Iso-surface sampler for developing high-dimensional engine controllers using high-fidelity CFD models

Developing controllers solely based on experimental testing is an industrious task as it is time-consuming and can be quite expensive. Instead, model-based controller development relies on a model, and hence the controller’s performance relies on the model’s accuracy. In the case of complex physical systems, physics-based high-fidelity models can be developed but cannot be directly used in controllers due to their complexity and computational overhead. Consequently, the use of data-driven models has garnered significant interest as they require minimal knowledge of the underlying physics for model development. However, these models heavily rely on the data based on which they are built. This paper discusses a method for developing reliable controllers by sampling data from high-fidelity computational fluid dynamics (CFD) models developed for internal combustion engine simulations. The developed iso-surface sampler (ISS) methodology samples data at high dimensions automatically in two stages, namely iso-surface generation and infilling. The framework and sampled data was then validated by performing simulations to develop a controller for the combustion phasing of a multi-fuel capable compression ignition engine.