Assessing An Aerospace Application Of Digital Twins For Multi-Agent Dynamic Decision Making
Author(s)
Marks, Ian
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
The concept of Dynamic Decision Making (DDM) is essential for achieving an overall goal by adapting to the results of previous decisions and unexpected environmental changes. Example applications of DDM in aerospace vary from individual predictive maintenance to multi agent tasking . When making dynamic decisions in a multi-agent scenario, the goal is to minimize uncertainty for future actions by predicting consequences for both the individual asset and the group. In a squadron with vehicles of the same type, it is expected that performance (e.g., fatigue rate and structural health ) vary form one vehicle to the next. Infusing individual performance capabilities and their uncertainties can overwhelm the decision maker. One approach to improve the decision-making process for multiple agents is by using Digital Twins, an authoritative virtual representation of a connected physical system. The digital twin’s aspects of computational, physical, and communications limits impact their overall utility. Furthermore, the aspects of fidelity, runtime, latency, and proximity (due to the physical requirements) need to be assessed to determine the value within multi-agent DDM. A vision for Digital Twins is to enable real time operational decision making by predictive and proactive measures while mitigating potential anomalies. This thesis seeks to evaluate the infusion of Digital Twins in a multi agent DDM architecture, the challenges with the infusion, and a comparison to historically deterministic decision-making processes for a relevant aerospace scenario to trade overall mission effectiveness. To that end, three steps are required: a method of evaluating different decision-making architectures, digital twin selection, and scenario definition. A structured decision-making process was developed such that both twinned and twinless multi agent DDM methods could be interchanged. The digital twin selected for evaluation was the airframe prognostic health of a remote-control aircraft. The digital twin determined how tightly a turn can be performed ( or ) as a function of health status mid-mission. A field surveillance/survey mission scenario was implemented with area surveilled as a metric. During the mission, each aircraft (twinned or twinless) defines their turn load, while a multi-agent coordinator modifies waypoints for agents. To ensure multi-agent interactions with DDM, a perturbance (treated as a gust event) occurs leading to one aircraft leaving the mission early and requiring the remaining aircraft to adapt their missions to mitigate the unexplored areas. Each aircraft leaves the mission area upon mission completion, digital twin health assessments or crashing. The assessment for permitting aircraft to leave the mission area is traded between the multi agent commander and by agents; both traded as a function of latency. Each agent has unique variations in both airframe life and digital twin architectures (instance vs aggregate) and are traded. The design of experiments enables trades across the agents factors of the digital twin fidelity (fit error with sensor to loads), initial health, and overall system latency. From the data generated, surrogate models were fit and analyzed to determine variable significance via ANOVA as well as a comparison between a turn only (treated as a twinless/human baseline) and various digital twin fidelities. Sensitivity analysis revealed that airframe life had the greatest impact on overall mission effectiveness among both digital twin-infused dynamic decision-making methods. Following closely was the influence of overall system latency, with digital twin fidelity being least important of the three. Additionally, the digital twin comparisons to human baseline show that digital twins significantly increase mission performance by longevity in the field as the entire fleet significantly ages. A simplified axiom for the digital twin’s infusion into multi agent dynamic decision making is as follows:
1) Having information is good (digital twin usage)
2) Having accurate information is better (digital twin fidelity)
3) Having information on time to make decisions is critical (data communication)
Sponsor
Date
2023-05-02
Extent
Resource Type
Text
Resource Subtype
Thesis