Effective Alerting for Multiple Cascading Alerts

Open Access
- Author:
- Bannon, Joshua
- Graduate Program:
- Aerospace Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- November 21, 2024
- Committee Members:
- Amy Ruth Pritchett, Thesis Advisor/Co-Advisor
Eric Norman Johnson, Committee Member
Joseph Horn, Professor in Charge/Director of Graduate Studies
Jacob Willem Langelaan, Committee Member - Keywords:
- Aviation
Aerospace
Aircraft
Alerting
Dependency
Algorithm
Machine Learning
Pilots
Hazards - Abstract:
- Within aircraft alerting, extensive work was dedicated to developing industry best practices for handling individual alerts with existing electronic central aircraft management systems. These systems can provide the flightcrew with the necessary checklists for resolving alerts, present them with detailed component diagnostics, and prioritize alerts based on available aircraft sensor metrics. However, research is lacking on how to best handle multiple alerts at once, especially when alerts are inter-dependent for resolution. Existing aircraft management systems lack the knowledge and logic to understand dependencies between alerts. There have been instances where such alert dependency logic may have benefitted pilots in more effectively resolving multiple cascading alerts and reducing aircraft incident severity. This thesis includes machine reasoning about multiple potentially cascading faults and integrating this machine reasoning effectively as a support to pilots. An algorithm prioritizes and logically forms chains of dependency between alerts. These chains were communicated to pilots via a prototype interface that mimics existing electronic centralized aircraft systems. In an experiment with nine pilots, three different interface versions implemented with various levels of dependency communication to determine which was most effective at resolving multiple cascading alerts in three different scenarios with various levels of dependencies between alerts. The interface version with the greatest ability to communicate dependencies was most effective at reducing pilots’ time to resolve all alerts and the number of times they were unable to resolve an alert due to dependencies.