A Novel Bearing Fault Diagnosis of Raw Signals Based on 1D Residual Convolution Neural Network

Diagnosing the faults of rolling bearing is an important thing, because it is the key component as well as the most easily damaged part in mechanical equipment. Whereas the traditional intelligent diagnosis methods rely heavily on the complex time-consuming feature extraction processes and rich diagnostic experience. Using the characteristics of one-dimensional vibration signals, we propose a one-dimensional residual convolution neural network (1D-RCNN) that directly uses time-domain waveforms as input. By introducing 1D CNN, residual blocks, batch normalization, and global average pooling layer, our model can effectively extract and identify features through deep convolution neural networks automatically without overfitting. A series of experiments were conducted based on the dataset of Case Western Reserve University (CWRU) to evaluate the performance of our proposed approach. According to the results, the proposed method is superior to state-of-the-art approaches using handcrafted features or raw data as input, and has certain practical potential in industrial production.