Diagnosis and Fault - Tolerant

Overview of Fault Detection Security and reliability are of great importance in control system technology. To ensure a reasonable level of security and reliability, the need for failure detection (diagnosis) techniques has long been recognized. Roughly speaking, a failure is any kind of malfunction of the system that leads to unacceptable performance. Failure detection consists of techniques for determining whether a failure has occurred. These techniques provide the means for avoiding major breakdowns and enable the system operator to take appropriate action. Failure detection is used in high-performance systems such as aircraft control systems, power plants, and chemical plants. Not surprisingly, the development of reliable failure detection techniques has become an intensive field of research. In model-based failure detection, the assumption is that a mathematical model of the behavior of the system under surveillance, in the absence of failure, is available. Failure detection then consists of deciding whether or not the measured inputs and outputs are consistent with this model. In most cases, a realistic model must account for uncertainty and noise, which implies that the input-output behavior is not just a single time trajectory but rather a set of time trajectories. This set is often difficult to characterize, making the detection decision difficult. Failure detection has been widely studied, and the literature includes numerous books [3], [4], [7]–[9], [15], [20], [21] and survey articles [2], [10]–[14], [16], [19], [23]. In particular, the book [8] gives an up-todate overview of the subject. The book reviewed here presents modelbased methodologies for fault detection and fault-tolerant control; some of the methodologies can be found in the above literature while others are based on recent research papers.