This paper proposes a study and a method for automatic detection and diagnosis of defects of rolling element bearings. We use classification techniques (QDC and neural networks) and classifier fusion. We exploit experimental data consisting of vibration signals represented in the frequency domain by means of the Fast Fourier Transform, registered by two accelerometers. We consider one defect, namely indentation on the roll, with three different severity levels, with the data related to the lowest severity level collected in four subsequent days. We achieve high classification accuracy in all the experiments, which aim, respectively, to identify the defects as soon as they appear, to identify the defects as time passes, to train the classifier on defects collected in the first day and test it on signals collected in the following days, and, finally, to analyze how a specific defect evolves over time. In particular, by analyzing how the vibration signals of a damaged bearing evolve over time, we observe that, as time passes, the signals representing the least severe damage get more similar to those related to the same defect but with a higher severity level. This study can be profitably used to define when bearing maintenance should be performed.
Time Evolution analysis of bearing faults
LAZZERINI, BEATRICE;
2009-01-01
Abstract
This paper proposes a study and a method for automatic detection and diagnosis of defects of rolling element bearings. We use classification techniques (QDC and neural networks) and classifier fusion. We exploit experimental data consisting of vibration signals represented in the frequency domain by means of the Fast Fourier Transform, registered by two accelerometers. We consider one defect, namely indentation on the roll, with three different severity levels, with the data related to the lowest severity level collected in four subsequent days. We achieve high classification accuracy in all the experiments, which aim, respectively, to identify the defects as soon as they appear, to identify the defects as time passes, to train the classifier on defects collected in the first day and test it on signals collected in the following days, and, finally, to analyze how a specific defect evolves over time. In particular, by analyzing how the vibration signals of a damaged bearing evolve over time, we observe that, as time passes, the signals representing the least severe damage get more similar to those related to the same defect but with a higher severity level. This study can be profitably used to define when bearing maintenance should be performed.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
