Modeling of Fuel InjectionSystem for Troubleshooting

Sammanfattning: With the technology progressing further, making heavy duty vehicles more complex, more computerized, it becomes necessary to update the troubleshooting process of such vehicles. From the vehicles´ computers, diagnosis trouble codes can be extracted, informing the mechanic about the health of the vehicle. Using those codes, together with physical observations, as input for a diagnosing software, the software can give educated troubleshooting advice to the mechanic. Such diagnosing software requires a model of the vehicle or one of its system, which mimics the behavior of the real one. If there would be a one-to-one correspondence between observations and diagnosis, the model would be completely accurate. However, no such one-to-one correspondence exists. This makes the system non-deterministic. therefore the model has to be constructed using another approach. This master thesis presents a statistical model of a fuel injection system called XPI. The XPI-system is modeled using a statistical model called a Bayesian network which is a convenient way to model a non-deterministic system. The purpose of this model is to be able to make diagnosis of the XPI-system, but also to ease the understanding of the dependency between components and faults in the XPI-system. The model is also used to evaluate detectability and isolability of faults in the XPI-system.