Multiscale Stochastic Modeling, Conditioning, and Simulation of Rare Events

Abstract

This proposal presents new approaches for dealing with rare events in material systems. The realmoffailureisdescribedalongtwoscales: thescaleofmeasurement,andthescaleofnucleation, andfundamentalresearchisdevelopedforstatisticalconditioningofeitherofthesetwoscalesonthe other. Inparticular,newphysics-basedmodels,statisticalmodelswithmachinelearningalgorithms will be developed to address long-standing challenges in the prediction of rare-events relevant to failure. In the context of physics-based analysis, reduced-order models that area adapted to the predictionofextremeswillbedeveloped. Inthecontextofstatisticalanalysis, anovelmulti-scale di?usion process (the switching di?usion) will be used to connect behaviors at the measurement and nucleation scales. These processes will provide insight into the dynamics of observables on eachscaleandthemannerinwhichtheyrelate. Alsointhestatisticalsetting,adaptedpolynomial chaos representations will be developed that bridge scales, permitting explicit representation of the measurement scale as function of the nucleation scale. Finally, in the machine learning (ML) context, we will develop new probabilistic ML algorithms that are adapted to querying either of these two scales based on information from the other scale. This will be valuable for design and prognosis.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 21, 2022
Source ID
FA95502110015XX0

Entities

People

  • Roger Ghanem

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Southern California

Tags

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Distributed Systems and Data Platform Development
  • Theoretical Analysis.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Neural Networks