In previous work, static models were successfully applied to diagnosis tasks for moderately complex systems based on qualitative constraint equations [1] and the signed directed graph (SDG) [20]. However, underconstrained models used in these approaches caused combinatorial problems. Furthermore, system dynamics were not part of the model that was used for diagnosis, therefore, temporal feedback effects could not be dealt with, or had to be re-introduced on an ad hoc basis [5, 19].