Speakers
Description
Salle A
The degradation of industrial systems is a natural and often inevitable process. Among the methods employed to estimate this degradation, the Hidden Markov Model (HMM) stands out for its wide application. This paper focuses on maintenance strategies for rotating machines, introducing a comprehensive framework for preprocessing methods alongside a novel adaptation of the HMM known as the Extended Multi-Branch HMM (EMB-HMM). To illustrate its efficacy, the FEMTO bearing dataset was specifically selected. Initially, abnormal signals, often identifiable by their pronounced noise in frequency zones, are pinpointed. Subsequent preprocessing steps are then executed. Moving forward, the EMB-HMM framework, distinguished by its four branches and five hidden states per branch, is applied. The determination of the active branch relies on both prior and posterior probabilities, with these probabilities and branch topologies linked to the four fault frequencies. Finally, the EMB-HMM serves as the assessment model, facilitating the evaluation of bearing performance degradation and the detection of the First Predicting Time (FPT) of initial degradation.
