Engine Controller using Implicit Fault-avoidance Learning of Control Parameters for Mixed-fuel Combustion

Kanako Esaki, Yuzo Shirakawa, Hiroto Naito, Kohsei Matsumoto, Kiyoto Ito, Takao Ishikawa

Abstract: An engine controller for mixed-fuel combustion has been developed that learns the control parameters while implicitly avoiding faults by setting the parameters so that thermal efficiency increases. In explicit fault-avoidance learning, the control parameters are learned while sensing faults such as abnormal combustion, misfiring and combustion fluctuation with sensors. Since the faults that can be detected with sensors can damage the engine, we used the fact that stable combustion is observed when the control parameters are set so that thermal efficiency is increased to develop an engine controller that autonomously determines the control parameters by alternately repeating parameter search and estimation. Application of this controller to an engine system operating on ethanol and hydrogen with the ratio of hydrogen increased in 2.5% increments from 0% to 30% (the engine's limit) resulted in a coefficient of variation in combustion fluctuation of less than 3%, demonstrating that combustion was stable and that faults were implicitly avoided.