Multi-Layer Surrogate Modeling via Bayesian Approach and Non-Contact Full-Field Measurements

Abstract

US Air Force is seeking delivery of an ultra-realistically detailed model for next generation of aircraft, in which a tailnumber specific computational model of individual airframe is pursued, together with uncertainties quantified. At the same time, online-acquired heterogeneous data are aimed to be incorporated into executions. The integration of existing state-of-the-art measurement technologies to physical systems with gigantic computational models will enable a virtual sensing scenario, which provides a global awareness of the system, as well as prediction of the futureperformance.This proposed project aims to apply Bayesian framework to help a fast and accurate model selection in digital-twin paradigm. Bayesian framework is adopted in which the surrogate model is kept updated by means of evaluating posteriori conditional on observing evidence. A specific concept of multi-layer surrogate model design is proposed in this project, which partitions the entire symbiotic model into multiple hierarchical functionality layers, some of which only concern key characteristics and some critical substructures yet provide near-real-time response with way lesscomputation consumptions and feasible for embedded systems. In collecting data as the evidence to support the surrogate modeling, non-contact full-field pictorial measurements will be adopted. By investigating the dynamics extraction algorithm, the characteristics of the system physics will be identified as a means to facilitate physics-layer establishment.The dynamical updating process of the surrogate model is proposed to be combined with decision costs, future loadings, and possible failure modes, and concludes with a predictive/forecasting package, so that the functionality and reliability will be enhanced.

Document Details

Document Type
DoD Grant Award
Publication Date
Sep 19, 2018
Source ID
FA95501810491

Entities

People

  • Zhu Mao

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Massachusetts Lowell

Tags

Readers

  • Computational Modeling and Simulation
  • Distributed Systems and Data Platform Development
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference