Le système de surveillance (HUMS) installé dans les hélicoptères permet d'anticiper les anomalies et de donner la possibilité d'effectuer des tâches de maintenance prédictive avant l'apparition de défauts critiques. Par ailleurs, HUMS est également destiné à détecter la propagation de défauts émergents. Ceci consiste à comparer les caractéristiques vibratoires en vol de l'hélicoptère aux caractéristiques d'un état normal prédéfini. L'inconvénient majeur de cette approche est que les caractéristiques de l'état normal sont relatives au type de l'hélicoptère et changent après les tâches de révision et de maintenance, ce qui nécessite un réapprentissage de ces caractéristiques. Cette étude présente des méthodes d'évaluation de la progression temporelle des signatures vibratoires. L'étude de l'évolution de la signature vibratoire dans le temps permet de détecter des événements comme des interventions de maintenance ou des propagations de défauts sans avoir à définir un modèle de l'état de bon fonctionnement de l'appareil. Des méthodes fondées sur des modèles paramétriques et des bancs de filtres d'analyse vibratoire ont été testées et validées. Finalement, une méthode de détection de défauts a été mise en oeuvre et a donné de meilleurs résultats que les méthodes traditionnelles utilisées.

Les progrès scientifiques réalisés ces dernières années ont produit des bases de données de plus en plus grandes et complexes. Ceci amène certains classificateurs à générer des règles de classification basées sur des attributs non pertinents, et dégrader ainsi la qualité de classification et la capacité de généralisation. Dans ce contexte, nous proposons une nouvelle méthode pour l’extraction d’attributs afin d’améliorer la qualité de la classification. Notre méthode consiste à effectuer une classification non supervisée des attributs afin de retrouver les groupements d’attributs similaires. Une nouvelle mesure de similarité à base d’analyse de tendance est alors conçue afin de retrouver les attributs similaires dans leur comportement. En effet, notre méthode cherche à réduire l’information redondante tout en identifiant les tendances similaires dans les vecteurs attributs tout au long de la base de données. Suite à la formation des clusters, une transformation linéaire sera appliquée sur les attributs dans chaque groupement pour obtenir un représentant unique. Afin de retrouver un centre optimal, nous proposons de maximiser l’Information Mutuelle (IM) comme mesure de dépendance entre les groupements d’attributs et leur centre recherché. Des expériences réalisées sur des bases de données réelles et artificielles montrent que notre méthode atteint de bonnes performances de classification en comparaison avec d’autres méthodes d’extraction d’attributs. Notre méthode a été également appliquée sur le diagnostic industriel d’un procédé chimique complexe Tennessee Eastman Process (TEP). / Scientific advances in recent years have produced databases increasingly large and complex. This brings some classifiers to generate classification rules based on irrelevant features, and thus degrade the quality of classification and generalization ability. In this context, we propose a new method for extracting features to improve the quality of classification. Our method performs a clustering of features to find groups of similar features. A new similarity measure based on trend analysis is then designed to find similarity between features in their behavior. Indeed, our method aims to reduce redundant information while identifying similar trends in features vectors throughout the database. Following the construction of clusters, a linear transformation is applied on each group to obtain a single representative. To find an optimal center, we propose to maximize the Mutual Information (IM) as a measure of dependency between groups of features and the desired center. Experiments on real and synthetic data show that our method achieved good classification performance in comparison with other methods of extracting features. Our method has also been applied to the industrial diagnosis of a complex chemical process Tennessee Eastman Process (TEP).

In order to understand the dynamics of desertification after the 1980s droughts, the trends and changes of photosynthetically active fraction of the vegetation cover of the semi-arid region of West Africa have been analyzed across three ecoclimatic entities. This study focuses on six countries (Senegal, Mauritania, Mali, Burkina-Faso, Niger and Benin) that reflect quite well the human and environmental context of semi-arid regions. The main objective of this thesis is to check in the before-mentioned biophysical and human context, if the state of the vegetation improves because of more favourable rainfall conditions, or if, conversely, the long environmental deterioration during recent decades has a healthy dose of irreversibility for several years. The process of investigation proceeds in three stages defined by geographical scales and a geoclimatic gradient. At the regional level, we compared vegetation productivity data from 1982-1999 time series of NOAA-AVHRR NDVI data to rainfall data. We analyzed data from 315 rain gauges distributed across five countries (Senegal, Mauritania, Mali, Burkina-Faso and Niger) with annual average isohyets ranging from 100 to 900 mm. For trends analysis, we used the ratio of the integrated vegetation index (iNDVI) during the crop-growth period (June to October) to the May to October sum of rainfall (RR). This ratio (iNDVI/RR), a proxy of the Rain Use Efficiency, is widely used by ecologists as an indicator for detecting desertification processes. Overall, our results show a significant increase of the net primary production as a response of post-drought rainfall increase. However, the trends of iNDVI/RR ratio suggest that most of the studied stations (54%) in sahelian and sahelo-sudanese belts were stable and 31.4% showed strong to very strong negative change in iNDVI/RR ratio, while 13.3% of the stations showed a moderate decrease and only 1.3% showed a positive trend. At the country level, similar trends were recorded throughout 128 stations in Burkina Faso located between the 500 mm and 1100 mm isohyets. In fact, more than half the studied stations showed a stability of iNDVI/RR ratio. However, 39.8% of them show a negative trend from low to high, while only 2.4% of them show a slight positive trend. In addition, a comparison with more detailed local case studies confirmed these observed trends. However, the obtained results for wetter stations in the southern part of the country should be taken with precaution, as the relationship between NDVI and rainfall tends to weaken when annual rainfall is higher than 1000 mm. Overall, the negative trends highlight a gradual decline in plant productivity. These results recorded in 44.7% of the analyzed stations may reflect ongoing desertification processes in the sahelian and sahelo-sudanian zones over the past two decades. At the local level, a structural characterization of woody units in three bioclimatic regions of the sudanian zone (900 mm 1200 mm) in the north of Benin was conducted to assess their degradation status. We recorded the structural characteristics of stands (basal area, density), species diversity as well as disturbances type and intensity. Multivariate analysis revealed a gradient of productivity between the three regions: there was a high diversity of woody stands in the south-sudanian sector, while the north-sudanian and sudano-sahelian sectors were dominated by savannas and shrub, which had low productivity. The productivity gradient is influenced by a disturbance gradient suggesting that the decline in productivity is stronger from south-sudanian to the sudano-sahelian region. In addition, the spatial component of the observable changes in vegetation cover has been mapped by remote sensing in a restricted area of the sudanian zone in northern Benin with SPOT-XS data over the period 1986-2005. Over the past two decades, 19.6% of the woody stands have completely disappeared; 12.9% of this extinction of woody stands was due to deforestation, and 13.9 % due to degradation processes. In contrast to these trends, 21.8% of the study area were stable, while less than a third (31.7%) of the area were experiencing woody recovery (reforestation). The analysis also reveals significant disparities in the rates of change of the identified land use class units. These variations are more pronounced for the woody units and agricultural land than in villages. These results suggest that land cover degradation throughout the study area is primarily due to anthropogenic factors (livestock and agricultural expansion, logging, breeding). In fact, this area is a preferred destination for agricultural migrants fleeing the unpredictable climatic conditions of the drier semi-arid areas. Overall, our results highlight the rapid decline of vegetation resources, challenging assumptions that the impact of ongoing desertification processes is mixed, outside of the arid and semi-arid regions of West Africa. The developed framework is easily reproducible and allows the implementation of a reliable diagnosis on the state of the vegetation cover from accessible and inexpensive data. Its implementation should facilitate the development of managerial techniques for better assistance to the poor, especially vulnerable to the gradual degradation of their environment.