퍼지 및 신경망을 이용한 터보축 엔진의 다중손상

Abstract

Helicopter propulsion system operates in severe flight conditions such as hot and cold temperature, heavy snow and rain, foreign particles due to dust, sand and birds, etc. This severe operating condition may increase possibility of foreign object ingestion such as sand, dust, birds, etc., which can give rise to damages of engine gas path components. Types and severities of most engine faults being so complex, conventional model based fault diagnostic approaches like the GPA (Gas Path Analysis) method may not monitor precisely all engine fault conditions. Recently soft computing methods such as Genetic Algorithms, Fuzzy Logic, Neural Networks and Expert System have been applied to the advanced gas turbine HMS (health monitoring system). Moreover, the on-line HMS has been developed for immediate and effective action on identified faults rather than the ground health. On-condition monitoring and fault detection of the engine can promote reliability and availability of the helicopter propulsion system. A hybrid health monitoring program using Fuzzy Logic and Neural Network Algorithms is proposed. In this hybrid method, such as Fuzzy Logic, it easily identifies the faulted components from changes of engine measuring parameters and the Neural Networks can accurately quantify the identified faults. In order to use effectively the fault diagnostic system, a GUI (Graphic User Interface) type program is proposed. This program is composed of the real time monitoring part, the engine condition monitoring part and the fault diagnostic part. The real time monitoring part can display measuring parameters of the turboshaft engine under study. The parameters are : power turbine inlet temperature, exhaust gas temperature, fuel flow, torque and gas generator speed. The engine condition monitoring part can evaluate the engine condition through comparison between monitoring performance parameters and the base performance parameters. The analysis was carried out using base performance analysis program, as well as look-up tables for component maps. The fault diagnostic program developed in this report can identify and quantify the single and multiple faults from the monitoring parameters using hybrid method.