摘 要:針對旋轉機械故障征兆與故障模式映射的復雜性,以及BP網絡容易陷入局部極小、收斂速度慢等缺點,提出基于徑向基(RBF)神經網絡的風機故障診斷方法。以風機振動信號的7段頻譜能量峰值作為故障特征,采用訓練好的RBF網絡進行故障辨識。結果表明,RBF網絡能滿足風機故障診斷的準確性,并在避免局部極小和節約訓練時間方面有較好的實用性。關鍵詞:RBF神經網絡; 故障診斷; 風機; 故障特征
中圖分類號:TN919-34文獻標識碼:A
文章編號:1004-373X(2010)18-0141-02
Fault Diagnosis of Rotary Machines Based onRBF Neural Network
WANG Qing-hua1, WANG Jing-tao2, DENG Dong-hua3
(1.School of Mechatronic Engineering, Xi’an Technological University, Xi’an 710032, China;
2.Guangxi Petrochemical Company, Qinzhou 535008, China; 3. China Petroleum Pipeline Engineering Corporation, Langfang 065000, China)
Abstract: Aiming the mapping complexity between fault symptoms and fault patterns of rotary machines, and the problems of falling easily into part minimums and low velocity of convergence in BP neural networks, a fault diagnosis method of fan based on Radial Basis Function (RBF) neural network is put forward. Making the seven frequency bands peak energy of vibration signals of a fan as fault symptoms, RBF network is trained to diagnose a fan, the results show that RBF network is a valid method of the fault diagnosis of fan in proving accuracy, repressing the network to sink local minimum and shortening the study time.Keywords: RBF neural network; fault diagnosis; fan; fault features
0 引 言
隨著旋轉機械大型化、自動化、高速化和復雜化的發展,其運行的可靠性和安全性日益受到重視,對其進行可靠、準確的故障診斷是石油、化工、冶金、礦山、機械等各行業安全生產的重要保障。設備故障診斷其實就是故障設備運行狀態的辨識,即對運行狀態進行分類識別,判斷設備有無故障,如果有,則要進一步判斷故障屬于哪一類,所以設備故障診斷實質上是一個模式識別問題[1]。學習機器是實現模式智能識別的最主要手段,訓練或獲得學習機器的方法稱為機器學習方法。機器學習方法是在一個由各種可能的函數構成的空間中尋找一個最接近實際分類函數的分類器。在設備故障診斷中,故障征兆與故障模式并不是簡單的一一對應的關系,其構成的故障特征空間比較復雜,常常不是線性可分的,有時甚至是完全不可分的。而學習機器中神經網絡能夠映射任意復雜的非線性關系,具有自學習、自組織、自適應等特性,并且具有極強的容錯和聯想能力、較快的計算速度,所以神經網絡被廣泛用于機械故障診斷識別中[2]。……