Abstract
Failure detection in partially observable model-based Discrete Event Systems requires modelling failures as unobservable events within the system. Representing failures as events is not always realistic. For example, some classes of failure are in form of violations of constraints such as Service Level Agreement (SLA) and Quality of Service (QoS). These forms of failures do not represent events by themselves. They have to be modelled as additional events. Modifying the plant model is not always acceptable. Firstly, this may make the models large, causing extra computational complexity. Secondly, adding extra transitions is not always acceptable from engineers' perspective, because these constraints may change over the time leading to alternations of models every time these constraints are changed. To address this issue, this paper presents a new definition of diagnosability which extends the existing definition. In the new definition, a formalism has been introduced which captures failures as logical constraints instead of events. We show that starting from a Petri net, if the failure is expressed in Yen's logic, we can create a new Petri net with additional transitions, including transitions modelling failure, such that detection of violation of the constraint in the first Petri net is converted to diagnosis of failure in the second.