Probabilistic Fault Isolation in Embedded Systems Using Prior Knowledge of the System

Abstract: Nowadays truck engines are controlled by an embedded control system which is a specially designed computer system. It is important that the embedded system is robust against possible faults that could appear when driving the vehicle since faults may cause the vehicle to stop unintended or even worse, crash. To prevent this tests in the system are designed to detect faults and further isolate the faulty behaviors for components. After the isolation fault tolerant control software is used to control the engine in the presence of the faults until the faults have been attended to by a mechanic. In this thesis an application of a probabilistic isolation method that ranks possible faults on their likeliness is presented. The method uses a Bayesian approach for the probability computations based on prior knowledge for ranking the faults in order to improve the result. The probabilistic isolation method is analyzed to show how the isolation performs and how the isolation differs when changing different parameters for the isolation such as test sensitivities and prior knowledge. Different solutions for problems that appear due to different circumstances are also described and evaluated. The solutions handle cases such as limited RAM and execution time, multiple faults and incomplete observations. The result shows a good performance for the probabilistic isolation method and the different solutions. However the method still needs further developments in order to achieve adequate trust for an implementation in vehicles. Future work is proposed and should include further improvements in the isolation of multiple faults.