Model-based fault detection and diagnosis for the fuel system of a six-cylinder heavy duty diesel engine

Abstract

Nowadays, the fault diagnosis in modern internal combustion engines is becoming increasingly important. The constant development of engines, particularly in terms of fuel efficiency, and more stringent regulations of exhaust emissions, are leading to more complex systems. This enormous increase in complexity restricts the efficiency of conventional diagnosis systems, such as the limit checking of sensors. This thesis deals with the development of a fuel diagnosis system for a heavy duty diesel engine using advanced signal model- and process model-based methods. The diagnosis system has been developed for serial operation, which results in various limitations. One of these limitations is the lack of integration of additional sensors to monitor intermediate states. Furthermore, the low pressure pump, as well as the rail pressure control operate in closed loop control to ensure best possible results for the controlled variable. However, closed loop controls compensate for minor faults. In this thesis, physical models for various components in the fuel system have been developed which form the basis for model-based development. These models are used for developing algorithms to monitor the low pressure and high pressure pump, fuel filters, various leakages and the rail flow valve. Additionally, the model-based parameters generated additional information to help characterize the faults during the fault diagnosis. For signal model-based development, the frequency components of the periodic rail pressure and exhaust pressure signals were analyzed in high resolution. With this it is possible to extract algorithms to monitor injector flow and compression losses in internal combustion cylinders. This signal model-based methods provide additional information not only during fault detection but also during fault diagnosis, which allows the exact location of the observed fault to be determined. Residuals have been created using the developed algorithms, which represent a deviation of the system to be monitored from the normal state. Residuals formed using process and signal models have been created, which represent the inputs of the fault diagnosis. An inference based fault diagnosis was used to determine the fault characteristics, such as the type and location of the fault, in order to isolate these faults. The fuel diagnosis system was implemented, parameterized, tested and validated at the engine test bench. Various faults in the engine fuel system were generated as authentic as possible. These were detected by the process and signal model-based fault detection and the individual faults were identified and isolated. The tools for identification and isolation were fault-trees, which belong to the category of inference methods. Also useful for isolating individual faults are fault-symptom-tables that contain the full symptoms of all implemented faults. In particular, the method for isolating the injector flow faults provided an appropriate contribution to clearly identify the location of the fault. Finally, a strategy was developed for the online compensation of various injectors flow faults, which allows the short-term compensation of these faults. This ensures a reliable operation of the engine until the next workshop stay. In the commercial vehicle industry this is of great importance in order to minimize the additional costs of machine downtime.