A Neural-Symbolic approach to Real-time Decision-making in Complex Aerospace Systems
Abstract
The goal of the proposed research is to build a novel neural-symbolic machine learningframework for early detection of anomalies, degradations and faults in complex aerospace systemsand taking mitigating control actions in real time for enhanced safety, robustness and resilience.Recent advancements in deep learning shows that while neural approaches are excellent at lowlevelfeature extraction from multi-modal raw data without meticulous hand-crafting, such modelsstill may not be well-suited for logical reasoning, interpretation and domain knowledgeincorporation. On the other hand, symbolic approaches can potentially alleviate such issues as theyare shown to be effective in high-level reasoning and capturing sequences of actions. The mainhypothesis of the proposed research is - a hybrid neural-symbolic learning architecture willpotentially have transformative impact on the critical problem of monitoring and control ofcomplex aerospace systems. Upon detection of anomalies, the proposed research will developmachine learning-based control strategies for preventing catastrophic failure by enacting resiliencyfor recovery to a gracefully degraded condition.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 02, 2017
- Source ID
- FA95501710220
Entities
People
- Soumik Sarkar
Organizations
- Air Force Office of Scientific Research
- Iowa State University
- United States Air Force