<p>Vehicles of today contain many advanced and complex systems, systems that make it hard for the mechanics working with them to keep an overview. In addition, new systems are introduced at an increasingly higher pace, which makes it hard or impossible for the mechanics to keep a both broad and deep competence. Consequently, to maintain a fast and efficient repair process, there is a need for computer-aided diagnosis.</p><p>In this thesis we develop a method for choosing the best "next action" in a repair process, using observations and a probability model. We describe the state of the system as a belief-state, a probability distribution over the faults that can occur on the system. An AND/OR-tree is used when searching for the optimal repair plan. We use entropy to speed up the algorithms. To avoid expensive validation actions, the system functionality is only inspected if the probability of having a fault free system is above a certain level.</p><p>The method is compared with two implementations of an existing method, with good results. The method can favorably be used on systems with many possible faults.</p>

Vehicles of today contain many advanced and complex systems, systems that make it hard for the mechanics working with them to keep an overview. In addition, new systems are introduced at an increasingly higher pace, which makes it hard or impossible for the mechanics to keep a both broad and deep competence. Consequently, to maintain a fast and efficient repair process, there is a need for computer-aided diagnosis. In this thesis we develop a method for choosing the best "next action" in a repair process, using observations and a probability model. We describe the state of the system as a belief-state, a probability distribution over the faults that can occur on the system. An AND/OR-tree is used when searching for the optimal repair plan. We use entropy to speed up the algorithms. To avoid expensive validation actions, the system functionality is only inspected if the probability of having a fault free system is above a certain level. The method is compared with two implementations of an existing method, with good results. The method can favorably be used on systems with many possible faults.