Computational fluid dynamics (CFD) modelling of thermal propulsion system concept performance

Low-cost transport is one of the leading drivers for human development and routes out of poverty, but burning hydrocarbon fuels in internal combustion engine (ICE) vehicles results in various emissions which contribute to climate change and are toxic to humans. The challenge is therefore to develop affordable powertrain technology that can facilitate human development while minimising impacts on the environment and human health. Several recent studies have shown that a mixed-technology approach to the future fleet, which includes the use of hybrid-electric vehicles and alternative fuels for ICE vehicles, is the fastest way to save the most amount of carbon in a robust and sustainable fashion. Continued research into improving the ICE is therefore imperative for reaching national and international climate targets. A crucial part of the powertrain design process are CFD simulations, which provide opportunities to quickly optimise performance, diagnose problems, and provide explanations for observed experimental results. Characterised by complex multi-physics phenomena inherent in turbulent combustion, and spanning very broad space- and time-scales, CFD models of an ICE need to be carefully validated against measured data in order to produce reliable predictions. In particular, the present work answers two questions:
1. What constitutes a fair comparison between CFD results and experimental measurements?
2. When has a CFD model been sufficiently validated?
This work pioneers a novel use and interpretation of cutting-edge machine learning techniques to extract features from complex fluid flow datasets. Particular focus is placed on sparsity-promoting dynamic mode decomposition, which is a recent dimensionality reduction technique capable of correlating sections of a flow with specific frequencies. Therefore, steady structures can be separated from turbulent structures, which allows new insights to be gained from experimental data. The feature extraction capabilities are also robust against artificial diminishing of quantity magnitudes, which is a common issue with datasets relevant to propulsion research. This therefore enables the construction of fairer validation targets for the CFD models. The next step for the research is to use these tools to conduct a more in-depth investigation into important physical phenomena present in the flows, which will highlight key processes that need to be captured by the CFD as well as highlight any deficiencies in the model accuracy.
This project falls within the EPSRC 'Fluid dynamics and aerodynamics' research area, as is part of the EPSRC Prosperity Partnership - Centre of Excellence for Hybrid Propulsion Systems. Collaborators include Jaguar Land Rover, Siemens Digital Industries, and the University of Bath.