1D Convolutional Neural Network-based Signal Tracking Algorithm with Feature Extraction and System Diagnosis

Diverse adaptive algorithms have been investigated and developed in an effort to mitigate the vibration and noise produced by mechanical systems. A reference signal plays a crucial role in adaptive algorithms for monitoring the target signal, however it is difficult to identify the reference signal when its spectrum is relatively complex. In addition, if the state of a mechanical system changes due to a disturbance or malfunction, signal tracking may not perform as well if the reference signal for the changing condition has not been redefined. This work focuses on constructing a signal tracking mechanism based on deep learning to increase tracking performance. The suggested algorithm consists of the following two parts: 1) automatically defining the reference signal segment and 2) system state feedback segment. These portions are managed by convolutional neural networks in one dimension. The Case Western Reserve University (CWRU) bearing dataset was used for training and testing to validate the proposed technique. In addition, a neural network-based approach for signal tracking is applied to validate the signal tracking performance. The suggested approach demonstrates that the signal tracking performance is great regardless of the system's state.