Temporal Diagnosis of Discrete-Event Systems with Dual Knowledge Compilation

Abstract

Diagnosis aims to explain the abnormal behavior of a system based on the symptoms observed. In a discrete-event system (DES), the symptom is a temporal sequence of observations. At the occurrence of each observation, the diagnosis engine generates a set of candidates, a candidate being a set of faults: such a process requires costly modelbased reasoning. This is why a variety of knowledge compilation techniques have been proposed; the most notable of them relies on a diagnoser and requires both the diagnosability of the DES and the generation of the whole system space. To avoid both diagnosability and total knowledge compilation, while preserving efficiency, a diagnosis technique is proposed, which is inspired by the two operational modes of the human mind. If the symptom of the DES is part of the knowledge or experience of the diagnosis engine, then Engine 1 allows for efficient diagnosis. If, instead, the symptom is unknown, then Engine 2 comes into play, which is far less efficient than Engine 1. Still, the experience acquired by Engine 2 is then integrated into the temporal dictionary of the DES, which allows for diagnosis in linear time. This way, if the same problem arises anew, then it will be solved by Engine 1 efficiently. The temporal dictionary can also be extended by specialized knowledge coming from scenarios, which are behavioral patterns of the DES that need to be diagnosed quickly. As such, the temporal dictionary is open and relies on dual knowledge compilation.

Twin‐engined diagnosis of discrete‐event systems

“…Moreover, the method in the aforementioned 44 for extending the open dictionary performs a determinization, whereas no such operation is carried out by the method described in this paper. Finally, the open dictionary generated by the algorithm Dictionary Extension in Section 5 is always a subgraph of the symptom dictionary, whereas this is not guaranteed by the method presented in the work of Bertoglio et al 44 The notion of a symptom pattern has been introduced in another work of Bertoglio et al, 45 where a method analogous to that presented in this current paper is adopted to expand the open dictionary. However, Bertoglio et al 45 addressed the task of diagnosis during monitoring and enforced this task to fulfill a so-called consistency requirement, whereas the task considered in the current paper is a posteriori diagnosis.…”

“…The notion of a symptom pattern has been introduced in another work of Bertoglio et al, where a method analogous to that presented in this current paper is adopted to expand the open dictionary. However, Bertoglio et al addressed the task of diagnosis during monitoring and enforced this task to fulfill a so‐called consistency requirement, whereas the task considered in the current paper is a posteriori diagnosis.…”

“…The notion of a symptom pattern has been introduced in another work of Bertoglio et al, where a method analogous to that presented in this current paper is adopted to expand the open dictionary. However, Bertoglio et al addressed the task of diagnosis during monitoring and enforced this task to fulfill a so‐called consistency requirement, whereas the task considered in the current paper is a posteriori diagnosis. Consequently, the dictionaries handled by the two papers are slightly different: a temporal dictionary is managed for diagnosis during monitoring, whereas a symptom dictionary is managed both here and in our different work …”

“…§ § This notion of a trajectory is specific to the task of a posteriori diagnosis of active systems; in fact, the constraint that a trajectory has to end in a final state is removed when the task of diagnosis during monitoring is performed, as in our other work. 45 One may also speculate that scenarios look like supervision patterns. 32 In fact, this similarity is only superficial because supervision patterns are meant to extend the notion of a fault to a relevant event that arises when a subtrajectory of the DES matches a given pattern, like in the works of Lamperti et al 48,49 As such, the relevant event is part of the candidate diagnosis.…”

“…In order to face this question, recent work by the authors has tried to combine the two mentioned approaches and to make them cooperate with each other, both for a posteriori diagnosis and diagnosis during monitoring: this paper moves a further step in this direction as far as a posteriori diagnosis is concerned. The main findings can be summarized as follows.…”