Hierarchical fault diagnosis and recovery in discrete-event systems

点击翻译按钮获取中文摘要

Fault detection and recovery systems play a crucial role in protecting life and property and in increasing the autonomy and reliability of industrial systems, and thus, are of great importance in various industries such as process, power and aerospace. The design of fault diagnosis systems and recovery procedures is a complex problem requiring a systematic and formal approach. Recently, a framework for solving these problems based on "discrete-event" models has been developed by the applicant and his students. One of the challenges associated with the use of discrete-event models in diagnosis is the computational complexity of the design procedure. One way of reducing complexity is to take advantage of the hierarchical structure present in many complex systems. We have devised a methodolgy for designing diagnosis systems for hierarchical discrete-event systems. Furthermore, we have developed a technique for diagnosis in which the complexity of the model used for diagnosis is adjusted according to the required diagnosis resolution. In the proposed research program, we plan to merge the above hierarchical and multi-resolution frameworks, and thus arrive at a methodlogy that takes full advantage of system architecture to control computational complexity. In the proposed research, we also study the synthesis of recovery procedures. We would like to develop a hierarchical solution for the fault recovery problem in which a high-level supervisor (controller) issues high-level commands that are translated and implemented by a low-level controller. The high-level controller is designed based on a simplified high-level model of the plant and hence, should have a simpler structure compared to a controller designed based on the low-level, detailed model of the plant. The above-mentioned recovery procedures are designed "off-line" (before the system becomes operational). In many cases, storing all recovery procdures in computer memory may not be feasible due to the large number of failure scenarios. An alternative approach that we intend to develop is the "on-line" generation of recovery procedures in which a suitable recovery sequence is computed "on-line" once a failure is diagnosed.