نظارت غیر مستقیم از شکست از یک سیستم تولید انعطاف پذیر تحت برنامه ریزی چرخه ای

We are interested in flexible manufacturing systems and more particularly in their well functioning. In such a context, we opted for a deterministic and cyclic command, which optimizes several criteria such as Work in Process and workflow. In order to supervise the system functioning, we propose to survey the system output and compare it to the deterministic schedule. We will give a diagnostic principle to identify the root cause of the detected symptoms.

In control architecture of an APS (automated system of production), the monitoring aims to identify the state of the system at any moment. It is an informational function (within the meaning of model OID —operative, information, decision— [LEM90]) which supplies data decisional modules within the functions of supervision and maintenance. Classically, the monitoring consists of two distinct tasks: detection and diagnosis. The role of detection is to determine any failure of the supervised system. Normally, the failures can be localized as well within the command or the operative part. In this paper, we will focus on the monitoring of failures of the operative part. Our approach of detection consists of making sure that the system evolves in accordance with the expected behavior specified by its designer. Any variation compared to the awaited behavior is interpreted as the proof of an abnormal operation. From a practical point of view, the role of detection is to determine these variations and to communicate them to diagnosis. The function of diagnosis is to analyze these variations in order to identify their causes. The causes are then communicated to the decision functions of the supervision and maintenance, which decide the procedure of recovery to set up taking into account dependability goals. This general outline of survey/reaction depends on the techniques of monitoring used. We can classify these techniques in two families called direct and indirect.

In this study, we have demonstrated that it is possible to monitor a 1-cyclic scheduling by limiting our observations to measure the delay of the parts at the output of the production system. In the b of the study context, we have limited ourselves to the hypothesis of a single failure. This limitation seems to be coherent with the classical study context of cyclic scheduling where it is generally admitted that the situations of breakdown are rare. If we agree with this hypothesis, it is clear that to have two faulty operations occurring on two distinct machines during two consecutive cycles is very improbable. So, our single failure hypothesis during two-cycle duration is realistic. However, in future works we will study how to manage the case of multiple failures. To understand the importance of this hypothesis on the diagnostic process, let us go back to the example given by Fig. 9. In the context of multiple failures, the system can have simultaneously self-failures W3/M2/OP13 and W4/M3/OP22. With the proposed resolution mechanism, we would diagnose failure W3/M3/OP13. A way to refine the diagnostic conclusions could be to reason on the delay of each operation. Indeed, in our example the two symptoms are characterized by the same delay ε. In case of multiple independent failures their delay would be different.