A data-driven distributed fault detection scheme based on subspace identification technique for dynamic systems

With the aid of the subspace technique and the average consensus algorithm, the main objective of this article is to develop a data-driven design of distributed fault detection for dynamic systems using the measurement in a complex sensor network. Specifically, the design process consists of two stages: distributed off-line learning and distributed online fault detection. Among them, the distributed off-line learning stage involves the average consensus algorithm and parameter identification by subspace technique. It is worth mentioning that, the distributed fault detection approach has the same performance as the centralized fault detection approach and avoids complex information exchange. In the end, a numerical simulation example and a case study of the three-phase flow facility are illustrated to show that the proposed distributed approach can accomplish the fault detection task successfully.

Automated summary

This article aims to develop a data-driven design of distributed fault detection for dynamic systems using the measurement in a complex sensor network, utilizing the subspace technique and the average consensus algorithm. The design process includes distributed off-line learning and distributed online fault detection, and the former involves the average consensus algorithm and parameter identification by subspace technique. The proposed distributed approach achieves the same performance as the centralized fault detection approach and avoids complex information exchange, as demonstrated by numerical simulation and case studies.