本論文針對智慧型馬達旋轉故障診斷系統提出一個無感測器的轉速偵測方法,適用於馬達變轉速下的旋轉頻率偵測,並結合了混合判別分析法與動態結構類神經網路,以提升判斷馬達故障的鑑別度。 傳統的馬達故障診斷系統,只針對定轉速進行量測。當轉速改變時,則須加裝轉速計,才能量測到正確的旋轉頻率。利用振動諧波頻譜的物理特性,本論文提出了一個以無感測器方式就能由系統自動估測出旋轉頻率的轉速量測方法,進而達到節省成本的目的。本論文還引入了混合判別分析法,考慮到資料整體、同類別及不同類別之間的關係,經過轉換後再結合動態結構類神經網路作學習,可同時縮短網路的訓練時間及次數,提升故障診斷的鑑別度。 本論文在實驗中透過無線感測器節點傳輸振動資訊,並使用MATLAB軟體撰寫訊號處理、無感測器轉速估測和類神經網路等功能模組,並以Visual Basic軟體進行人機介面整合,建構出適用於變轉速下的智慧型馬達故障診斷系統。由測試機台故障實驗結果可知,本論文所提方法可有效估測轉速並得到較佳的故障鑑別度。
For the intelligent motor fault diagnosis system, this thesis proposes a sensorless speed estimation algorithm to calculate the motor rotary frequency during speed-changing operations. The new system incorporates the hybrid discriminant analysis and the dynamic structure neural network to increase the accuracy of fault diagnosis. Traditional motor fault diagnosis system only focuses on the fixed rotary speed measurement. When the motor speed changes, the system requires a tachometer to measure the proper rotary frequency. Using the physical spectrum characteristic of vibration harmonics, this study proposes a sensorless method to automatically calculate the new rotary frequency. Additionally, this work introduces the hybrid discriminant analysis (HDA) to investigate the relations among the whole data, the same and different categories. After the HDA translation, the translated data incorporates the dynamic structure neural network for learning. The network training time and learning counts are thus reduced and the accuracy of diagnosis is increased. The testing platform transmits vibration information via wireless sensor nodes in the experiment. MATLAB software is used to develop the modules of the signal process, sensorless rotary estimation, and neural network. The human-machine interface is programmed and integrated by Visual Basic software. Finally, from the faulty experiments conducted by using the testing platform, the experimental results verify the correctness of the speed estimation and the diagnosis system gains better diagnosis accuracy.