L'analyse logique de donnees appliquee a la maintenance conditionnelle

Condition-based maintenance consists of planning maintenance tasks depending on the machine state.;We show that LAD's use to condition-based maintenance can be fruitful as it exhibits several advantages: LAD can be used on a great variety of classification problems, extracted knowledge is comprehensible by maintenance experts and LAD reacts well to signal noise and measurement errors. Most important, is the fact that LAD's classification accuracy is comparable and often superior to other classification methods. (Hammer, P.L. et al., 2000);The main objective of this thesis is the exploratory use of LAD to condition-based maintenance as well as development of a software application that allows automation of the analysis and classification of maintenance observations according to the aforementioned technique. To the best of our knowledge, LAD has never been used in the context of condition-based maintenance.;This research paper is organized as follows: In the introduction, the theoretical foundation of condition-based maintenance and of LAD is presented. In chapter 2, LAD' s different stages are described and illustrated with examples.;However this state is not directly observable and must be determined from indicators' measurements taken from the monitored machine at a given moment (observation). Logical Analysis of Data (LAD) is a data-mining technique that allows classification of observations according to the corresponding state.;The design and implementation of the different modules of which the software application is comprised, accompanied by diagrams and the corresponding algorithms are presented in chapter 3.;In the following chapter, two maintenance applications are studied using the software application that was developed. Finally under "Conclusions" we present a summarized view of the advantages as well as the challenges and limitations faced when building and using the software in the condition-based maintenance context. We also include a series of future research proposals which cover aspects that are not explored under the present research effort, as well as suggest ways to improve the performance and flexibility of the software that was built.