Filtering Pressure Oscillations in a Common Rail Pressure Signal with IIR Notch Filters for Injected Fuel Amount Estimation

Sammanfattning: This project concerns an inline six-cylinder (I6) diesel truck engine equipped with common-rail injection system. It has been done at the Swedish truck manufacturer Scania CV AB. The main focus is digital signal processing of the rail pressure sensor signal from the injection system. A low-pass filter and a band stop filter of type infinite impulse response (IIR), both of order two, has been used to attenuate high frequency noise and pressure oscillations that are seen as unwanted information. The pressure oscillations occur in the fuel rail after an injection due to the water hammer effect. These have been analyzed in two different engines which showed that they vary depending on the engine used. An increase in frequency for the oscillations was seen when the rail pressure increases. It was also found that the oscillations from a specific injector had an almost identical behavior from one injection to another. It is suggested to use narrow band stop filters about the oscillation frequencies. A requirement of maximum one injection cycle was set for the phase delay of the filter to make it usable in real time. The phase shift could be kept significantly smaller without compromising any other filtering aspects. The resulting phase shift from the filters can be seen in chapter 5.4. The filter algorithm has been implemented with fixed point precision in MATLAB and in a microcontroller unit from NXP called MPC5777C. Digital hardware filters and direct memory access is used in the MPC5777C to avoid load on the central processing unit. In [1], it was advised that the sampling of the rail pressure signal needed to be at least 10 kHz to retain a high resolution of the injections. 20 kHz has been used in this project. An evaluation of the filtering through injected fuel amount estimation has been made based on the pressure drop in the filtered signal during an injection. The goal of the injected fuel estimation was to have a mean relative error below 20 %, as this was achieved in a similar study by Namir Askari and MathiasNiemand [2], which also utilized the pressure drop for the injected fuel amount estimation. This was achieved for two out of three tested injection amounts independently of the rail pressure. For the last and highest injection amount, it was above 20 % for most rail pressures.