SEI Machine Learning Prototype Helps the Air Force “Fuel More Fight”

When the U.S. Department of the Air Force (DAF) fuels its aircraft, it feels pain at the pump at a large scale: 1.5 billion gallons of fuel annually at a cost of about $5.5 billion. Aircraft modifications, or interventions, that reduce fuel consumption by even a small amount result in significant savings. However, faced with voluminous and variable data from in-service sorties, estimating fuel rate reduction is a laborious challenge for DAF experts. The SEI developed a prototype machine learning (ML) model to estimate fuel savings from aircraft interventions.

The DAF applies fuel cost savings to improve its combat readiness, or “fuel more fight,” as those within the department say. The DAF asked the Defense Department’s Defense Innovation Unit (DIU) to investigate a data-driven way to optimize flights for fuel savings. At the DIU’s suggestion, the DAF asked the SEI to apply ML to flight data and derive fuel savings from aircraft interventions. This automation approach promised to speed an expensive process currently based on expert experience.

The SEI’s expertise in ML, data science, and Air Force operations made it the perfect choice for tackling this problem. “This work is a good example of the SEI’s applied research and prototype development for mission,” said Keltin Grimes, an associate researcher in the SEI’s AI Division and lead of the operational energy model project.

The SEI team recognized the challenges: collecting clean flight data, processing it into a usable format, overcoming data noise, and controlling for confounding hardware, software, and operational variables. Using common open source tools, the SEI team built and trained ML models and developed a prototype tool the DAF can use to specify aircraft type, input data from flights with and without aircraft intervention, and output the intervention’s estimated fuel savings down to 0.5 percent. Even reductions that small in a single fleet of aircraft could save millions of gallons of fuel per year.

After validating the tool against expert estimates of a real-world intervention, the SEI delivered the prototype in May 2024 and trained DAF personnel to use it. Since then, the DAF has been applying the model to various aircraft interventions.

Dr. Jordan Eccles, former data scientist at the DAF’s Operational Energy Program, said, “SEI’s effort dramatically accelerated Air Force Operational Energy’s efforts to analyze flight data recorder files to automate detection of fuel efficiency improvements. The tools developed by SEI demonstrated that machine learning can successfully identify very small improvements in efficiency that nonetheless have significant impacts on combat capability. Critically, cost savings identified with these tools can be recovered and reinvested in energy initiatives, a program expected to provide over $35 million for DAF initiatives in fiscal year 2024 alone.”

While the operational energy model is tuned for aircraft interventions, the data-driven approach could allow the DAF to identify other fuel-saving factors, such as flight location and performance parameters. Such techniques could be applied to any government or industry fleet that feels pain at the pump.

Accelerating Technology Adoption

The Office of the Undersecretary of Defense for Research and Engineering (OUSD(R&E)) has been fostering technology innovation initiatives, such as the Rapid Defense Experimentation Reserve (RDER), to rapidly develop solutions for warfighters. The SEI works alongside defense technology accelerators to support this research and development tempo.

The Defense Innovation Unit (DIU), which accelerates adoption of commercial and dual-use technology across the Department of Defense (DoD), is one of the SEI’s most frequent innovation sponsors. The SEI provides technical advising and support as DIU evaluates potential vendors to address DoD mission needs. The SEI also collaborates with DIU on original projects:

• Responsible Artificial Intelligence Guidelines in Practice, a 2021 framework for building AI systems that align with DoD AI Ethical Principles. In 2025 the SEI will co-author updates that cover test and evaluation frameworks, generative AI, and operationalizing assured AI.
• xView2 Machine Learning Competition, a 2019 competition for computer vision algorithms examining satellite imagery to assess disaster damage. The SEI is part of a 2025 project to develop the xView 2 program into a tool for warfighters to assess battle damage.

By partnering with innovation organizations, the SEI helps the DoD deliver solutions at the speed of relevance.

Photo: U.S. Air Force, Senior Airman Joseph Morales