Automated Partitioning of Concurrent Discrete-Event Systems for Distributed Behavioral Identification

The aim of behavioral identification of discrete-event systems is to build, from a sequence of observed inputs/outputs events, an understandable model that exhibits both the direct relations between inputs and outputs events (i.e., the observable behavior of the system) and the internal state evolutions (i.e., the unobservable behavior). Since parallelism hinders the construction of monolithic models, distributed identification builds instead the models of subsystems. This paper proposes an automated partitioning of the system and optimal regarding the readability of the identified distributed models, thus fitting reverse-engineering purposes. To solve the optimization problem, a first solution is extracted from the observable behavior; then additional solutions are computed by agglomerative clustering. The approach is applied to a benchmark, resulting in an adequate functional partition.