Gaussian Mixture Model and Bond Graph tools for the elaboration of an efficient fault diagnosis procedure

Abstract

In this study, we exploit the advantages of an integrated approach built upon a model-based method (bond graph) and a data-based method (Gaussian Mixture Model) for the elaboration of a novel diagnostic procedure. The BG model is adopted for two purposes; firstly to verify the ability to diagnose the system components by investigating the structural/causal properties of the Bond Graph model. Secondly, it is employed for the creation of a database after transforming the Bond Graph model into a block diagram model. Furthermore, the Gaussian Mixture Model is implemented thanks to the created database for data clustering. For validation, the process has been assessed using a permanent magnet DC motor.