Fault-Tolerant Control Using Recursive Gaussian Processes and Observer-Structured Robust Output Feedback Controller

Abstract

This paper discusses fault-tolerant control using recursive Gaussian processes (RGP) and observer-structured robust output feedback controllers. The proposed method can detect variations in plant dynamics based on estimated plant state using RGP. The paper gives a design method for robust dynamic output feedback controllers based on generalized plants considering uncertainties due to some faults. When output feedback control is required, running RGP that use the plant state is complicated. In addition, the existence of the uncertainties makes conventional observers unreliable. These issues motivate us to adopt observer-structured controllers, which are obtained from a priori designed dynamic output feedback controllers through only appropriate state transformations without control performance degradation. Running this scheme online makes it possible to guarantee robust stability for plants after dynamics changes due to some faults. Two numerical examples of a web server model and an airplane model illustrate the effectiveness of the proposed method.