Fuzzy Learning Control of Rail Pressure in Diesel-Dual-Fuel Premixed-Charge-Compression-Ignition Engine

Abstract

Common rail systems have added a new degree of freedom in controlling diesel engines. A dieseldual-fuel, premixed-charge-compression-ignition (DF-PCCI) engine was modified from a diesel engine by injecting compressed natural gas (CNG) into the intake ports as the main fuel and injecting a smaller amount of diesel directly into the cylinders. The diesel injection timing was advanced to early in the compression stroke creating a mixture of diesel, CNG and air before being ignited almost simultaneously in the combustion chamber. The DF-PCCI engine had several modes of fueling; only diesel was used during idling, both diesel and CNG were used at low load with cylinder skipping, and both diesel and CNG were used with various energy replacement ratios during medium and high loads. As a result, a rail pressure set point was required to vary over a wide range and with a more abrupt change than that of a diesel engine mainly to obtain appropriate diesel atomization and to avoid excessive combustion. The rail pressure set point was also used as a factor in choosing the appropriate injection timing and duration during calibrations; therefore, it was necessary to track the set point of the rail pressure even more accurately. A novel rail pressure control system was presented based on fuzzy logic. One standard fuzzy system, having the tracking error and its integral as inputs, produced a necessary variation of the common-rail duty cycle to minimize the tracking error. The other fuzzy learning system, connected in parallel with the first fuzzy system, having engine speed and load as inputs, received this variation and used it to adjust centers of output membership functions to produce an appropriate mean value of the common-rail duty cycle to the engine. The fuzzy learning system’s rule-base was initialized from scratch, that is, with output membership functions centered at zeros. The rule-base can also be pre-programmed with the best human experience obtained during steady-state engine calibrations. A DF-PCCI engine, modified from a Toyota 2KD-FTV diesel engine, was connected to an engine test-bed. A new European driving cycle test was performed. Substantial improvement of the common-rail pressure tracking was observed during subsequent urban cycles because the fuzzy learning system was able to learn from the earlier urban cycle. Transient tracking results were also improved.