Adaptive Surrogate Modelling & Optimisation (ASMO) of Problems Described by Real or Synthetic Data

Lead ResearcherDr Mauro S. Innocente

The aim of this project is to design a computational toolbox for the Adaptive Surrogate Modelling & Optimisation (ASMO) of problems described by data, where surrogate model refers to both meta-models and data-driven models. In the former case, they are emulators of other models whilst in the latter they can be viewed as surrogates for unknown mechanistic models. Thus, the ASMO toolbox would receive input data −either experimental or synthetic via High Fidelity (HF) models− and return meaningful optimal response(s).

While the ASMO toolbox may be applied to an unlimited number of real-world problems, the focus in this project will be on “optimal design” and “decision-making” within a comprehensive spectrum of applications spanning disciplines such as operational research; aerospace and civil materials and structures; automotive powertrains; and electrical engineering. In particular, we are interested in the following specific objectives:

  • Surrogate-based optimal design of civil and aerospace structures composed of complex engineering materials (e.g. composites, concrete, wood, soil) initially modelled using classical and multi-scale methods.
  • Optimal design of FRP-reinforced concrete beams from experimental data as well as from computationally intensive numerical models.
  • Data-driven modelling and optimisation in soil mechanics.
  • Optimal Management of Waterflooding in Petroleum Fields.
  • Decision-making in smart manufacturing.
  • Surrogate-assisted optimal design of power converters.
  • Best aircraft engine-frame matching.

Optimum management of Waterflooding in Petroleum Fields

Collaborators:
Prof. Silvana M. Bastos Afonso (Federal University of Pernambuco, Brazil)
Prof Johan Sienz (Swansea University)

  • High Fidelity (HF) Commercial Simulator (at Universidade Federal de Pernambuco, Brazil) used to simulate oil extraction from petroleum reservoir via waterflooding.
  • Latin Centroidal Voronoid Tesselation DoE used to train Kriging model offline.
  • Particle Swarm Algorithm used to find optimum of surrogate model.
  • Adaptive Constraint-Handling Techique.

Best Aircraft Engine-Frame Mathching

AVAILab actively participated in the InnovateUK & KTN Uncertainty Quantification & Management (UQ&M) Study Group with Industry in Liverpool 2017, leading the Set-Based & Multi-Objective Design work-package of a problem presented by Airbus titled “Climb-Cruise Engine Matching – An Optimisation Approach“. The latter involved Surrogate Modelling, Uncertainty Propagation, and Chance-Constrained Set-Based and Multi-Objective Optimal Design of Airframe-Engine Matching aiming for minimum noise, fuel consumption, and gaseous emissions.

%d bloggers like this: