Fault diagnosis of hybrid systems with applications to gas turbine engines

Stringent reliability and maintainability requirements for modern complex systems demand the development of systematic methods for fault detection and isolation. Many of such complex systems can be modeled as hybrid automata. In this thesis, a novel framework for fault diagnosis of hybrid automata is presented. Generally, in a hybrid system, two types of sensors may be available, namely: continuous sensors supplying continuous-time readings (i.e., real numbers) and threshold sensitive (discrete) sensors supplying discrete outputs (e.g., level high and pressure low).;The problem of diagnosability of failure modes in hybrid automata is also studied here. A notion of failure diagnosability in hybrid automata is introduced and it is shown that for the diagnosability of a failure mode in a hybrid automaton, it is sufficient that the failure mode be diagnosable in the extended DES model developed for representing the hybrid automaton and residual generators. The diagnosability of failure modes in the case that some residual generators produce unreliable outputs in the form of false alarm or false silence signals is also investigated. Moreover, the problem of isolator (residual generator) selection is examined and approaches are developed for computing a minimal set of isolators to ensure the diagnosability of failure modes.;The proposed hybrid diagnosis approach is employed for investigating faults in the fuel supply system and the nozzle actuator of a single-spool turbojet engine with an afterburner. A hybrid automaton model is obtained for the engine. A bank of residual generators is also designed, and an extended DES is constructed for the engine. Based on the extended DES model, a hybrid diagnoser is constructed and developed. The faults diagnosable by a purely DES diagnoser or by methods based on residual generators alone are also diagnosable by the hybrid diagnoser. Moreover, we have shown that there are faults (or groups of faults) in the fuel supply system and the nozzle actuator that can be isolated neither by a purely DES diagnoser nor by methods based on residual generators alone. However, these faults (or groups of faults) can be isolated if the hybrid diagnoser is used.;It is assumed that a bank of residual generators (detection filters) designed based on the continuous model of the plant is available. In the proposed framework, each residual generator is modeled by a Discrete-Event System (DES). Then, these DES models are integrated with the DES model of the hybrid system to build an Extended DES model. A "hybrid" diagnoser is then constructed based on the extended DES model. The "hybrid" diagnoser effectively combines the readings of discrete sensors and the information supplied by residual generators (which is based on continuous sensors) to determine the health status of the hybrid system.