Data-driven fault detection for large-scale network systems: A mixed optimization approach

Abstract

This paper considers the fault detection (FD) problem for large-scale network systems with unknown system dynamic matrices. Compared with single systems, the FD problem is hardly solved due to the unmeasurable interconnection signals composed by neighboring subsystems states. To overcome this difficulty, the unmeasurable interconnection terms are estimated within the data-driven framework firstly. Then, a residual generator is designed in terms of the input and output data. Moreover, considered the freedom degree in design of the residual generator, an H−/H∞ mixed optimization scheme is proposed to enhance the sensitivity to the actuator faults as well as the robustness against the measurement noises. Based on it, actuator faults with smaller magnitude can be detected. Also, the advantages and effectiveness of the proposed FD approach are verified by a numerical example.