Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
STATISTICAL INDICATORS-BASED MACHINE LEARNING METHOD FOR CLASSIFICATION OF VIBRATION SIGNALS
Submission Author:
Gisele de Fátima Lima Camargo , SP
Co-Authors:
Gisele de Fátima Lima Camargo, Eurípedes Nóbrega
Presenter: Gisele de Fátima Lima Camargo
doi://10.26678/ABCM.COBEM2021.COB2021-0579
Abstract
This paper proposes a vibration analysis methodology, based on statistical indicators and Machine Learning algorithms, for feature selection, fault detection, diagnosis and localization. To assess the proposed methodology, it was initially applied to a rotating machine and then to a Lamb wave-based fault localization in an anisotropic structure. In the preprocessing phase, the statistical indicators are calculated from the signals and normalized. They are then analyzed through scatter and box plot diagrams. Features are represented by the calculated indicators, which are selected to best characterize the state of the respective monitored system, using unsupervised clustering in the hyperplane where the selected features are the coordinates. Each cluster is represented by its centroid. The centroids of each identified cluster, defined through an initial offline procedure, represent together the normal behavior of a system. Distances from the online acquired signals to the centroids are then calculated, in order to detect the eventual anomalies caused by faults. The methodology is first applied to the Case Western Reserve University rotating machine dataset and then to the experimental analysis of Lamb waves propagation in a flexible structure. For the rotating machine, an algorithm based on euclidean distance was developed for data classification. In the case of an anisotropic structure, a fully connected Artificial Neural Network was adopted to locate the faults, which also led to good classifications. The results suggest that the proposed methodology simplify the classification process yielding a good performance in both applications, permitting to expect successful application to monitor mechanical systems.
Keywords
Structural Health Monitoring (SHM), machine learning, statistical indicators, condition monitoring, Vibration Analysis
