EEMD-CNN based Method for Compound Fault Diagnosis of Bearing

Abstract

After a prolonged use of a faulty bearing, cracks are created on more than one parts of the bearing, which is a compound fault condition. This situation is tougher than the single fault condition. This combined faulty bearing creates a complex vibration signal with significant amount of noise, where it becomes very difficult to identify the fault frequencies by signal processing methods. This paper deals with a novel machine learning method for the compound fault diagnosis of Rolling bearing, where compound fault signals are decomposed into Intrinsic Mode Functions (IMF) by Ensemble Empirical Mode Decomposition (EEMD). The proposed method uses Convolution NeuralNetwork (CNN) based technique, which receives the decomposed signals of compound fault signal as input to CNN. These IMFs consists of groups of different frequencies. When these IMFs are given as input to CNN it classifies it effectively into different faults existing on bearing. CNN yields almost 96% accuracy which is better than any other previous performance for compound faultclassification.