A Fully Transparent Deep Signal Operator Network for Intelligent Fault Diagnosis

Intelligent Fault Diagnosis (IFD) based on Artificial Intelligence (AI) has made significant breakthroughs in recent years. However, the black-box nature of AI-enabled IFD hinders its interpretability, which poses a risk to safe-critical industrial assets. In this paper, we propose a fully transparent deep network for IFD using a basic neural node with a signal operator. The signal operator layer uses probability metrics to determine the connection of these signal operators. The signal operators used in the layer include Hilbert transform and interpretable neural Fourier operators. Additionally, to strengthen the sparsity of the metrics, we use L1 norm to learn a more compact structure. Through a case study of self-powered condition monitoring for bearings, we verify the effectiveness of the training and transparency of the proposed method. Moreover, transparent IFD is of utmost importance in industrial applications and the proposed deep signal operator network opens up a new and promising research sub-field of transparent IFD.