Bearing fault detection and diagnosis by fusing vibration data

This article presents a simple method for the detection and diagnosis of bearing faults, by fusing the information coming from two accelerometers. The method relies on three simple and intuitive features, extracted from the data coming from accelerometers placed at two different locations of the system under investigation. Our preliminary results indicate that by using simple statistical measures, such as the elements of the covariance matrix of the two sensors, faults at an early stage can be detected. In the proposed scheme, the extracted features are fed to a k-nearest neighbor classifier for diagnosis purposes or to an ensemble of one-class detectors, if only the information from normal situation is available. As it is proven, based on experimental results, in both scenarios a remarkably high detection/diagnostic performance is achieved.