Date of Award

2017

Publication Type

Master Thesis

Degree Name

M.A.Sc.

Department

Mechanical, Automotive, and Materials Engineering

Keywords

Automated; Body Force Model; Expert System; Nelder-Mead; Optimization; Turbomachinery

Supervisor

Defoe, Jeff

Rights

info:eu-repo/semantics/openAccess

Abstract

Simulating the ingestion of non-uniform inflow to a fan or compressor requires enormous computational resources if the full details of the flow in the blade rows being studied is to be resolved, since full-wheel unsteady computations are required. A simplified modelling approach exists as an alternative computational option, which is the use of volumetric source terms (body forces) in place of the physical blades. Typically, body force models are manually calibrated with reference to single passage simulation results, and demands significant user experience and expertise. The objective of this thesis is to eliminate the need for experience and expertise during model calibration as much as is practical by employing an automated expert system. The modelling approach employed in this work is the combination of an existing turning force model, and an adaptation of an existing viscous force model. The automated system is implemented into Matlab and makes use of Ansys CFX as the flow solver. User input is required to initialize the system but the procedure then runs through to convergence of the final, calibrated model. Viscous force model coefficients that are traditionally found through an iterative procedure, are instead subjected to a Nelder-Mead optimization process. The machine studied as an example of the application of the automated technique is the NASA stage 67 transonic compressor. At peak efficiency, the isentropic rotor and stage efficiency, and the rotor work coefficient are matched within 1% of their single passage counterparts, a result that is on par with a manually generated body force model. A key finding in this thesis is that the stage efficiency is not the optimal parameter used for calibration of the stator's viscous force model. Despite this finding, the model produced performs sufficiently at off-design conditions not nearing choke. Across the speedline simulated, the model predicts the rotor total temperature ratio, total pressure ratio, and the stage total pressure ratio to within 1.3% of the single passage result. The computational time required for the calibration of the model produced from this work is 23 core-days. Although this computational cost remains relatively high, the removal of nearly all required user experience is achieved.

Download

Share

COinS