ESSENCE

Project Overview 

Galois, in partnership with Rutgers University, Purdue University, and Charles River Associates, has developed the Exploration Service for Synthesis and Evaluation of Novel CPS Emergent designs (ESSENCE). This innovative project introduces a groundbreaking approach to cyber-physical system (CPS) design, where AI/ML tools and techniques drive rapid iteration, optimization, and simulation.

 The Challenge: CPS Design is Slow, Costly, and Too Limited in Scope

Traditionally, designing or modifying cyber-physical systems, such as aircraft, power plants, or medical devices, is cumbersome, slow, and costly. Exploring a given design space is typically a manual process, with human engineers mixing and matching various component combinations and testing designs one by one. 

But what if artificial intelligence (AI) and machine learning (ML) could lend a hand, making the design and testing process faster and more efficient while maintaining – or even improving – the quality and diversity of engineering solutions?

The Solution: Expanded Design Space Exploration with AI

In the ESSENCE project, part of DARPA’s Symbiotic Design of Cyber-physical Systems (SDCPS) Program, Galois and its partners developed a machine-learning-driven method and toolset for quickly iterating on novel unmanned aerial vehicle (UAV) designs.

With ESSENCE, Engineers can interactively select and modify UAV components such as propellers, wings, and motors, and receive immediate feedback on the impact of that change on the UAV’s performance. Alternately, ESSENCE can use its bespoke machine learning algorithms to rapidly evaluate potential UAV designs against desired performance metrics or mission-specific requirements, generating numerous design variations at the push of a button, each tailored to achieve specific goals.

Speedy Simulation

To fast-track the testing phase, ESSENCE incorporates advanced surrogate modeling, enhanced by Koopman Theory and deep learning. This innovative approach creates simplified computational models that quickly predict UAV system behaviors, effectively serving as a “virtual wind tunnel” and drastically reducing the time and resources required for design validation.  

The resulting software is called DLKoopman, an open-source Python package that can not only efficiently test novel UAV designs, but can perform simulations under theoretical conditions that are difficult or impossible to replicate physically. What’s more, these capabilities are generalizable to a widespread variety of systems found in physics, engineering, and beyond, differentiating DLKoopman from previous efforts to model complex systems and making the potential for practical applications nearly limitless.

Value-Add

• Rapid Iteration: Near-instant feedback allows for quick iterations, pushing the boundaries of UAV and other CPS innovation.
• Enhanced Performance Analysis: Immediate insight into how design changes affect UAV capabilities, including fuel efficiency, speed, altitude, and stability.
• Reduced Costs and Time: Machine learning and surrogate modeling minimize the need for expensive, time-consuming physical prototypes and simulations.
• Optimized Designs: Ability to generate UAV designs optimized for specific tasks or conditions, not readily achievable with traditional methods.
• Generalizable Applications: While the ESSENCE proof case focused on UAV design, our ML algorithm, design space exploration tool, and DLKoopman software package can be applied to any CPS.