Black Swift Technologies (BST) has unveiled an innovative artificial intelligence-driven solution aimed at identifying anomalies in unmanned aerial systems (UAS) and forecasting potential failures before they occur. This initiative is backed by a contract from the U.S. Air Force.
A report from the Center for Strategic and International Studies (CSIS) highlights that the U.S. Air Force’s Remotely Piloted Aircraft (RPA) or UAS serve as cost-effective, highly capable assets compared to conventional manned aircraft. The lower operational and maintenance expenses associated with UAS are critically important factors driving their increased utilization. Though, UAS face similar challenges as manned aircraft, notably concerning reliability. The USAF has reported numerous incidents where RPA/UAS have crashed due to pilot mistakes, mechanical issues, or electrical malfunctions. To enhance the detection of potential failures in critical systems of small UAS, the USAF has recently granted Black Swift technologies an SBIR Grant to create a machine learning software solution aimed at optimizing UAS maintenance schedules.
“failures in systems can incur significant costs—in terms of time, finances, and equipment,” states Jack Elston, Ph.D., CEO of Black Swift Technologies. “Our approach employs unsupervised learning for anomaly detection, utilizing algorithms that construct a model of expected aircraft behaviour across various missions and flight conditions, subsequently monitoring for deviations from these models.”
Enhancing the understanding of a UAS’s physical condition and its essential subsystems is crucial for ensuring the aircraft’s reliability and readiness for missions. Typically, UAS lack onboard monitoring systems or structured maintenance protocols. Many operators depend on limited guidance found in owner’s manuals to establish maintenance schedules. In contrast,manned aircraft benefit from extensive maintenance logs and schedules,while small UAS frequently enough lack detailed subsystem state information. Key components, such as servos, frequently operate in an open-loop manner without monitoring.
Certified technicians manage maintenance schedules for manned aircraft, significantly reducing the risk of equipment failure. Manned aircraft also enjoy the advantage of redundant systems and experienced pilots with extensive flight hours who can identify potential hazards, such as ice accumulation on wings. The absence of these safeguards in UAS can lead to costly failures, considering the value of lost vehicles, avionics, and payloads. More critically, such failures can pose risks to life and property, especially if they occur over populated areas or beyond visual line of sight (BVLOS).
“By harnessing artificial intelligence and machine learning, we can develop a more intelligent predictive maintenance schedule for UAS,” Elston emphasizes. “This will ensure that these UAS remain operational, safe for those on the ground, and always mission-ready.”
Mechanism of Action
BST’s solution employs unsupervised machine learning algorithms to provide early warnings and diagnostics for potential critical system failures in small UAS. Essential data is collected from existing avionics data gathered by the USAF, and if this data is insufficient, BST has created a series of monitoring nodes (Figure 1) that can be integrated into their proprietary avionics systems. These nodes enhance data collection and enable real-time analysis through machine learning algorithms.
Figure 1: BST’s modular network approach for UAS system monitoring
“We utilize a web-based platform that features a straightforward red/yellow/green diagnostic rating for each subsystem,” explains Elston. “Users can delve deeper into each subsystem for more detailed insights.”
BST’s user-friendly dashboard is designed for easy comprehension, eliminating the need for highly specialized UAS technicians to interpret the data (Figure 2).Drawing from their experience in creating custom modular solutions,BST aims to deliver decision-quality information that is accessible to everyday users.
Figure 2: Dashboard information display screenshot
an effective anomaly and maintenance tracking system will encompass three primary components:
Direct Failure Tracking: Most UAS are equipped with systems to monitor issues such as sensor failures,low battery levels,and interaction losses. This foundational aspect is crucial for maintenance and emergency responses to subsystem failures.
Supervised Learning: BST has been refining tools to monitor an expanding array of known failures using labeled telemetry data. This robust statistical approach can directly correlate onboard telemetry with specific subsystem failures or anomalies requiring maintenance. Examples include malfunctioning servo motors, damaged propellers, and adverse weather conditions like icing. The benefit of this method is that it can directly ground an aircraft and alert maintenance teams about necessary repairs or replacements.
Unsupervised Learning for Anomaly Detection: This emerging area of UAS research focuses on maintenance and anomaly tracking. BST has begun applying these techniques to specific subsystems,such as monitoring communication performance. However, the potential of this methodology is vast. When applied to the entire aircraft, unsupervised learning can identify maintenance needs based on performance deviations from expected norms.
