Diagnosis of concurrent and asynchronous systems, such as large telecommunication or information systems, requires powerful mathematical models. The use of Petri net unfoldings allows to formalize diagnosis using partial order semantics, a generalization from the global state model imposed by the use of automata. If, in addition to asynchronicity and distribution, the network topology itself is subject to dynamic changes, all static models, including Petri nets, reach their limits. Then, graph grammars can be used, encoding in the current local states not only the current values of state variables but also the current topology of the network connections; the fact that unfolding semantics is available allows to carry over the diagnosis algorithms to this setting.