Resilient UAVs Traffic Operation Using Fluid Queueing Models

Dengfeng Sun
School of Aeronautics and Astronautics
Purdue University

Xiao Wang
Department of Statistics
Purdue University

Abstract
In this project, we address the issue of resilient operations, specifically under weather uncertainties, in future Unmanned Aerial Vehicle (UAVs) Systems (UAS). We model the traffic of UAVs as fluid queues and study the traffic dynamics using three basic traffic modeling components, namely the single link, tandem link, and merge link. The impact of weather uncertainty is captured as fluctuations of the saturation rate in fluid queue discharge (capacity), while the weather uncertainty is assumed to follow a continuous-time Markov process. We view the resilience of the UAS as the long-run stability of the traffic queues and address the resilience by a designed strategy for the optimal throughput. We derive the necessary and sufficient conditions for the stability of the traffic queues in the three basic traffic components. Both conditions can be verified in practice, and the optimal throughput can be calculated via the stability conditions.

Distributed Resilience for Swarms of Autonomous Agents

Shaoshuai Mou
School of Aeronautics and Astronautics
Purdue University

Shreyas Sundaram
School of Electrical and Computer Engineering
Purdue University

Abstract
By working as cohesive whole in an optimal way under consensus-based distributed optimization algorithms, large-scale inter-connected autonomous agents can offer better intelligence and a wider range of operations than single monolithic systems. On one hand, large swarms are inherently robust against individual agent/link failures; on the other hand, high dependence on local coordination raises the possibility that the whole swarm may be impacted when a few vulnerable agents are compromised by attackers. One main challenge in achieving resilience in such large-scale swarms comes from the fact that each agent has limited capabilities (in computation, storage, communication, etc.) while the attacker may be intelligent (hard to identify, highly mobile, perform arbitrarily bad actions). In this talk, we will discuss methods towards addressing this challenge for consensus-based algorithms in a fully distributed scenario.

Accelerating Post-Event Data Collection and Analysis Using Artificial Intelligence

Shirley Dyke
Schools of Mechanical and Civil Engineering
Purdue University

Ilias Bilionis
School of Mechanical Engineering
(presented by Ali Lenjani)

Abstract
A post-disaster reconnaissance mission is a valuable opportunity for engineers to study and extract knowledge about structural integrity of buildings during natural disasters. A reconnaissance mission often consists of two phases, a preliminary survey and a detailed survey. The preliminary survey is conducted by driving in the field and manually collecting information to provide a rapid understanding of the overall building conditions in an affected community and to select buildings that require closer investigations during the detailed survey. However, currently the preliminary survey requires extensive manual processes for data collection, observation, classification, and documentation, which are time-consuming and labor-intensive. Recent advances in vision-based visual assessment evolving around artificial intelligence (AI) can improve and streamline the current processes of the preliminary survey. In this research, we developed AI-based data collection and analysis system that directly supports the preliminary survey in a rapid and efficient manner. First, we developed an automated technique to extract multiple view pre-event images of the buildings from Street View 360 degree panoramas. Second, we developed a probabilistic method to fuse the results of analyzing several images, using our CNN-based scene (image) classifiers, to generate ultimate decision regarding the attributes or condition of a building. We demonstrate and validate the proposed system using post-hurricane images collected after hurricanes Harvey and Irma (2017) by structural wind and coastal engineering reconnaissance teams from the Structural Extreme Events Reconnaissance (StEER) Network.