Power systems are safety critical so that faults in such systems are rare. Consequently, faulty conditions typically cannot be used to learn patterns from. Additionally, operating conditions are varying a lot so that training samples captured over limited time periods may not be representative for the entire operating profile. However, if faults occur, they can have important consequences. Fault detection and isolation are, therefore, central tasks in system health management of power systems. Moreover, after detecting the fault, a decision needs to be made about fault mitigating actions. The talk will give insights into current challenges and potential solutions to unsupervised data-driven fault detection and isolation, and how decisions can be supported by multi-agent systems.