured data. The result is a set of data clusters whose members are in some way related, but it may not be obvious how.

Balancing act

The most common example of unsupervised machine learning is what’s known as topic modeling, in which a system clusters documents together according to their most characteristic words. Since the data is unlabeled, the system can’t actually deduce the topics of the documents. But a human reviewing its output would conclude that, for instance, the documents typified by the words “jurisprudence” and “appellate” are legal documents, while those typified by “tonality” and “harmony” are music-theory papers.

The MIT researchers made two major modifications to the type of algorithm commonly used in unsupervised learning. The first is that the clustering was based not only on data items’ shared features, but also on their similarity to some representative example, which the researchers dubbed a “prototype.”

The other is that rather than simply ranking shared features according to importance, the way a topic-modeling algorithm might, the new algorithm tries to winnow the list of features down to a representative set, which the researchers dubbed a “subspace.” To that end, the algorithm imposes a penalty on subspaces that grow too large. So when it’s creating its data clusters, it has to balance three sometimes-competing objectives: similarity to prototype, subspace size, and clear demarcations between clusters.

“You have to pick a good prototype to describe a good subspace,” Kim explains. “At the same time, you have to pick the right subspace such that the prototype makes sense. So you’re doing it all simultaneously.”

The researchers’ first step was to test their new algorithm on a few classic machine-learning tasks, to make sure that the added constraints didn’t impair its performance. They found that on most tasks, it performed as well as its precursor, and on a few, it actually performed better. Shah believes that that could be because the prototype constraint prevents the algorithm from assembling feature lists that contain internal contradictions.

Suppose, for instance, that an unsupervised-learning algorithm was trying to characterize voters in a population. A plurality of the voters might be registered as Democrats, but a plurality of Republicans may have voted in the last primary. The conventional algorithm might then describe the typical voter as a registered Democrat who voted in the last Republican primary. The prototype constraint makes that kind of result very unlikely, since no single voter would match its characterization.

Road test

Next, the researchers conducted a set of experiments to determine whether prototype-based machine learning could actually improve human decision-making. Kim culled a set of recipes from an online database in which they had already been assigned categories — such as chili, pasta, and brownies — and distilled them to just their ingredient lists. Then she fed the lists to both a conventional topic-modeling algorithm and the new, prototype-constrained algorithm.

For each category, the new algorithm found a representative example, while the conventional algorithm produced a list of commonly occurring ingredients. Twenty-four subjects were then given 16 new ingredient lists each. Some of the lists were generated by the new algorithm and some by the conventional algorithm, and the assignment was random. With lists produced by the new algorithm, subjects were successful 86 percent of the time, while with lists produced by the conventional algorithm, they were successful 71 percent of the time.

“I think this is a great idea that models the machine learning and the interface with users appropriately,” says Ashutosh Saxena, an assistant professor of computer science at Cornell University. Saxena leads a research project called Robo Brain, which uses machine learning to comb the Internet and model the type of common-sense associations that a robot would need to navigate its environment.

“In Robo Brain, the machine-learning algorithm is trying to learn something, and it may not be able to do things properly, so it has to show what it has learned to the users to get some feedback so that it can improve its learning,” Saxena says. “We would be very interested in using such a technique to show the output of Robo Brain project to users.”