Bearing Fault Detection Using Beamforming Technique and Artificial Neural Networks

Abstract

The importance of predictive maintenance optimization has been recognized over the past decades. A relevant aspect in the process of machinery noise control is the proper identification of noise sources. Microphone-array-based methods are known as alternatives for noise source identification in machines. In this work, the “Beamforming” technique is used to visualize the directionality pattern of the noise emitted by a rotating machine and a study is presented to compare the performance of machine condition detection using different architectures of classifiers based on Artificial Neural Networks. Sound maps from a rotating machine are used as inputs to classifiers for two-class (normal or fault) recognition. The classifier is trained with a subset of the experimental data for known machine conditions and is tested using the remaining data set. The procedure is illustrated using data from experimental sound maps of a rotating machine. The effectiveness of the classifiers and the network training is improved through the use of the Karhunen-Loève transform on the sound map data.