針對偵測馬達旋轉故障,本論文建立一個智慧型診斷系統,利用無線感測器節點讀取馬達振動訊號,再利用動態結構類神經網路進行故障分類,並配合統計製程管制技術以鑑別出馬達旋轉故障類別。 由於馬達於工廠實際運轉時,容易遭受到雜訊干擾,使得類神經網路誤判故障特徵或是產生暫時性的異常情況,因此已有學者利用統計製程管制技術進行輔助判斷。傳統的Shewhart管制圖技術只適用於變化量較大的情況,當馬達發生故障訊號產生時,無法馬上偵測得到故障,所以本論文引入指數加權移動平均(EWMA)管制圖技術以解決上述問題。此技術是利用過去及現在觀察值的加權平均,配合權重值設定來調整管制的上下界限,因此可以診斷出比較小的變化量,尤其當馬達有故障趨勢時,可以即早偵測。 最後,本論文建立使用無線感測器的馬達故障診斷系統並進行實驗以分析、比較EWMA與Shewhart管制圖。所研發的系統可以無線的方式收集振動資料並輸入至用MATLAB撰寫的濾波、故障分類診斷及統計製程管制等功能模組,然後在Visual Basic所製作的人機介面中整合及呈現分析結果。由實驗結果可知,和Shewhart管制圖相比較, EWMA管制圖具有較高的正確率。
This thesis is mainly devoted to developing an intelligent diagnosis system for motor rotary faults. The vibration signal measurement is done by mounting the wireless sensor on the motor, and then adopts the dynamic structure neural network to establish the diagnosis functionality. Using the control chart technique in statistical process control to effectively classifies the fault characteristics. The motor operating in the factor often suffers from the noise, and a temporary fault signature may appear in the spectrum of the measured vibration signal. Therefore, scholars have been using the statistical process control to detect the motor fault. The traditional Shewhart control chart is suitable for detect the large variation and hence the trend of motor fault can not be found quickly. Therefore, this thesis uses the exponentially weighted moving average control chart to solve this problem. The EWMA control chart uses the measured values in the past and present time to calculate the weighted moving average, and uses that to adjust the upper and lower control limits. As a result, the control chart can detect the small variation, and quickly classifies the motor fault signal. Finally, this thesis develops a diagnosis system for motor rotary faults using wireless sensor nodes and compares the performance of the EWMA and Shewhart control chart. The wireless sensor nodes measure the motor vibration signals and transmit these signals to the modules built by MATLAB software. The modules include the filtering, fault characteristics extraction, diagnosis function, and statistical process control chart. All the functions are integrated and executed by using the friendly graphical user interface written by Visual Basic. From the experimental results, it is seen that the EWMA control chart outperforms the Shewhart control chart in fault classification.