신경회로망과 퍼지추론을 이용한 선박디젤기관의 고장진단 예측시스템

Abstract

In recent years, the ship's propulsion engines have tend to be equipped with high efficiency diesel engines because of expensive fuel cost even though they are different according to the cargo.

Their measurement points are substantially increasing over 10 thousands although there are some difference according to degree of automation for propulsion diesel engine room.

Because it is difficult for operators to manage the huge data obtained from various kinds of monitoring and control systems, it is nearly impossible to determine the faults before monitoring systems make alarm. Some faults of systems can be found easily by analyzing one or two data from monitoring systems, while some are very difficult because many data are affected each other, they are called interactive data. But expert engineer can infer the faults by analyzing these various kinds of interactive data obtained from monitoring systems for fault diagnosis of complex system using their skillful experiences and decision making tools.

Therefore in this paper, an predictive fault diagnosis system of marine diesel engines using neural networks and fuzzy inference technique is introduced. The huge data from the monitoring systems are classified into combustion system which is most primitive in diesel engine, heat exchanger systems which are important to operate diesel engine safely and continuously, and motor and pump systems which are inevitable to operate heat exchanger systems. Specially, this paper makes fault diagnosis models by analyzing methods which engineer with expert knowledges infers the faults by analyzing various interactive data and shows to build automatic predictive fault diagnosis systems with three classified subsystems by managing collected data from various monitoring systems using neural networks, fuzzy inference and decision making technique by answer tree. Also this paper shows simulation results and ascertains proposed fault diagnosis systems being appliable to real diesel engine room for three classified subsystems.