Data-driven Modeling and Control for Multi-Fuel Compression Ignition Engine

Controlling the combustion phasing of a multi-fuel compression ignition engine in varying ambient conditions, such as low temperature and pressure, is a challenging problem. Traditionally, engine control is achieved by performing experiments on the engine and building calibration maps. As the number of operating conditions increase, this becomes an arduous task, and model-based controllers have been used to overcome this challenge. While high-fidelity models accurately describe the combustion characteristics of an engine, their complexity limits their direct use for controller development. In recent years, data-driven models have gained a lot of attention due to the available computation power and ease of model development. The accuracy of the developed models, which in turn, dictates the controller's performance, depends on the dataset used for building them. Several actuators are required to achieve reliable combustion across different operating conditions, and obtaining extensive experimental datasets across all these conditions can be difficult. This work proposes utilizing a dataset from a high-fidelity model such as CFD to build an approximate Gaussian Process Regression (GPR) model with the Variational free-energy (VFE) methods. The GPR model is then used for guided engine testing and controller development. Simulations using an experimental dataset were presented to demonstrate the improved tracking performance and a much-reduced number of experimental tests.