Advanced combustion technologies such as PCCI (Premixed Charge Compression Ignition) combustion for diesel engines and HCCI (Homogeneous Charge Compression Ignition) combustion for gasoline engines have the potential to reduce fuel consumption and emissions. These combustion technologies basically depend on the auto-ignition of pre-mixture and have sensitivity for operation conditions, then precise control systems are required. On the other hand, the conventional control system uses control maps designed by the huge number of experiments. For advanced combustion technologies, further experiments are required to develop control maps to ensure the precise control of the advanced combustion technologies. Then, the model-based control is attractive on behalf of conventional control maps. The model-based control basically derives adequate control inputs for actuators such as injectors and EGR valves with the consideration of the operation condition. The authors developed a control-oriented model for advanced premixed diesel combustion with multiple fuel injections and the heat release of which shows two-stage and the heat release shape realizes lower combustion noise with high thermal efficiency. The developed model is based on physics as much as possible with several statistic sub-models. Even in the physical sub-model, they have several model parameters and the parameters also depend on operation condition and change for aging. In addition, the model structure is very simple for achieving light calculation cost to run on-board, the parameters include and absorb various influences. Then, such parameters are preferable to be adapted on-board with considering operation conditions. In this paper, an automatic adaptation method for the developed control-oriented model for the advanced diesel engine with two-stage combustion was developed. A neural network was employed for the adaptation of the ignition model and the combustion model, its optimization algorithm, selection method for learning data, proper input parameters were studied by using experimental data. The successive backward selection was performed based on candidate input parameters for the networks for the ignition model and the combustion model and preferable inputs for the networks are clarified. The data accumulated for a certain period of time was used as learning data and it was effective to perform mini-batch learning with randomly selected data while ignoring time order, and the model with the neural network adaptation shows better prediction accuracy than the model with the model parameters obtained by multiple regression. In addition, the network switching algorithm using multiple neural networks depending on the driving area defined by the throttle position was proposed and its prediction accuracy was better than the prediction result without the network switching algorithm.


Study on Automatic Adaptation for a Control-Oriented Model of an Advanced Diesel Engine

Division / Sponsors: 
Internal Combustion Engine Division (ICED)

Paper Proceedings: https://asmedigitalcollection.asme.org/ICEF/ICEF2020/volume/84034

ICEF2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.