Alexander Braafladt
(Advisor: Prof. Mavris]

will propose a doctoral thesis entitled,

A Reduced-Order Modeling Approach Supporting Accelerated Emergent Behavior Exploration Using Simulation in Military Capability Design

On

Monday, December 13 at 11:00 a.m.

In the
Collaborative Visualization Environment (CoVE)
Weber Space Science and Technology Building (SST II)

And

https://bluejeans.com/261189762/3438

Abstract
The capability design tasks ongoing through the Air Force Warfighting Integration Capability in the Capability Development Planning process are designed to support decision-making on investments and to identify and fill capability gaps in the forces expected to be operating in future conflicts. The focus in early capability design is on using constructive operational simulation to explore a large trade space of potential policy design alternatives - in terms of applications of emerging technologies and innovative concepts of operations. The analysis process in this context is analogous to an emergent behavior exploration process, and both have the goals of avoiding catastrophic problems and exploiting emergent opportunities. This prompts a framework for simulation-based exploration in capability design, and a central research problem of enabling this exploration at the speed needed for parametric, interactive, scenario-driven design exercises.

In response, an approach for reduced-order modeling (ROM) adapted to the use of operational simulation is proposed built up from techniques from engineering, statistical, and computer science applications. These are needed to handle the central dimensionality problems that arise in working with the key elements identified as needed for capturing emergent behavior - probability distributions and system element interaction dynamics. The first proposed adaptation involves developing a representation and order reduction combination that can encode an acceptably sized latent space to work with for the simulation output state from operational simulation. Specifically, a combination of histograms and principal component analysis is proposed. The second adaptation involves an alternative approach to direct Monte Carlo sampling from uncertainty propagation applied in engineering and reliability fields in order to deal with the expense of generating stochastic distributions of the outputs of operational simulation. Specifically, an adaptation of arbitrary polynomial chaos is proposed to meet the requirements of operational simulation. Lastly, once emergent behavior is identified using the leading, applicable methodology from the complex systems engineering literature, time series of the simulation state outcomes and variables are considered, and an approach to adaptive resolution in time using Mixture Gaussian Processes is proposed with adaptations for the discrete outputs expected from operational simulation.

A demonstration of the overall approach is planned using a contested reconnaissance scenario based in an open-source agent-based model. This demonstration is planned to directly showcase the added capability of the framework and ROM approach proposed in this work based on comparison to previous application of the simulation to a traditional surrogate modeling case.

Committee

• Prof. Dimitri N. Mavris – School of Aerospace Engineering (advisor)
• Prof. Daniel P. Schrage – School of Aerospace Engineering
• Prof. Mark S. Whorton – School of Aerospace Engineering (GTRI)
• Dr. Alicia M. Sudol – School of Aerospace Engineering