A comparison study on the different SNR levels to the accuracy of two deep learning techniques in fault diagnosis of planetary gearbox

For most of the current deep learning models, the noise contaminated data are usually used in both model training and test. The ability of deep learning models to withstand different levels of noise becomes an important research topic. In this work, a recent developed deep learning model called DCAE-CNN is studied under different noise levels of training and test samples. The model combines a one-dimensional De-noising Convolutional Auto-encoder (DCAE-1D) and a one-dimensional Convolutional Neural Network (CNN-1D). The DCAE-1D network enables the noise reduction ability embedded in the model and may accommodate noisy training and test samples. The effectiveness of this model is studied by using different levels of SNR to planetary gearbox fault diagnosis. The DCAE-CNN and CNN model are compared by using different SNR noise contaminated training and test samples, the results demonstrated the effectiveness of the DCAE-CNN.