Audine Subias

Process situation assessment: From a fuzzy partition to a finite state machine

Abstract Process situation assessment plays a major role in supervision of complex systems. The knowledge of the system behavior is relevant to support operators in their decision tasks. For complex industrial processes such as chemical or petrochemical ones, most of supervision approaches are based on data acquisition techniques and specifically on clustering methods to cope with the difficulty of modeling the process. Consequently, the system behavior can be characterized by a state space partition. This way, situation assessment is performed online through the tracking of the system evolution from one class to another. Furthermore, a finite state machine that is a support tool for process operators is elaborated to model the system behavior. This article presents theoretical aspects according to which the intuition that the trajectory observation of a dynamical system by a sequence of classes, to which the actual state belongs, gives valuable information about the real behavior of the system is substantiated. Thus, practical aspects are developed on the state machine construction and illustrated by two simple applications in the domain of chemical processes.

Classification Tool Based on Interactivity Between Expertise and Self-Learning Techniques 1

Abstract The present work proposes a learning methodology to identify process situations from measured data with a direct participation of the expert. The classification technique is based in LAMDA fuzzy method of conceptual clustering. The proposed tool has been developed in such a way that the learning phase (Supervised or not) is strongly associated with the expert knowledge. The application chosen for state identification and validation is the Universitat Politecnica de Catalunya fluidised bed gasifier selected as benchmark process in the CHEM European project.

Chronicle modeling by Petri nets for distributed detection of process failures

In this paper, distributed detection of process failures is studied. Erroneous evolutions, corresponding to failure situations, are described by chronicles. Every chronicle is composed of a set of events, and of a set of time constraints between these events. A chronicle can be decomposed into sub-chronicles distributed into several monitoring sites (systems). The objective of a monitoring site is to recognize a sub-chronicle. A specific reasoning considering relative time between event occurrences is proposed to provide time consistency of the mechanisms set Combining t-time Petri nets and p-time Petri nets, p-t-time Petri nets are used to represent these chronicles. When the Petri net reaches a particular state, a sub-chronicle is recognized. A symptom is detected when all the sub-chronicles associated to the initial chronicle have been recognized.

INTERMITTENT FAULT DIAGNOSIS: A DIAGNOSER DERIVED FROM THE NORMAL BEHAVIOR

Abstract This paper deals with an approach for the localization of intermittent faults in discrete events systems with partial observability. The proposed methods are based on a discrete events model representing the normal functioning of the observable behavior of the monitored system. This model based on automata formalism is built from the design data. The detection step consists of a comparison between the flow of observable events emitted by the monitored system and the flow foreseen by the model. A localization mechanism, based on diagnoser approach, points out the set of events potentially responsible for the faults. These two mechanisms are designed in order to operate on-board, in real time. An example from the automotive domain is presented.

Intermittent Fault Detection Through Message Exchanges: A Coherence Based Approach

Abstract: This paper deals with an approach for the detection and localization of intermittent faults in discrete events systems with partial observability. The proposed methods are based on a discrete events model representing the normal functioning of the observable behavior of the monitored system. This model based on the Petri net formalism is built from the design data. The detection mechanism consists of a comparison between the flow of observable events emitted by the monitored system and the flow foreseen by the model. A localization step complete the detection mechanism and points out the set of events potentially responsible for the faults. These two mechanisms are designed in order to operate on-board, in real time. An example from the automotive domain is presented. Copyright ©2006 IFAC

Iterative hybrid causal model based diagnosis: Application to automotive embedded functions

This paper addresses off-line diagnosis of embedded functions, such as that made in workshops by the technicians. The diagnosis problem expresses as the determination of a proper sequence of tests and measures at available control points, which would lead to greedily localize the fault quickly and at the lowest cost. Whereas anticipated discrete faults can be properly addressed by fault dictionary methods based on simulation, a consistency based method designed for hybrid systems is proposed to address parametric faults and non-anticipated faults. This method uses those same inputs as the fault dictionary method and the only additional information is the structure of the reference models in the form of a causal graph and the interpretation of the simulation results into qualitative values and events. The consistency based diagnosis method is combined with a test selection procedure to produce an original iterative diagnosis method for hybrid systems that reduces diagnosis ambiguity at each iteration. The method is illustrated in the automotive domain with a real case study consisting in the electronic function commanding the rear windscreen wiper of a car.

Learning chronicles signing multiple scenario instances

Chronicle recognition is an efficient and robust method for fault diagnosis. The knowledge about the underlying system is gathered in a set of chronicles, then the occurrence of a fault is diagnosed by analyzing the flow of observations and matching this flow with a set of available chronicles. The chronicle approach is very efficient as it relies on the direct association of the symptom, which is in this case a complex temporal pattern, to a situation. Another advantage comes from the efficiency of recognition engines which make chronicles suitable for one-line operation. However, there is a real bottleneck for obtaining the chronicles. In this paper, we consider the problem of learning the chronicles. Because a given situation often results in several admissible event sequences, our contribution targets an extension to multiple event sequences of a chronicle discovery algorithm tailored for one single event sequence. The concepts and algorithms are illustrated with representative and easy to understand examples.

Supervision Patterns: Formal Diagnosability Checking by Petri Net Unfolding

Abstract This paper addresses the problem of checking diagnosability of supervision patterns in discrete-event systems. With a supervision pattern, it is possible to represent a complex behavior of the system, and especially a faulty behavior. As opposed to classical diagnosability analyzers that check by exploring the marking graph of the underlying net, the proposed method relies on Petri net unfoldings and thus avoids the combinatorial explosion induced by the use of marking graphs. The method is an adaptation of the twin-plant method to net unfolding: a pattern is diagnosable if the unfolding representing the twin-plant does not implicitly contain infinite sequences of events that are ambiguous.

A Discrete Event Model for Situation Awareness Purposes

Abstract: In this paper we propose a framework for Situation Awareness (SA) of complex processes using data mining techniques and discrete event models. Referring to the well-known Endsley’s model of SA we present the different stages of our framework and show how this framework can be used for failure detection purposes.

Timed Diagnosability Analysis Based on Chronicles

Abstract Automated chronicle recognition is an efficient and robust method for fault diagnosis in timed discrete-event systems (TDES). This paper addresses the problem of diagnosability of TDES with regards to such a diagnosis method. We propose a fully automated chain to a priori check whether faults can be identified with certainty based on a given set of chronicles. To deal with the time aspects inherent to the chronicles, we first propose an automated translation of chronicles into a set of Labeled Time Petri Nets with Priorities. The diagnosability analysis is then performed on the state class graph of these nets and consists in determining whether the recognition of a chronicle is exclusive or not.