Analysis on Combustion Variations Using Neural Network with High Readability

Abstract

Combustion variation exists as one of the problems for improving the thermal efficiency of spark ignition engines. If the mechanism of combustion variation is clarified, effective control can be performed and it is expected to achieve higher thermal efficiency. this study aimed to elucidate the mechanism of combustion variation by constructing an IMEP prediction model for each cycle using the RF method which is a readable neural network. To evaluate the validity of the model, the model performance was verified by changing analysis target to a data set with no physical meaning, the physical equation with known the relationships, and the engine combustion data. Using the model, it is shown showed the factors that affect the cycle-by-cycle variation and the relationship between the factors, and it is clarified that the engine combustion may be influenced by the residual gas two cycles before, and is affected by intake and exhaust gas.