coming soon

This page is currently under construction

Publications

REDUCED ORDER MODEL UPDATING

PhD Thesis (University of Bristol)

  • Laurence M. Griffiths
Across all engineering disciplines, the differences found between experimental results and computational simulations gives rise to various degrees of uncertainty in the solutions. In fluid dynamics these differences can be broadly split into issues of boundary conditions and numerical accuracy. Within the computational fluid dynamics (CFD) community a great deal of effort has been invested in reducing numerical error, yet large discrepancies with experimental data persist. The nature of experimental and computational studies often dictates the application of different boundary conditions applied in each. Further- more, for aerospace applications often both experimental and computational methods are attempting to model a free flying aircraft but doing so by applying fundamentally different conditions. Model updating provides the opportunity to modify the behaviour of the system to reduce these discrepancies. Initially this research concentrates on the update of reduced order models (ROMS). These models are a major area of research in CFD and promise the accuracy of CFD with much reduced computational cost. A novel framework is developed by which the steady state gradients of an unsteady eigenvalue based ROM may be updated. The new updating process is applied to remove tunnel wall interferences for Euler and RANS (Spalart-Allmaras) ROMS and to add the effects of viscosity to an inviscid Euler based ROM. Multistage updates are also demonstrated whereby a ROM is updated for both viscous and wind tunnel wall interference. A novel method is developed whereby the pulse input sizing for the production of ROMS from the nonlinear Euler and RANS equations equations may be automated. The method is proved accurate for a range of test cases. Finally a parameter study, investigating the suitability of a viscous-inviscid interactive model for updating, is performed. The study demonstrated that the equations in their original form are not suffciently robust for an automated model updating process. PDF

PULSE INPUT SIZING FOR CONSTRUCTING REDUCED ORDER MODELS OF THE EULER EQUATIONS,

In INTERNATIONAL FORUM OF AEROELASTICITY AND STRUCTURAL DYNAMICS, PARIS 2011 - IFASD-2011-195

  • Laurence M. Griffiths
  • Doran P. Jones
  • Micheal I. Friswell
Sizing the pulse magnitude for constructing dynamically time linear reduced order models can become a labourious process of trial and error for the aerodynamicist. Improper sizing of the pulse may lead to poor convergence or breakdown of the flow equations. In this paper we present a method to size the pulse input using classical one dimensional piston theory.

This work has been superseded the work presented in my PhD thesis.
PDF

MODEL UPDATING OF DYNAMICALLY TIME LINEAR REDUCED ORDER MODELS,

In INTERNATIONAL FORUM OF AEROELASTICITY AND STRUCTURAL DYNAMICS, PARIS 2011 - IFASD-2011-6

  • Laurence M. Griffiths
  • Doran P. Jones
  • Micheal I. Friswell
Here we present a novel approach to applying model updating to reduced order models (ROM) of computational fluid dynamics (CFD) codes. Differences in the boundary conditions and simplifications to the model often mean the steady state behaviour of the the underlying CFD scheme is incorrectly identified. By updating both the steady state magnitude and gradient of the ROM from higher order CFD or experimental data we can reduce these discrepancies. In this paper we present results of model updating of an unsteady aerofoil in an internal flow to a free flight condition.

My PhD thesis has significantly more improved methodology than that presented in this paper.
PDF

Source Codes

REDUCED ORDER MODELING TOOLBOX

PYTHON

  • Laurence M. Griffiths
  • Chris Wales
Toolbox for creating reduced order models used in my thesis. Code for restarting reduced order models is from chris wales request

PULSE INPUT SIZING DEMO SCRIPT

PYTHON

  • Laurence M. Griffiths
Example code to accompany the paper "Pulse input sizing for constructing reduced order models of the euler equations" Uses one dimensional piston theory request