Fault diagnosis based on identified discrete-event models

Abstract

Fault diagnosis of Discrete-Event Systems consists of detecting and isolating the occurrence of faults within a bounded number of event occurrences. Recently, a new model for discrete-event system identification with the aim of fault detection, called Deterministic Automaton with Outputs and Conditional Transitions (DAOCT), has been proposed in the literature. The model is computed from observed fault-free paths, and represents the fault-free system behavior. In order to obtain compact models, loops are introduced in the model, which implies that sequences that are not observed can be generated leading to an exceeding language. This exceeding language is associated with possible non-detectable faults, and must be reduced in order to use the model for fault detection. After detecting the fault occurrence, its isolation is carried out by analyzing residuals. In this paper, we present a fault diagnosis scheme based on the DAOCT model. We show that the proposed fault diagnosis scheme is more efficient than other approaches proposed in the literature, in the sense that the exceeding language can be drastically reduced, reducing the number of non-detectable fault occurrences, and, in some cases, reducing also the delay for fault diagnosis. A practical example, consisting of a plant simulated by using a 3D simulation software controlled by a Programmable Logic Controller, is used to illustrate the results of the paper.

Fault detection of Discrete-Event Systems based on an identified timed model

“…(i) the TAOCT is an extension of the DAOCT model, which has been shown in [21] to be more efficient for fault detection than the NDAAO model, which is the basis for the TAAO model presented in [29];…”

“…(ii) the timing information added to the TAOCT model improves the capability of detecting faults of the fault detector system in comparison with the one obtained using the underlying DAOCT [21].…”

“…The fault detector proposed in this paper is inspired by the one presented in [21] based on the DAOCT model. Thus, as long as a sequence of timed events s t , whose corresponding untimed behavior s u = ψ(s t ) belongs to L Obs , is executed by the system, the fault detector observes the events, and plays the model following the behavior of s t .…”

“…Fault detection techniques based on an identified automaton model of the system have been proposed in the literature [16,25,20,21]. In these works, the two main ideas are: (i) to automate the process of obtaining the fault-free automaton model of the system by identification; and (ii) when a fault has been detected through a discrepancy between the system behavior and the model, to use a technique based on residuals for fault localization [26,21].…”

“…The use of the path estimation function reduces the exceeding language generated by the model in comparison with the NDAAO, and, consequently, it reduces the number of nondetectable faults. In [21], an algorithm for fault diagnosis using the DAOCT model proposed in [20] is presented.…”