Fault Detection of Planetary Gearboxes Based on Deep Convolutional Neural Network

Due to the reliance on the expert experience and the signal processing approaches, traditional diagnosis methods lead to uncertainty in feature extraction and fault detection results. deep learning is a great method to overcome the shortcomings of traditional fault diagnosis. For the other side, the accelerometers in a single direction are not suitable enough to position-shift damages and the vibration data is generally nonstationary and noisy, which impacts the accuracy of fault detection. Therefore, as the reason that different measurement locations provide different sensitivity degree or complete data of the damages, this research presents a method based on deep convolutional neural network (DCNN) of vibration signal for early fault detection. The accuracy of this approach is validated based on the sensor data sets collected from a experimental rig. The results show that the DCNN based fault detection method presented in this paper could obtain promising identification results.