Fusion de données multi-capteurs pour le diagnostic des défauts des roulements

Abstract

In order to improve classification performance, data from multiple sensors can be merged to create more representative information. Compared to a single sensor data, multi-sensor signals always have redundant and complementary information that could be used for a better fault diagnosis. In this study, we propose to use indicators, calculated using signal processing techniques, from different sensors measuring non-proportional quantities such as motor phase currents and vibration signals. All indicators are merged to find an optimal subset of features which are then fed to a Support Vector Machine (SVM) classifier to perform bearing fault diagnosis.