Industrial Big Data Analytical System in Industrial Cyber-Physical Systems Based on Coarse-to-Fine Deep Network

In smart factories, there have been increasing requirements for industrial Big Data analysis of complex systems. With the rapid development of industrial cyber-physical systems (ICPS) and communication techniques, the scale and complexity of industrial data are growing explosively, which not only provides massive operational information of industrial systems but also brings challenges in Big Data analysis. In this paper, to overcome the intra/inter-class distance unbalance and local minima problems in traditional deep learning-based methods, an industrial Big Data analytical system based on a coarse-to-fine network (CTFN) is proposed for intelligent industrial Big Data analysis and condition monitoring of complex system. In addition, considering the gap between semantic comprehension and natural characteristics of different failures, a structure learning algorithm is proposed to get rid of the complicated hyper-parameters and implement intelligentization authentically. Finally, an experimental verification was carried on a nuclear power system dataset with 362,994 samples from 66 fault categories. The results demonstrate the effectiveness and superiority of the proposed method in condition monitoring of industrial systems, which provides a promising tool for industrial Big Data analysis in ICPS.