A Simple Physical Model of ICE Mechanical Losses 2011-01-0610

The current state-of-the-art offers two extremes of engine mechanical loss models: pure empirical models, using, e.g., regression models based on experimental results, and full-sized 3-D hydrodynamic friction models, solving Reynolds-like lubrication equations for complicated geometry of piston ring/cylinder liner or load-distorted shapes of crankshaft/connecting rod bearings and journals. Obviously, the former method cannot be reliably extrapolated while the latter is too complicated, especially for the early stage of design.

The aim of the current paper is describing the development and experimental calibration of the physical cranktrain model for FMEP prediction, based on simplified phenomenological model of mixed friction. The model uses simply defined shapes of Stribeck curves (friction coefficient) in dependence on Sommerfeld number, i.e., on effective sliding velocity, oil viscosity, dimension scaling factor and the normal force load. The piston rings are loaded by pre-tension and by pressure difference at external and internal surfaces. The instantaneous pressures at ring-pack are simulated using GT Power modules. Forces and torques in the cranktrain are calculated iteratively taking into account the additional friction effects and the non-uniformity of crank speed, reflected by inertia forces.

The model has been calibrated using data from engine tests at different loads, speeds and oil/cooling water temperature levels. The experimental FMEP was found by comparison of IMEP and engine torque. Top dead centre position of the piston was corrected according to the thermodynamic results. The calibration itself is done applying optimization methods for finding the best fit to experimental values, using different load and speed conditions. The sensitivity of model parameters to load conditions is tested in advance, yielding hints to the arrangement of experiments. The same approach is being applied to the valve gear and injection pump torque estimation.

The results show the predictive accuracy better than four percent considering mechanical efficiency.