Virtualization and deep recognition for system fault classification

Efficient gearbox health monitoring and effective representation of diagnostic results of dynamical systems have remained challenging. In this paper, a new approach to using deep learning for translating diagnostic results of one-dimensional time series analysis into graphical images for fault type and severity illustration is presented, with gearbox as a representative example. Specifically, time sequences are first converted by wavelet analysis to time-frequency images. Next, a deep convolutional neural network (DCNN) learns the underlying features in the time frequency domain from these images and performs fault classification. Experiments on gearbox data demonstrates effectiveness and efficiency of the developed approach with a classification accuracy better than 99.5%.