This thesis addresses the problem of understanding change and reducing the work needed to estimate and respond to change in families of safety-critical embedded systems. Explicit family feature modelling techniques are developed that record the context within which a feature is valid for each family member. These features are combined with a description of their allowed variation among different members, to provide a complete family feature model.
These techniques are used to create a family feature model for a number of industrial projects. Comparisons are made between the ability of the family model and the project processes to accurately estimate change impact. Results show that the family model provides more accurate change impact estimation than the existing project processes. It also provides an understanding of the role of domain knowledge in impact estimation, a method by which different types of specification may be traced to one another throughout the development process, and a process by which individual feature descriptions are transformed into single family descriptions.