The accuracy of the mean value engine model affected little by the accuracy of compressor mass flow rate model for a marine diesel engine

Abstract

The digital twin concept proves to be highly advantageous for smart ships as it allows the digital model to adapt in real time with changes in the physical ship. The mean value engine model (MVEM) is investigated with three different compressor mass flow rate models: a 2D interpolation model (2D), an experience-based M-JENSEN (MJ) model, and a physics-based Kang Song (KS) model. Each model is studied on a MAN 7S80ME engine with ABB A270 turbochargers. The root-mean-square errors (RMSE) for the test data are approximately 2.5 ${\mathrm{m}}^{3}/\mathrm{s}$ for the 2D model, 2.8 ${\mathrm{m}}^{3}/\mathrm{s}$ for the KS model, and 0.56 ${\mathrm{m}}^{3}/\mathrm{s}$ for the MJ model. For the training data, the RMSEs are around 0.0, 3.0, and 0.69 ${\mathrm{m}}^{3}/\mathrm{s}$, respectively. When coupling each model to the MVEM, it is observed that the calculated values differ only in the turbocharger speed. The RMREs under steady-state processes are 0.54% for MVEM with the 2D model, 1.63% for MVEM with the KS model, and 0.48% for MVEM with the MJ model. Under dynamic processes where the engine load is changed from 100% to 80%, the RMREs are 1.59%, 2.40%, and 1.58%, respectively. Among the studied models, the 2D model exhibits instability, the MJ model shows the highest accuracy, while the KS model performs the worst. However, all models demonstrate comparable results when used in the MVEM. Moreover, the physics-based KS model is a favorable choice for digital twin applications due to its excellent extrapolation ability and low dependence on measured data.