本論文研製智慧型馬達旋轉故障診斷系統,以無線感測器及電流鉤表分別擷取馬達的振動與電流訊號,並結合免疫演算法與動態結構神經網路進行訓練與故障分類,以提高判斷馬達旋轉故障之鑑別度。 在振動訊號分析部份,本論文使用免疫演算法對動態結構類神經網路隱藏層權重部份進行最佳化。根據免疫演算法特性,將初始神經元設為抗原,使系統穩定之神經元當成抗體。經適應度計算後,將最佳解保留至記憶細胞,並且利用免疫系統中的細胞突變特性,保持群體多樣性,避免落入局部最佳解。針對電流訊號分析部份,本論文根據馬達定子電流故障頻譜特性,發現當故障發生時會在特定頻率點產生較大振幅,可進一步與健康馬達比對,作為判斷轉子斷條故障之依據。 最後,本論文經由馬達故障診斷系統進行實驗與分析,利用MATLAB R2010a撰寫動態結構類神經網路、田口方法、免疫演算法、電流訊號分析等功能模組,接著利用Visual Basic軟體整合製作出人機介面,建構出智慧型馬達故障診斷系統。透過機台故障測試實驗結果可知,本論文所提出之方法的確有效減少學習疊代數至112代及降低收斂時間至13.84秒,分別改善53.48%和50.45%。
In this thesis, the intelligent motor rotary fault diagnosis system with immune algorithm is proposed. Wireless sensor networks and a current clamp meter are used to capture motor vibration and current signals. The dynamic structural neural network, immune algorithm, and Taguchi method are adopted to increase the training convergence rate and the accuracy of motor fault classification. For signal analysis section, this work applies the immune algorithm to optimize the weight update ability in the hidden layer of dynamic structural neural network. According to the characteristics of immune algorithm, the initial neurons are set as antigens and the neurons stabilizing the system are set as antibodies. After the adaptation calculation, the optimal solutions are kept in memory cells. Also, the cell mutation in the immune system is utilized to well maintain the diversity of the cell population to prevent local optimization difficulty. Considering the current signal analysis of broken bars, larger magnitudes are observed in specific frequency points as compared with that of healthy motors. This frequency characteristic can be adopted a discriminating condition of rotor broken bar fault. The developed motor fault diagnosis system performs experiments and analysis to verify the correctness and practicality. MATLAB R2010a software is used to establish functional modules such as fault diagnosis classification, Taguchi methods, neural network, Immune algorithm, and current signal analysis. Moreover, Visual Basic software is adopted to accomplish the graphical human-machine interface. In light of the experimental results of fault classification test, the proposed method improves the convergence speed to 112 epochs and reduces the computing time to 13.84 seconds. The improvements are 53.48% and 50.45% respectively.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。