Iterative Learning on Dual-fuel Control of Homogeneous Charge Compression Ignition

Abstract

An Iterative Learning Controller (ILC) is used to control a dual-fuel Homogeneous Charge Compression (HCCI) engine. The engine is a CFR engine with in-cylinder pressure measurement ports and is operated at 100°C intake heating, 800 RPM and a compression ratio of 11:1. To control combustion timing and load, the amount of iso-octane and n-heptane injected into the manifold are used as inputs. The metrics used for combustion timing and load are CA50, crank angle when 50% of the fuel is burned, and gross IMEP, respectively. Using these inputs and outputs a system identification is performed and an ARMAX model is obtained. This model is then used to generate a norm optimal control. The norm optimal control is compared to a model-less control strategy that involves populating the off-diagonal of the learning matrix using a Jacobian estimate inverse. Both systems are used to follow a reference trajectory involving a step input in IMEP then CA50. The model-less control outperforms the norm optimal in both convergence speed and final iteration error. Application of non-causal filters within the iteration is also tested using a zero-phase filter and a Gaussian filter. The zero-phase has faster convergence than either the Gaussian or filter-less and has better final iteration error. This gives the best ILC control as model-less with zero-phase filter. This control is then compared with two PI controllers. It is found that the ILC outperforms the PI controllers after 3 iterations.