A data-driven approach for predicting overall vibration severity of diesel engine

Abstract

With the trend toward higher boosting and lightweight design of diesel engines, the issue of vibration-induced failure of typical structural components has become increasingly prominent. Rapid and accurate prediction of overall engine vibration severity is key to evaluating and improving diesel engine reliability. To address the problems of low prediction accuracy of diesel engine vibration severity and insufficient vibration assessment capability for power density extension, this study comprehensively considers the combustion characteristics of cylinder pressure and dimensionless impact parameters. Multi-dimensional characteristic parameters highly sensitive to vibration severity (correlation coefficient > 0.8) are constructed and screened, namely average speed, cylinder pressure peak, maximum singular value of cylinder pressure, fourth-order cumulant of cylinder pressure, and kurtosis factor of cylinder pressure. Using the selected cylinder pressure and rotational speed characteristic parameters as inputs, a PO-SVR method for predicting diesel engine vibration severity was proposed. The prediction results for the test set samples are MAPE=3.04% and R2=0.98. The proposed vibration severity prediction method can provide guidance for rapid vibration assessment in the serial development of diesel engines.