WVU engineer earns federal award for safety validation of AI-based avionic systems

Research Assistant Professor Ali Baheri has been awarded a grant to develop safety verification frameworks for learning-based avionic systems. (Submitted photo)

Artificial intelligence and machine learning have been adopted widely in the aviation domain in the last decade. Today, researchers are vying to overcome the challenge of establishing trust in AI-based systems that are largely responsible for life, property and the environment.

Story by Olivia Miller and Adrianne Uphold

Benjamin M. Statler College of Engineering and Mineral Resources

A researcher at West Virginia University has recently been awarded a $400,000 grant from the Federal Aviation Administration to focus on the design and implementation of a safety verification framework for learning-based aviation systems.

Ali Baheri, research assistant professor in the Department of Mechanical and Aerospace Engineering, explained that AI has shown promising success in avionic systems such as Airbus’ Autonomous Taxi and Take-Off Landing project. The project demonstrated a world-first in aviation — successfully achieving autonomous taxiing, take-off and landing of a commercial aircraft through fully automatic vision-based flight test using on-board image recognition technology, according to Airbus. 

“AI methods have unique characteristics that do not fit within existing FAA certification guidelines,” Baheri said. “Current certification policies are based on extensive testing and were not written with the needs and capabilities of learning-based autonomous systems in mind.”

According to Baheri, from the verification and validation perspective, this research offers feasible solutions for safety verification at various stages throughout the system deployment cycle, enabling the continuous learning and assessment of the system product. If successful, the new framework will accelerate the testing process and reduce the overall testing cost for certifying complex AI-enabled avionic systems. 

“We believe this framework provides the FAA a viable path to certify the AI-based avionic systems that possess unique features of unexplainably and mission environmental uncertainty compared to traditional systems,” Baheri said.

Baheri, from the Statler College of Engineering and Mineral Resources, will partner with Honeywell Aerospace and George Washington University to provide offline verification and runtime monitoring solutions to enhance overall safety of those systems.

“We are excited to make progress toward the challenging problem of AI safety in the air,” Baheri said. “We need to trust AI systems if we aim to employ them in safety-critical applications such as avionic systems.” 

---