Diagnostics of advanced diesel fuel injectors

Ranjit S More, Purdue University

Abstract

During manufacture, variations in fuel injector dimensions result in variations in fuel metering. Thus injectors need to be calibrated at the end of the production-line and then matched with engines to achieve desired fueling characteristics. Over time, injectors can wear in, resulting in changes in fuel dosing. Since measuring injected fuel quantities can be inconvenient, various estimation techniques exist. The Virtual Sensor Coefficient method for injected fuel quantity estimation is based on the concept of Bulk Modulus. This concept uses the fact that outflow of fuel from the rail at every injection event leads to corresponding measurable pressure changes inside the rail. Consequently, the magnitude of pressure change can be used to predict the change in fuel flow and hence the amount of fuel actually injected during an injection event. The most crucial parameter in this concept is Rail Pressure. It is known that disturbances in the pressure signal lead to inaccurate measurements of pressure drop. This leads to inaccurate computations of the system bulk modulus and hence the injected quantity of fuel. This project explores ways to process the pressure signal appropriately so that the pressure drops corresponding to the injection events could be measured accurately. Three methods are proposed, implemented and their results compared. These methods are the two-window based dominant frequency method, filtering using IIR filters and de-noising using wavelet transforms. It is seen that the first method is the simplest method to implement relative to the other two methods while the results obtained after using the second and the third methods are relatively better. For higher rail pressures (2000-2600 bar) the de-noising method fares the best with consistent % error numbers of +/-4 over a large range of on-times. While for the same pressure range, the two-window based dominant frequency method, on the other hand, shows % errors that fall in the range of +/-12 for the same range of on-times. This project also identifies the issues that affect the implementation of the VSC concept. These are quantization and corruption of data due to pumping effects. These are more significant at the lower rail pressures and on-times. An algorithm is devised in order to detect the pumping affected data with an aim to 'weed-out' such data and work with only the good data so that the VSC concept could give acceptable results.

Degree

M.S.M.E.

Advisors

Meckl, Purdue University.

Subject Area

Mechanical engineering

COinS