A Unified Mathematical and Algorithmic Framework for Managing Multiple Information Sources of Multi-Physics Systems

Project: Research project

Description

1. Identication of the research and issues Analysis and decision-making processes for complex, multidisciplinary systems often begin with low-fidelity models and progressively incorporate higher fidelity tools. In many cases, however, there is not just a set of computational models clearly ranked in terms of fidelity. Rather there are multiple models with different types of distortion of the true system. Moreover often the analysis and decision will involve other information sources, e.g., expert opinion or experiments. Typically, these different information sources are combined in an ad hoc manner. We propose to go beyond this current state of the art by creating therst set of fully principled approaches to analysis and decision-making for multi-physics systems that explicitly integrate the breadth of information sources. Existing multifidelity optimization approaches calibrate low-fidelity models or replace low-fidelity results using data from higher fidelity analyses. There is a need for new approaches that permit significantly more flexibility while maintaining mathematical rigor. We highlight three major challenges not addressed by current multifidelity approaches: (1) Man- aging a broad range of information sources (multifidelity models, historical, operational and experimental data, expert opinions) to select which next information sources to invoke and at what inputs as the decision process unfolds. Solving this problem requires a trade-off between cost and benefit: we must manage our limited resources (money, time, experimental facilities, computational power) while exploiting the respective strengths and accuracies of the information provided by the different sources. (2) Certifying overall analysis and design results, including assessment of risk. This requires new methods to define and assess the fidelity of information and to incorporate the associated uncertainties in the decision process. (3) Adapting to new decision goals, so that quantification of uncertainty and management of information sources are driven explicitly by the decision goals at hand, can adjust to dynamic system behavior, and can account for possibly changing resource constraints. These three challenges are further complicated in the setting of multi-physics systems, where different information sources relate to different parts of the system (e.g., to a single discipline, to multiple disciplines, or to coupling among disciplines), and where there are choices for how to couple (or decouple) the various aspects of the system. New formulations and methods to address these challenges will lead to better decisions, as well as improvements in the decision process itself, achieved through harmonization of multiple information sources, systematic ways to incorporate existing knowledge and expertise, and better management of risk by quantifying uncertainty throughout the decision process.
StatusActive
Effective start/end date1/1/1512/31/20

Funding

  • DOD-USAF-AFRL: Air Force Office of Scientific Research (AFOSR): $1,037,435.00

Fingerprint

Physics
Decision making
Large scale systems
Dynamical systems
Aging of materials
Uncertainty
Costs
Experiments