Applying neural networks (NN) to the improvement of gasoline turbocharger heat transfer modeling

Abstract

Precise prediction of turbine outlet temperature is important for the design of after-treatment systems, two-stage turbochargers and exhaust energy recovery. An accurate turbocharger model is necessary for the prediction of turbine outlet temperature. Turbine and compressor models are important for the behavior of both engine and turbocharger predicting. An predicted of turbine outlet temperature using a related model is significantly different from the actual turbine outlet temperature due to the neglect of turbocharger heat transfer [1]. In this study, we developed an improved method for correcting a turbocharger simulation model by considering turbocharger heat transfer. The differences between the predicted and actual turbine outlet temperatures depend on three parameters: turbocharger speed, turbine expansion ratio, and the turbine inlet temperature. A neural network was used to correct turbocharger heat transfer based on these three parameters, and then it coupled with a one-dimensional (1D) turbocharger model in the GT-Power software. Further, this method was validated by performing additional experiments. The predictions of turbine outlet temperature were improved, the errors are less than 15 K when the turbocharger detailed geometry model and heat transfer coefficients is not clear, and the NN method is more easier than the traditional 1D turbocharger heat transfer model.