Virtual Crank Angle based Cylinder Pressure Sensor

Sammanfattning: Closed-loop combustion control is an on-going field of research for improving reducing engine emissions and increasing efficiency. Cylinder pressure is a key parameter to monitor for combustion feedback. Measuring pressure with a transducer is an option, although being able to estimate the pressure based on the crank angle measurement instead would be beneficial in terms of costs. A virtual crank angle based pressure sensor was therefore developed within this thesis. It was studied how the in-cylinder pressure trace for a full closed cycle could be modelled from a pressure trace from a rigid crankshaft model, the angular velocity measurement and heat release modelling. The pressure trace from the crankshaft model was subjected to a singularity at TDC and torsional oscillations, it was therefore of interest to study whether the singularity could be avoided by modelling the heat release. Further on, the indicated work and total heat released during combustion were estimated from the angular velocity measurements as they are important parameters for determining the heat release trace.   It was found that the indicated work could be approximated by comparing the kinetic power trace, obtained from the measured angular velocity, with the piston power trace, estimated using isentropic pressure curves for the compression and expansion within the cylinder. Accurate results were obtained for operating points at 800 rpm while large deviations were seen for higher speeds as a consequence of larger torsional effect on the angular velocity trace; on the form of perturbed oscillations. The results could be improved from local averaging of the kinetic power trace at the occasions of deceleration, although it could be concluded that only the low speed operating were still accurate enough.   The kinetic power trace was attempted to be corrected for torsional power using angular displacement estimations of the crankshaft nodes from a dynamic crankshaft model. Even though the model seemed to capture the torsional behaviour at parts of the cycle, the oscillations could not be completely removed and it was determined that the final work estimate could not be improved from the torsional power estimate. The torsion was further studied regarding frequency and amplitude of the oscillations within the angular velocity and acceleration trace. No clear relations between the torsional behaviour and operating speed and load could be concluded. Further, since inversion of the dynamic crankshaft model for pressure estimation resulted in an improper solution since before, the model was iterated instead. The pressure trace could thereby be derived accounting for torsion, however the trace still contained oscillations which highlights the challenge of estimating the torsion accurately. The torsion is a complex phenomenon to describe and further development of a model for estimating the torsion with high accuracy for all operating points would improve the virtual pressure sensor significantly.   The heat release was, as a first step, modelled as isochoric and isobaric. These models gave information of the limits of SOC by comparing the indicated work from the resulting pressure trace with the work estimate from the angular velocity measurement. Further, one Wiebe function was parametrised such that the resulting pressure derivative during late combustion was adapted to the trace from the crankshaft model in a least-square sense. This allowed for better adaption as the partial pressure trace was subjected to torsional oscillations. The fitted Wiebe function described the diffusive combustion well but missed out the shape of the premixed combustion. Lastly, a double Wiebe function parametrisation was done where the diffusive combustion function was fitted to the late combustion data and the premixed combustion function was adapted such that the resulting indicated work matched the estimated work. To receive more accurate results, the premixed SOC and duration had to be approximated beforehand from the kinetic power trace. The virtual pressure sensor and most of the sub models were most accurate for low speed operating points. It was concluded that the reason is most probably the torsional effect on the input data to all sub models. It was shown that the crankshaft model can be complemented with heat release estimations which improved the final pressure trace and removed the singularity present around TDC.