LAUSR

Rub-impact is a kind of serious malfunction, which often occurs in rotating machinery. The non-stationary rub-impact signals are always submerged in the background and noise signals, which makes it difficult to accurately diagnose the rubbing based on the hand-designed features extracted by the traditional methods. This paper presents a 1-D convolutional neural network (CNN) based approach to automatically learn useful features for rub-impact fault diagnosis from the raw vibration signals of a rotor system. The proposed model is trained on a dataset of vibration signals obtained from an industrial hydro turbine rotor. The results show that timely and accurate rub-impact fault detection can be achieved by a simple 1-D CNN configuration.