Machine Learning Techniques for the Prediction of Combustion Events in Cooperative Fuel research Engine (CFR) at Homogeneous Charge Compression Ignition (HCCI) conditions. 2020-01-1132
This research assesses the capability of data-science models to predict the combustion events occurring for certain input conditions in Cooperative Fuel Research Engine (CFR) at Homogeneous Charge Compression Ignition (HCCI) conditions. The experimental data from CFR engine of University of Michigan (UM), operated at different input conditions for various gasoline type fuels was utilized for the study. The current study developed a capable machine learning framework to predict the auto-ignition propensity of a fuel under HCCI conditions. The combustion events happening at HCCI conditions in CFR engine are primarily classified into four different classes depending on the combustion phasing and pressure rise during the combustion in engine. The classes are: no ignition, normal combustion, high MPRR and early CA 50. Two machine learning (ML) models, K-nearest neighbors and Support Vector Machines, are compared for their classification capabilities of combustion events. Seven conditions are used as the input features for this ML models viz. Research Octane Number (RON) of fuel, Sensitivity of fuel (S), fuel rate (J/L/cycle), oxygen mole fraction, intake temperature and pressure, and compression ratio. The entire data set consisting of 10,395 cases corresponding to 15 different fuel blends, operated at various input conditions was divided in to three sets – 70% for training, 15% for validation and 15% for testing. The accuracy metric used in this study is micro-precision, ideal for multi-classification problem with the amount of data for different classes in data set is not uniform. A good accuracy/precision of 94% was obtained with SVM model. Finally, a sensitive analysis was performed to deduce the most effective input features on establishing the combustion event. This study enables the engine researchers to implement a design of experiments (DOE) beforehand, carefully targeting the right input conditions for a fuel to enable a controlled combustion phasing at HCCI regime.
Fabiyan Angikath Shamsudheen, Kiran Yalamanchi, Kwang Hee Yoo, Alexander Voice, Andre Boehman, Mani Sarathy
King Abdullah University of Science &Technology, University of Michigan, Aramco Services Co.