Abstract

The use of algorithmic optimisation techniques whereby several designs are evaluated automatically in
batches using Computational Structural Mechanics (CSM) and or Computational Fluid Dynamics (CFD)
modelling to improve performance, has become an integral part of turbomachinery design process.
Designing radial compressors for better performance as well as manufacturing the impeller such that there
are no discrepancies between the designed surface and machined surface represents a significant
challenge for the industry. Accounting for geometric variability (to model manufacturing errors) during
the design/optimisation phase where hundreds of candidate geometries are evaluated is costly due to the
large number of calculations required to analyse the possible combinations of manufacture errors for each
new geometry design. This thesis addressed the problem by separating the design phase from the
manufacture uncertainty calculations phase, focusing on a common 5-axis milling type error – undercut;
and using a low cost high throughput computing cluster to meet the computational requirements of both
phases.
A bespoke parametric CAD algorithm was developed to automate the geometry creation during the
optimisation phase. The Differential Evolution for Multi-Objective Optimisation (DEMO) algorithm was
used to drive the optimisation calculations. In-house meshing software from Napier Turbochargers Ltd,
subsequently referred to as Napier, was used to mesh the computational domain, which was then solved
using a commercial CFD solver. The compressor in the high-pressure (HP) stage of a two-stage
turbocharger was optimised, and shows significant improvements in measured parameters - up to 1.6
points of efficiency gain and 20% increase in map width, respectively. The calculations were carried out
on a HTCondor cluster of 8 Linux workstations.
Moreover, a study on the sensitivity of radial compressor aerodynamic performance to the presence of an
undercut on the impeller surface was also carried out. In-house software from Napier was used to create
an undercut on the impeller surface by modifying the CAD geometry file. The impact of the undercut on
performance was quantified using detail 3D CFD analysis. Various undercut height and width levels at 13
different locations on the blade surface were analysed for three compressor designs. A unique sensitivity
distribution for each compressor impeller is calculated and used to create a variable tolerance map on the
impeller surface. This approach was shown to facilitate savings in cost by reducing scrap rate.
In addition, a bespoke 1-D algorithm for estimating the size of a radial compressor impeller required to
meet a design mass flow and pressure ratio at a given rotor speed was developed. The model can be used
PhD Thesis – O. F. Okhuahesogie Page 5
as a preliminary tool when designing a new compressor (where there is no previous experimental or
numerical data). The algorithm is based on a combination of fundamental turbomachinery physics
equations, correlations extracted from literature and statistical modelling.
Finally, an algorithm for calculating the flow area and air mass flow of the low pressure (LP) and high
pressure (HP) compressors and turbines in a two-stage turbocharger required to meet a diesel engine
specification was developed. The algorithm was used to validate the flow specification of a two-stage
turbocharger for a test diesel engine.