A Novel Framework for Early Pitting Fault Diagnosis of Rotating Machinery Based on Dilated CNN Combined With Spatial Dropout

Pitting corrosion of rotating machinery is one the most common faults in industrial engineering. The convolutional neural network (CNN) increasingly applied to fault diagnosis. However, conventional CNN method will reduce feature dimension collected signal and cause loss information during pooling process. In this paper, a new based on dilated combined with spatial dropout (DCSD) proposed diagnose early machinery. By filling convolution kernel, DCSD can increase receptive field without increasing number parameters while retaining more features raw vibration machine. To avoid eliminates adjacent elements strong correlation, Spatial Dropout adopted overfitting problem deep networks. pitting gears experiment was designed verify paper. data 6 different healthy states were effectiveness method. experimental results show that effectively distinguish pitting, diagnostic accuracy better than other popular learning methods.