AbstractThe wide range of operating conditions and the multiscale nature of the physical processes makes numerical replication of in-cylinder fuel injection, mixing and combustion events a difficult task. In commercial engine development, these difficulties are compounded with limited computational resources and the requirement for fast development cycles. Combined, they are the main reason that for commercial purposes, simplified computational models with user-defined modelling coefficients are preferred over physically accurate methods without tuning dependency. Producing reliable in-cylinder simulations for quick turnaround engine development is a challenging task for modern CFD tools, mostly because of the tuning effort required for the sub-grid scale (SGS). Common practice is to prepare simulations for a baseline key point, tune the simulation to match available experimental data, and then use the tuned setup to predict other operating conditions or even engine configurations. The underlying assumption made when employing this approach is that all used sub models and their respective modelling coefficients are physically accurate and can therefore be representative for changing boundary conditions. This is however rarely the case.
This thesis covers four objectives. First, a combination of industry standard and relatively simple simulation sub models is selected, and their limits identified. Secondly, it challenges the common practice of matching a baseline experiment and swinging the boundary conditions with a fixed simulation setup. Next, a novel modelling coefficient table is developed that can match a wide range of idealised experiments from the Engine Combustion Network. Finally, the performance of simulations, whose settings are based solely on the novel modelling coefficient table, are tested selected on experimental data from two small-bore LDD DI Diesel engines at two load conditions.
The approach in this thesis shows that it has the potential to remove the necessity of lengthy and tedious tuning iterations by standardising and accelerating the simulation preparation. The modelling coefficient table is derived using a combination of Design-of-Experiment and stochastic process modelling. This statistical approach applied to a large range of operating conditions and a variety of computational models visualises the physical multivariable interactions between modelling coefficients and governing boundary conditions and lays the groundwork for novel auto-tuned and predictive in-cylinder simulations.
|Date of Award||2019|
|Supervisor||Konstantina Vogiatzaki (Supervisor), Robert Morgan (Supervisor) & David Mason (Supervisor)|