Intelligent Fault Classification of a Misaligned Geared-Rotor Machine Equipped with Active Magnetic Bearings

Gear can be considered as one of the most vital components in any rotating machine. Like any mechanical system, geared-rotor systems can experience various types of faults or failures. One of the most commonly experience faults is shaft misalignment, which can cause due to improper installation or wear. In recent years, machine learning and deep learning techniques have sparked great interest in accurately diagnosing geared-rotor faults. Therefore, in this paper a fault detection intelligent method is implemented to predict the type of fault class provided with a labeled set of input vibration and current data. A multi-class classification artificial neural network (ANN) model is developed with statistical features extracted from time-domain vibration. The vibration dataset is built by conducting an experiment on a geared-rotor test rig where angular misalignment is deliberately introduced with four fault conditions: no misalignment and three severity levels of misalignment. Transverse vibration data are recorded with proximity probes. The set-up is also equipped with two active magnetic bearings (AMBs) mounted on the input and output rotors, which are used for vibration suppression. The control current signal running through the AMB coils and time-domain vibration signal at three different speeds are used for misalignment diagnosis. Features like mean, root mean square and entropy have been found to perform the best. The optimum tuning of the hyper parameters of the ANN model is done to achieve about 98.33% prediction accuracy.