Atten-DilatedNet: fault diagnosis method of bearing for mine drilling rig

Currently, bearing fault signal diagnosis of mine drilling rig suffers from strong time dependency and serious signal noises, which leads to low diagnostic accuracy. Aiming at solving the problem of inadequate extraction of key features hidden in strong noise, a temporal attention embedding dilated convolutional neural network (Atten-DilatedNet) is designed for fault diagnosis of mine drilling rig bearing. The method takes the original vibration signals of different faulty bearings of drilling rigs as the input of the model. It adopts a one-dimensional dilated convolutional neural network to automatically extract features from the vibration signals, with larger receptive field and longer time range than traditional convolutional operator. A temporal attention mechanism is further embedded to the network, which adaptively assigns different weights to above feature sequences and model their time dependency. Thus, Atten-DilatedNet can enhance attention to critical-to-fault feature, and restrain interference of noise. Case studies on public data sets and experimental drilling rig demonstrate the validaty and superiority of the proposed Atten-DilatedNet based fault diagnosis method. It shows higher accuracy, better generalization performance, and stronger robustness than some traditional machine learning methods.