Real-Time, Adaptive Machine Learning for Energy Applications


Adam Vaughan, Ph.D.

Adam Vaughan, Ph.D.

For over a decade, Adam Vaughan has focused his efforts on the intersection of physical systems and computer control.  His work has included engine control electronics for custom race cars, robotics, electric motor control, and applying adaptive machine learning algorithms to complex, near chaotic engine combustion control.  Adam earned his Ph.D. in mechanical engineering from The University of Michigan, and master and bachelor of engineering under full scholarship from The Cooper Union in New York City.  


Critical need: Many potentially transformative energy technologies are hampered by complex control challenges.  New methods for modeling and control are needed to solve these challenges and ultimately help bring these technologies to market.

Technology vision: We believe machine learning control can enable many next-generation energy systems.  Our vision is to make it simple to connect machine learning to these systems, to control them with fast real-time software and hardware, and to do so at a low enough cost that the market readily adopts the technology.

Current state-of-the-art: Although we are currently targeting broader energy applications, our original focus was in the field of advanced engine combustion.  In this field, state-of-the-art techniques are able to predict the general behavior of combustion, but often struggle to predict complex combustion dynamics in real-world engines.  Conventional techniques are also challenged by the increasing complexity of modern engines.

Key innovation: We provide low-latency (10-100 microsecond) machine learning algorithms that adapt over time to predict and control complex physical systems.  Our technology is seamlessly integrated from the base hardware to low-level assembly code, all the way up to a high-level, easy-to-use user interface that runs within a web browser.  An early prototype of this technology has proven capable of both predicting and controlling near chaotic, stability limit engine combustion in sub-millisecond real-time.


Competing technology: Conventional 20th century control techniques often struggle with complex, nonlinear systems and are particularly limited on systems that are difficult to model.  Machine learning offers new possibilities for both modeling and control in the 21st century.

First market hypothesis: Data is king for data-driven machine learning algorithms, so our short-term focus is on a minimally viable product that provides a quality data acquisition system at low cost. 

Potential for impact: Beyond enabling better fuel efficiency in advanced engines, there is potential for our technology to make a meaningful impact in other energy-related applications.  These areas include low-density plasma stability control for fusion reactors, wind turbine control in turbulent wakes, real-time power forecasting of intermittent energy sources, chemical reactor control, and fast battery charging.  Given the technology’s broad potential in many energy-related applications, it could have a significant impact on climate change, energy, and the economy.

We're looking for: 

  • Technical collaborators
  • Team members - scientist, engineers
  • Joint development partners
  • Funders


Machine Learning, Combustion Engines and Real-Time Control
Raspberry Pi Learns How to Control a Combustion Engine

Contact: Adam Vaughan (adam [at] dauntless [dot] io)

News Feed: