Identification of partially observable discrete event manufacturing systems

This paper deals with black-box identification of discrete event manufacturing systems that are automated using a programmable logic controller (PLC). The behavior of the system is observed during its operation and is represented by a single long sequence w of observed input/output (I/O) signals vectors. The identification method, conceived for addressing large and complex industrial systems, consists of two complementary stages; the first one obtains, from w, the observable part of an interpreted Petri net (PN) model composed of observable places and transitions describing the reactive behavior of the system. Afterwards, w is transformed into a sequence S of transition firings from which a PN model that reproduces S is inferred. This paper focuses on the second stage of the method in which a PN is built by adding non-labeled places and arcs that represent the non-observed behavior of the whole system by assuring the reproduction of w; this technique is based on discovering the causal and concurrent relationships between transitions in S.