因石油日漸枯竭與節能減碳的需求，以馬達取代傳統引擎來推進船舶漸成爲未來趨勢。感應馬達因具 苻構造簡単、反應快速、推進力量大等優勢較常被使用，而促使線I：即時馬達故障監控受到重視；然而' 故障監控的準確性與即時性是此項工作中較爲困難的。一般而言，馬達故障主要發生於定子、軸承與轉子 相關處•該故障又肇因於電力與機械兩部分，其中又有80%的機械相關故障會導致氣隙偏心的現象 ' 而降 低馬達功率因數與發生振動，進而損耗電能與擴大故障範圍，甚至衍生定子與轉子磨擦之問題。是故，若 能事先預判感應馬達之故障，以進行保養將可避免因馬達的突然故障而造成人員受傷或營運上的損失。本 文應用Matlab/Simu丨ink模擬比較一部30kW三相四極感應馬達於正常運轉、配置不均與轉子偏心等情況之電 流訊號的差異，並驗證其氣隙偏心的故障特徵，再與文獻作比對。最後，吾人藉由模糊運算實現無需將時 域信號先轉至頻域再用神經元演算法分析故障狀況的方法，進而能於時域下快速準確比對出感應馬達的偏 心故障訊息，並判定故障等級。

Owing to the gradual depletion of oil and a concern for environmental protection, motors have become the drive system of the future instead of the combustion engine. Among various types of motors, induction motors are often used due to their simple structure, fast reaction, and powerful propulsion force. Thus, the safe operation of motor vehicles is a major concern with the trends in the online fault monitor. However, the accuracy of online monitoring is the most difficult task. Generally, the motor failures caused by electric and mechanical parts may occur mainly in stator, bearing and rotor. Among mechanical faults, 80% of relevant troubles may cause the phenomenon of air-gap eccentricity, leading to the power factor decreasing, motor vibrations, power losses and enlargement of the fault area. Particularly, it may result in the rotor and stator sides to craft each other in a serious situation. If one can detect motor faults in advance, maintenance schedule or parts replacement can be properly arranged to reduce the risk of suddenly shut down in motor operation. This paper simulates the differences of the characteristics of electric current signals for a 30 kW, four-pole, three-phase induction motor under normal condition, uneven dispose fault and air-gap eccentric fault using Matlab/Simulink. To verify the feasibility of the models developed in the Matlab/Simulink environment, the simulated fault signals are compared with the results obtained by other research. Finally, this paper shows that the air-gap eccentricity degree can rapidly and reliably be diagnosed without transforming the fault signals into the frequency domain instead of analyzing them with a neural algorithm.