Is this project an undergraduate, graduate, or faculty project?
Graduate
Project Type
group
Campus
Daytona Beach
Authors' Class Standing
Michael Budihartono, Graduate student Francisco Bustamante, Graduate student Gabriela Gavilanez, Graduate student
Lead Presenter's Name
Michael Budihartono
Lead Presenter's College
DB College of Engineering
Faculty Mentor Name
Dr. Hever Moncayo
Abstract
This work proposes a comprehensive, adaptive framework for abnormal condition detection and flight envelope prediction in complex systems. The approach integrates data reduction techniques, including principal component analysis and lower-dimensional projections, to efficiently distinguish between nominal and abnormal operational states while maintaining computational efficiency through metacognitive interventions. A hybrid detection architecture combines bio-inspired real-value negative selection algorithms with support vector machines, augmented by generative machine learning models to synthesize failure data and support training through digital twin environments. Online fault trend analysis enables continuous monitoring of system degradation, while the system dynamically adapts to operational variations by minimizing offline training and incorporating learning during runtime. For identification, a dendritic cell-inspired algorithm based on danger theory is employed to infer affected subsystems from self/non-self projections, complemented by probabilistic models such as Naïve Bayes and nearest-neighbor methods for handling unknown failure patterns. The framework supports autonomous updating of the self/non-self space via incremental learning and decremental unlearning, enhanced through reinforcement learning and guided by metacognitive layer interventions for hyperparameter optimization. Additionally, the methodology introduces a generalized flight-envelope prediction mechanism that represents operational limits as a hyperspace of variables in reduced dimensions. This enables visualization and adaptation of safety margins under varying conditions, with runtime storage of abnormal condition patterns and associated constraints. The incorporation of dendritic memory cells further enhances prediction accuracy and responsiveness during repeated encounters, resulting in a robust, self-evolving system for fault detection, identification, and safe operation.
Did this research project receive funding support (Spark, SURF, Research Abroad, Student Internal Grants, Collaborative, Climbing, or Ignite Grants) from the Office of Undergraduate Research?
No
Included in
Aeronautical Vehicles Commons, Aviation Safety and Security Commons, Navigation, Guidance, Control and Dynamics Commons
Adaptive Health Monitoring for Runtime Safety Assurance of Advanced Air Mobility Applications
This work proposes a comprehensive, adaptive framework for abnormal condition detection and flight envelope prediction in complex systems. The approach integrates data reduction techniques, including principal component analysis and lower-dimensional projections, to efficiently distinguish between nominal and abnormal operational states while maintaining computational efficiency through metacognitive interventions. A hybrid detection architecture combines bio-inspired real-value negative selection algorithms with support vector machines, augmented by generative machine learning models to synthesize failure data and support training through digital twin environments. Online fault trend analysis enables continuous monitoring of system degradation, while the system dynamically adapts to operational variations by minimizing offline training and incorporating learning during runtime. For identification, a dendritic cell-inspired algorithm based on danger theory is employed to infer affected subsystems from self/non-self projections, complemented by probabilistic models such as Naïve Bayes and nearest-neighbor methods for handling unknown failure patterns. The framework supports autonomous updating of the self/non-self space via incremental learning and decremental unlearning, enhanced through reinforcement learning and guided by metacognitive layer interventions for hyperparameter optimization. Additionally, the methodology introduces a generalized flight-envelope prediction mechanism that represents operational limits as a hyperspace of variables in reduced dimensions. This enables visualization and adaptation of safety margins under varying conditions, with runtime storage of abnormal condition patterns and associated constraints. The incorporation of dendritic memory cells further enhances prediction accuracy and responsiveness during repeated encounters, resulting in a robust, self-evolving system for fault detection, identification, and safe operation.