A set of vital societal functions such as health and safety are necessary for today's society to function and to secure the life of its individuals. Infrastructure is required to provide and maintain these functions. This for society critical infrastructure includes electronic communication technology, transport systems, oil \& gas supply, water supply, and the supply of electric power. The electric power system plays a central role in the critical infrastructure since it is required to operate all others. Therefore, outages in the power system can have severe consequences not solely for the supply of electricity but also for the supply of water, gas, and food. To provide a reliable and safe power supply, power system operators are applying asset management strategies to investigate, plan, maintain, and utilize the system and its components while improving the performance under its own financial constraints.

One approach to increase the reliability of the power grid while decreasing costs is maintenance planning, scheduling, and optimization. To optimize maintenance, a reliability measure for power system components is required. The failure rate, which is the probability of failure in a predefined interval, is utilized in maintenance optimization. Thus far, an average failure rate has been assigned to all components of the same type due to a shortage of component failure data. However, this limits the accuracy of maintenance techniques since the component heterogeneity is neglected. Moreover, the actual failure rate is being underrated or overrated and it is a challenge to identify the impact of conducted maintenance tasks.

This thesis presents how the failure rate accuracy can be improved despite limited failure data available. Firstly, an introduction to failure rate modelling theory, concepts, and definitions is given to provide a common understanding for the later chapters and papers. Secondly, regression models are presented which can be used to model, predict, and characterise the failure rate and failure intensity for power system components. The Cox regression and regression models for count data are applied to two case studies of disconnector and circuit breaker failure data. The results contribute to an improved modelling of the failure rate on individual level but also improve the understanding of risk factor's impact on component failures. However, the aforementioned regression models have rarely been applied in the power system domain due to the limited failure data. Thirdly, the necessity to distinguish between population and individual failure rates is illustrated and risk factors and methods are presented, which are frequently used in failure rate modelling. Moreover, the thesis presents a method to calculate and predict individual failure rates despite the occurrence of actual failures which is of particular advantage for new components.