ENBIS: European Network for Business and Industrial Statistics

Optimizing the robustness of a diesel engine’s combustion system

23 September 2009, 10:25 – 10:45

Abstract

Submitted by

Öivind Andersson

Authors

Öivind Andersson

Affiliation

Lund University, Dept. of Energy Sciences, Sweden

Abstract

This paper describes how robustness was optimized during the development of
the combustion system for a new diesel engine from Volvo Cars. The focus is on
the injector nozzle configuration which has a great influence on the combustion
process and, therefore, on the formation of exhaust gas emissions and on the
thermodynamic efficiency of the engine.
The intended function of the combustion system is to meet a number of
targets, e.g. legislation levels for emissions and customer demands on fuel econ-
omy. During the development process much effort goes into tuning factors in
the combustion system for meeting those targets. Examples of such factors are
hardware dimensions or settings in the engine control system. What is often
overlooked in development work is that there are also uncontrollable factors
affecting the system—noises. Examples of such factors are piece-to-piece varia-
tion in components and variations in the assembly process. If the system is not
robust to these variations it may enter failure modes, e.g. exceeding legislated
emissions levels. In this study, robustness was optimized mathematically using
transfer functions obtained using Design of Experiments (DoE). Noises from
the manufacturing and assembly process were put into these transfer functions,
and the resulting variation in the engine attributes was minimized by choosing
optimal settings.
The key measures of the system considered was the spray target position
(STP) and the Spray Angle. The STP is essentially the location where the
fuel sprays interact with the combustion chamber wall, and the Spray Angle
is the angle at which the interaction takes place. Analysis of the engine sys-
tems adjacent to the combustion system showed that the STP was affected by
14 measures of different engine components. To assess the effect of the natu-
ral variation in these components, production data was collected for the mean
production outcomes and the standard deviations of their dimensions.
Random samples of measures for all components of interest were obtained
by Monte Carlo simulation. For some components the suppliers measured 100%
of the produced items and scrapped those outside the specification limits. For
such components, the random data generated were truncated at the specification
limits, to mimic the scrapping process. Finally, the measures were combined as
they would be during engine assembly. The result was a sample population of
engines with estimated distributions of STPs and Spray Angles.
To treat robustness analytically a numerical robustness metric is needed.
Here, the distance from the failure mode (DFM) was chosen [1, 2].
A failure mode is defined as a state where the system fails to meet its intended function.
For instance, if the combustion system must comply with an emission standard,
its intended function is to meet the standard and the corresponding failure mode
is to exceed the standard. The distance to the failure mode can be defined in
terms of the mean performance (e.g. the population’s mean emission level) and
the specification limit (the emission standard). If the distance between these
is measured in units of the population’s standard deviation in emissions the
measure contains information about the system’s robustness. This is because it
estimates the fraction of the population that falls outside the specification limit,
i.e. which is in a failure mode.
A computer experiment
was designed to investigate how the DFM varied as function of STP and Spray
Angle. A response surface model was fit to the data and the system’s Distance from the Failure Mode could easily be maximized using multiobjective optimization.
[1] Davis, T.P., Applied Stochastic Models in Business and Industry, 22 pp
401-430 (2006)
[2] Davis, T.P., SAE Paper No 2004-01-1130 (2004)

View paper

Return to programme