Abstract
Engine starting vibration level of hybrid vehicles has a variation in the same vehicle and there are various factors such as intake air, fuel, state of battery and backlash of gears. In analyzing the effect of these factors on the variation, it is necessary to solve two difficulties: multicollinearity caused by adopting conventional statistical analysis for large amounts of factors, and high measurement load for simultaneous acquisition of several types of time-series data. In this paper, we adopted machine learning for the factorial analysis and narrowed down the analysis target to control output of powertrain related to the exciting forces, which can be collected easily from ECU of engine and motors. Factors not found in the past investigation were found by random forest, which was a kind of machine learning and could clarify the contribution of the factors toward target variables. Load of learning huge amounts of the data was reduced by automation of analysis process. It was confirmed by simulation that the newly found factors were primary factors in the variation of engine starting vibration.
