研究生(外文):Heng-Tai Tien 論文名稱:永磁式同步馬達故障診斷之電流訊號特徵分析 論文名稱(外文):Fault Diagnosis of PMSM by Motor Current Signature Analysis 指導教授:劉孟昆 指導教授(外文):Meng-Kun Liu 學位類別:碩士 校院名稱:國立臺灣科技大學 系所名稱:機械工程系 論文出版年:2020 語文別:中文 論文頁數:88 論文摘要
在現今科技發達的時代中馬達無所不在，是工業界不可或缺的角色。在長時間運轉下馬達會產生疲勞故障等問題，嚴重時可能導致停機或損毀，所以馬達故障檢測與維護越來越受到重視。常見的診斷方式包含機械振動分析(vibration analysis)及馬達電流特徵分析(motor current signature analysis, MCSA)等，這些技術被大量應用於感應馬達(induction motor)的故障診斷。
與感應馬達比起，永磁式同步馬達(permanent magnet synchronous motor, PMSM) 有高效率、更加省電及功率因素高等優勢，因此為了降低用電量及提升馬達效能，人們逐漸使用永磁式同步馬達取代感應馬達。本研究使用馬達電流特徵分析方法計算健康與故障馬達的電流頻域訊號特徵，並利用兩種監督式學習的分類器識別馬達狀態。其中支持向量機(support vector machine, SVM)將主頻率與故障特徵頻率之峰值差作為輸入特徵，而深度學習(deep learning)演算法將MCSA之電流頻譜圖作為輸入特徵以識別馬達狀態。本研究並將深度學習建立之辨識架構延伸應用在時頻訊號上來比較分類效果。
論文外文摘要
With the development of advanced technologies, motors play an indispensable role and are widely used in the industry. During the long-term operation, the motor is subject to internal and external factors, such as fatigue failure, plant temperature, etc., which may seriously cause downtime or damage. Hence motor fault detection and maintenance become more and more important. The diagnostic methods widely used in the induction motor include mechanical vibration analysis and motor current signature analysis (MCSA).
Due to the rapid development of the industry, in order to reduce power consumption and improve motor performance, permanent magnet synchronous motors(PMSM) have the advantages of high efficiency, more power saving, and high power factor. Due to these reasons the induction motors are gradually replaced by PMSM. This research uses MCSA to detect the difference between the healthy and the faulty motor on the frequency domain. Two supervised classifiers are applied to determine the motor status. The support vector machine uses the peak difference between the main frequency and the fault characteristic frequency as the input feature, while the deep learning algorithm uses the image of current spectrogram as the input. The structure used by deep learning algorithm is extended to the time-frequency signal and their performances are compared.