CNN and GRU combination scheme for Bearing Anomaly Detection in Rotating Machinery Health Monitoring

This paper proposes a Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) combination scheme for anomaly detection to train feature extraction and to test anomaly prediction by using Stacked Convolutional Neural Networks (S-CNNs), Stacked Gated Recurrent Units (S-GRUs) as the typical model of RNNs, and a linear regression layer. In this proposed model, the S-CNNs layers firstly capture spatial feature extraction of the input sequence data of vibration sensor, and the result is used to temporal feature learning secondly with the S-GRUs. After this procedure, finally a regression layer predicts an anomaly detection. The experimental results of bearing data in NASA prognostics data repository not only show the accuracy of the proposed model in anomaly prediction for rotating machinery diagnostics, but also suggest the better performance than other state-of-the-art algorithms such as a plain RNN and GRU of the individual model.