A deep convolutional neural network based fusion method of two-direction vibration signal data for health state identification of planetary gearboxes

With the great ability of transforming data into deep and abstract features adaptively through nonlinear mapping, deep learning is a promising tool to improve the intelligence and accuracy of diagnosis. On the other hand, one acceleration sensor is not sensitive enough to position-variable faults and the collected signal is usually nonstationary and noisy. As different measurement locations provide complementary information to the faults, the paper proposes a deep convolutional neural network (DCNN) based data fusion method for health state identification. This method fuses the raw data from the horizontal and the vertical vibration signals and extracts features automatically. The effectiveness of the novel method is validated through the data collected from a planetary gearbox test rig, and experiments using DCNN, SVM and BPNN based model in different data processing methods are also carried out. The results show that the proposed method could obtain better identification results than the other methods.