Indirect measurements for control and diagnostics of IC engines

Abstract: The combustion process in internal combustion engines determines engine power, fuel consumption, exhaust emissions and combustion noise. Measurement and analysis of in-cylinder pressure plays an important role in the improvement and optimization of engine performance. Measuring pressure directly inside a cylinder is expensive and troublesome since an engine needs modifications to mount a pressure transducer and the harsh environment inside a cylinder will cause a transducer to have a short lifetime. An indirect, non-intrusive method based on external measurements has the potential to offer on-board control and diagnostics of combustion. This thesis focuses on cylinder pressure reconstruction and top dead center localization. For the pressure reconstruction two indirect approaches have been proposed in the literature; 1) engine structure vibration based reconstruction and 2) crankshaft angular speed based reconstruction. The crankshaft speed fluctuation has a high coherence with the cylinder pressure at low frequencies while structure vibrations have a high coherence for higher frequencies. Therefore, a new combined approach based on the two earlier approaches was developed. The combined approach was evaluated with two different methods; inverse filtering and neural networks. The proposed combined approach based on inverse filtering was compared to using either crankshaft speed or structure vibrations. The combined method gave the best results, due to combining the strengths of both methods in different frequency regions. The maximum pressure for the combined method was predicted with an RMS error of less than 5 bars (5%). A further development was to use a radial basis function (RBF) network, which combines a knowledge base with non-linear interpolation. The RBF network was able both to learn the training conditions and generalize this knowledge to untrained running conditions. This method can therefore be used generally, regardless of the actual running condition. For the untrained running conditions the maximum pressure was predicted with an RMS error of less than 3.5 bars (4%). The method based on a RBF network was also the method that gave the best agreement between reconstructed and measured pressure waveforms and was the only method in this thesis found to be able to predict unknown running conditions. Knowledge about piston positions is crucial when injection and ignition timing are set. In this thesis, a method to determine the top dead center from measurement of the crankshaft speed fluctuation was developed and evaluated. The TDC position can typically be determined with an error of less than 0.4 degrees.