A knock model for 1D simulations accounting for cyclic dispersion phenomena

Control of knock phenomenon is becoming more and more important in modern SI engine, due to the tendency to develop high boosted turbocharged engines (downsizing). To this aim, improved modeling and experimental techniques are required to precisely define the maximum allowable spark advance. On the experimental side, the knock limit is identified based on some indices derived by the analysis of the in-cylinder pressure traces or of the cylinder block vibrations. The threshold levels of the knock indices are usually defined following an heuristic approach. On the modeling side, in the 1D codes, the knock is usually described by simple correlation of the auto-ignition time of the unburned gas zone within the cylinders. In addition, the latter methodology commonly refers to ensemble-averaged pressure cycles and, for this reason, does not take into account the cycle-by-cycle variations. In this work, an experimental activity is carried out to characterize the effects of cyclic dispersion on knock phenomena for different engine speeds, at full load operations and referring to a spark advance of borderline knock. In each case, a train of 200 consecutive in-cylinder pressure traces is processed and the knocking cycles are identified through a standard FFT analysis, compared to an auto-regressive (AR) technique. The latter, proved to be more sensitive, is utilized to define the percentage of knocking cycles occurring in each operating condition, through the assignment of a proper threshold level. Then, a 1D model is set up to reproduce the above experimental pressure traces in terms of average IMEP and cycle-by-cycle variation. A kinetic sub-model is used to compute the heat released in the end-gas zone to be related to the knock occurrence. A new knock index is defined for each simulated cycle and its distribution is compared with the AR model outcomes. The above comparison proves a substantial congruence between the AR model-based knock detection methodology and the numerical one.