Observability Guaranteed Distributed Intelligent Sensing for Industrial Cyber-Physical System

Distributed sensing is a key process for acquiring system state information in the network environments of industrial cyber-physical system (ICPS). Considering the unknown complex industrial system models, the intelligent methods for distributed sensing are received extensive attention. In most existing works, the system observability is assumed strictly first to obtain complete sensing information for subsequent state estimation. But with the expansion of industrial monitoring network scale, the observability requirement is increasingly difficult to be satisfied in advance. Therefore, a new distributed intelligent sensing method with guaranteed observability is proposed for ICPS in this paper. Specifically, a distributed learning mechanism based on field level data is designed to dynamically approximate the distributed sensing process. Then, the learning weight complete update condition is provided to actively guarantee the observability, and the novel convex-set construction approach is proposed to handle the non-convex property of this condition. Besides, the learning convergence speed and error bound are analyzed in detail. Finally, the proposed method is applied into the industrial hot rolling laminar cooling process based on the established simulation system. Compared with state-of-the-art methods in distributed intelligent sensing, the proposed method can actively reduce the sensing cost while improving the sensing performance with guaranteed observability. An average overall improvement of 24.1% in the normalized sensing performance and selection number of sensing terminals is achieved, which provides a solution for the upgrade of intelligent sensing of key processes in similar ICPS.