Abstract— In this paper, an approach for fault diagnosis of Discrete Event Systems (DESs) is proposed. This approach builds a Boolean temporal model representing the normal and faulty behavior of a plant. This model is built by exploiting past events occurrences and their dates of occurrence. In addition, the proposed approach integrates a mechanism to detect a drift of the system functioning from normal to faulty conditions. This approach is illustrated and evaluated by using an example of a system composed of two independent tanks and equipped by discrete sensors and actuators. Keywords- Fault diagnosis; Discrete Event Systems; Temporal model-based diagnosis. I. INTRODUCTION HE task of diagnosis is to decide whether there is a fault in the system or not. In the case of a fault presence, the fault should be identified. Several diagnosis approaches have been proposed [1], [2], [3], [4], [5], [6], [8], [9]. They are generally based on the use of a model of nominal (desired) behavior G N of the plant and/or a model for each faulty behavior i F G in response to a specified fault { } 1 ,.., i d F f F F f Π Σ Π Π ∈ ⊆ = , where i F Π is the partition of faults associated with the fault label F i . Each fault partition corresponds to some kinds of faults in a plant element (e.g. sensor, actuator, a process component) which have the same effect according to either the configuration or maintaining procedure. f Σ is the set of all fault partitions which can appear in the plant. A fault is detected when the inputs u and outputs y are inconsistent with the nominal model. When a fault is detected, its identification is achieved by looking to a consistency between the measurements of inputs and outputs and one of the faulty models { } , 1,.., i F G i d ∈ . This consistency makes F i a fault candidate. A system can change its functioning mode from normal to faulty either abruptly or gradually. In the case of gradual change, the system begins to malfunction (degraded behavior) until the failure takes over completely. In this paper, we are interested on the diagnosis of both changes (abrupt and gradual) of the system behavior from normal to faulty. This paper is focused on qualitative model based approaches, in particular Discrete Event Systems (DESs) [7]. In DES, the system behavior is described by sequences of discrete events. There are several methods proposed in literature for the diagnosis of DES [1], [3], [4], [5], [6], [8]. This paper focuses on two popular methods to achieve diagnosis: chronicles [3], [4], [5] and templates [6]. These two approaches take into account time information in the model. A chronicle is a partial order of observable events constrained by time and associated to a particular faulty behavior while a template models the normal behavior. In these two approaches, the model is compiled off-line in order to minimize the work on-line. Then, the templates are used on-line to confirm the correct behavior of the system while the chronicles identify a fault belonging to a particular fault partition by on-line recognition. However, the templates and chronicles are generally built based on the human experience and their recognition is not a trivial task. In this paper, an approach based on the construction of Boolean temporal discrete event model is proposed to achieve the diagnosis. This approach builds off-line the normal and faulty behavior by exploiting the past observations about events and their dates of occurrence. These observations can be generated by simulation or are available in the history of the system function. The paper is structured as follows. In the second section, the proposed approach is detailed. Then in section 3, this approach is illustrated and tested using an example of two similar and independent tanks equipped with discrete sensors and proportional actuators. Some concluding remarks and the perspectives of the future work are provided in the last section. II. PROPOSED APPROACH Fig. 1 shows the various steps of the proposed approach. Each one of these steps will be detailed in the following subsections. Boolean Temporal Model-Based Approach for the Diagnosis of Discrete Event Systems N. Malki, M. Sayed-Mouchaweh Université de Reims Champagne-Ardenne, Centre de Recherche en STIC (URCA-CReSTIC), Reims, France {noureddine.malki, moamar.sayed-mouchaweh}@univ-reims.fr T 2011 IEEE International Conference on Automation Science and Engineering Trieste, Italy - August 24-27, 2011 ThB1.3 978-1-4577-1732-1/11/$26.00 ©2011 IEEE 175