- 學位論文
直流馬達之異常振動訊號檢測及辨識
Detection and Recognition of Abnormal Vibration Signals of DC Motors
摘要
由於馬達設備無預警之異常，往往將導致生產單位之停機損失。為預先瞭解馬達於運轉所產生警示訊號，本論文將以直流馬達作為研究對象，分析其運轉中所產生之振動訊號，以期瞭解馬達之運轉狀況，有效降低停機事故發生機率。 首先，為比較希爾伯-黃轉換與小波轉換於時間及頻譜之清晰程度，本研究以模擬訊號作為比較對象，獲得利用希爾伯-黃轉換於訊號清晰度上，優於利用小波轉換之初步結果。 其次，本研究為瞭解希爾伯-黃轉換及小波轉換於濾除突波及雜訊之適用性，分別提出並比較六種濾除策略；而模擬結果顯示，其中兩種策略（先以小波轉換於突波濾除，後以希爾伯-黃轉換於雜訊濾除；及先以經驗模態結合倒傳遞類神經網路於突波濾除，後以希爾伯-黃轉換於雜訊濾除）於突波及雜訊濾除上確具成效。 再者，以實際量測直流馬達之振動訊號作為分析對象，在完成突波及雜訊濾除程序後，則爰用前述對於希爾伯-黃轉換於頻率及時間清晰度優於小波轉換之初步結果，本研究採用希爾伯-黃轉換辨識馬達異常樣本；其分析結果顯示，利用希爾伯-黃轉換頻譜分析確可於希爾伯頻譜上可找出異常之特徵，並判斷出異常種類。 最後，為利用自動辨識方法取代前述之視覺化之功能，本研究分別利用經驗模態及多重解析度分法，分解異常訊號，而後利用特徵擷取結果進行倒傳遞類神經網路分類，將可達到自動辨識之目的。另外，為瞭解上述經驗模態及多重解析度應用於自動辨識之成效差異，本研究以實際直流馬達振動訊號作為分解對象，其分析解果顯示，結合多重解析度分解及倒傳遞類神經網路之程序，較可達到自動辨識之良好成效。
並列摘要
The unexpected malfunction in terms of motor equipment usually results in down-time cost. To understand the breakdown signal in the running process, the vibration signal produced from DC motor is employed to analyze the status of the DC motor in order to decrease the down-time probability. First, the simulated signals based on Hilbert-Huang transform (HHT) and Wavelet Transform (WT) are applied to compare the resolutions of time-frequency contour. The result has shown that the resolution based on HHT is better than that of WT. Second, to understand the performances of the spike and noise filtering, the six filtering strategies are compared. The simulated results have shown that the two strategies are effective to filter spike and noise. That is, for the first strategy, the spike filtering is applied by WT and then the noise filtering is applied by HHT, and for the second strategy, the Empirical Mode Decomposition (EMD) and Back-propagation (BP) algorithm are combined to filter spike and then the HHT is applied to filter noise. Third, the actual measured vibration signal of the DC motor is employed to filter spike and noise, and subsequently, HHT is applied to recognize the malfunction status of DC motor due to the better resolution of HHT. Finally, the method of automatic recognition is employed to substitute the aforementioned visualized inspection. The EMD and Wavelet Multiresolution Analysis (MRA) are applied to decompose breakdown signal and the extracted characteristics are employed to produce the BPNN feature vector for categorization. In addition, the actual vibration signals for DC motor are employed to implemented EMD and MRA and to compare effectiveness of the EMD mentioned above and MRA for automatic recognition. The results have shown that the combination of MRA and BPNN can achieve a better automatic recognition result.