Agnostic Multi-Fidelity Gaussian Process Regression for Modeling Complex Systems in Aerospace

Friday 10.9.2021, h.17.00
Sala Consiglio DAER – Edificio B12 – Second Floor, Via La Masa, 34 – Milano

Speaker: Dr. Giulio Gori

Download abstract >> pdf file
Download presentation >> pdf file

Multi-fidelity modelling methods leverage on the concatenation of data sets presenting enormous diversity in terms of information, size, and behavior. Pieces of information of diverse fidelity and complexity complement each other, leading to improved estimate accuracy. Multi-fidelity regression models bring clear advantages to the aircraft preliminary design phase, when data from computer simulations and preliminary experimental tests are exploited to define the best feasible configuration.
In a multi-fidelity setting, it is fundamental to establish the correct hierarchy in terms of data credibility w.r.t. the reality of interest. Unfortunately, the complexity characterizing aerospace applications generally makes the direct estimation of data fidelity difficult, if not intractable, leaving ground to modeling biases.
In this lecture, we focus on multi-fidelity co-kriging methods, developing an agnostic framework robust to prior modeling biases concerning the alleged fidelity of the available data sets.
It will be shown that this capability is particularly relevant whenever the Modeler does not select the appropriate ordering for the sequential construction of the multi-level surrogate. Moreover, we will show how the proposed formulation can also be exploited to obtain invaluable insights about the physics underlying the reality of interest.
Eventually, we also provide perspectives concerning the future deployment of the proposed methodology, with particular reference to Bayesian optimization methods and to the efficient construction of databases.

Share This Story, Choose Your Platform!