Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
IMPROVED HEALTH INDICATOR FOR LOW-SPEED BEARING FAULT DIAGNOSIS
Submission Author:
Thiago Barroso Costa , PA
Co-Authors:
Thiago Barroso Costa, João Lucas Lobato Soares, Elton Prestes de Souza , Jonatas Cruz da Silva, Walter dos Santos Sousa, Alexandre Mesquita, André Luiz Amarante Mesquita, Danilo Braga
Presenter: João Lucas Lobato Soares
doi://10.26678/ABCM.COBEM2023.COB2023-2304
Abstract
Vibration analysis and acoustic emission have been useful for condition monitoring and fault detection and diagnosis of low-speed slew bearings. Nevertheless, the use of vibration signal is a challenge due to low-speed bearings operate subjected to large loads and sometimes non-stationary condition. Moreover, the weak fault bearing signal can be smeared/masked by noise from other sources. Hence, processing signal tools and machine learning algorithms has been proposed to address those issues. Non-linear features have presented satisfactory results for condition monitoring and fault diagnosis of low-speed slew bearings. Among non-linear features, the maximal Lyapunov exponent (MLE) has showed clearer outcomes for low-speed slew bearing fault detection and prognosis. The Lyapunov exponent characterizes the rate of separation of infinitesimally close trajectories in phase-space. Since 1993, when the current method was proposed, the distance between those trajectories has been measured based on Euclidean distance. Aiming to enhance class separation, this work proposes the use of Pearson and Spearman distances as novel means to find the nearest trajectories improving MLE separability of classes. The Welch t test statistic value was used as a comparative index which is based on predictors’ separability. The methods were validated using vibration data from a controlled rotor test rig at shaft speed of 60 rpm, where healthy and damaged rolling bearings were tested. Finally, the results showed both Pearson and Spearman distances performed larger class separability, demonstrating that they can improve machine learning algorithm accomplishment for classification tasks.
Keywords
Low-speed bearing, Fault Diagnosis, machine learning, Feature Selection, Lyapunov exponent, Pearson distance, Spearman distance
