查詢結果分析
來源資料
頁籤選單縮合
題 名 | 專家系統振動訊號圖型判別之研究=Plot Identification of Vibration Signals Using an Expert System |
---|---|
作 者 | 黎文龍; 吳明川; 劉興華; 鄭增祥; 陳瑞麒; | 書刊名 | 台電工程月刊 |
卷 期 | 599 1998.07[民87.07] |
頁 次 | 頁72-80 |
分類號 | 448.13 |
關鍵詞 | 類神經網路; 診斷; 監測; Artificial neural network; Diagnosis; Monitoring; |
語 文 | 中文(Chinese) |
中文摘要 | 由於電廠中迴轉機械在不同狀況下有不同振動訊號,本研究以發展適用於此種迴轉 機械之診斷、偵測用專家系統為中程目標,最終則為配合臺電總體規劃,發展出線上自動偵、 檢測之專家系統,以提供發電廠更高效率、更正常供電與安全之運轉。本文對於振動訊號之辨 識,係採用類神經網路(ANN),又由於一般迴轉機械系統之振動訊號之不確定性、雜訊較高,故 本計畫採用較為穩健之ANN倒傳遞模式(BPN, Backward Propagation Network),本研究將一 些常見的迴轉機械異常缺陷用各種不同之圖形表現,粹取其訊號特徵,一方面當作訓練ANN之 知識資料;另一方面也可以當作在整體專家系統未完成前,現場人員訓練之用。由於本計畫採 用ANN之徵狀辨識,除了具有學習功能外,也可以根據後天環境變化,修正知識庫,作出較正確 的辨識,降低訊號量測上的干擾或量測誤差影響。本計畫之程式,從訊號擷取介面到使用者介 面,一律都採用C++語言,而整個架構是完全開放的,使用者可以完全依照實際的狀況作必須 的修改。最後,本研究所發展之系統以(1)正常運轉、(2)質量不平衡、及(3)對心不準等三種 模式作為測試,效果良好。此外,程式上仍然保留有擴充之空間,以方便使用及以後其他相關計 畫之發展與相容。 |
英文摘要 | It is well-known that a rotating machine may have different vibrating signals depending on many variables and running conditions of the machine. This is also the reason why an expert system used for fault diagnosis and protective is proposed in Taiwan Power Co. (TPC). In order to do so, a-computer-aided system which has capability to analyze the collected vibration signals is the best option. Therefore, the long-term goal of the present research is to develop an on-line diagnosis expert system that can be utilized for predicting possible faults or abnormal operations. The robust and knowledge-based artificial neural network (ANN) based on backward propagation (BPN) model is chosen to eliminate uncertainty and to reduce the noise effect that is very common in all operating machines. In addition, both the lab and field collected signals provided by TPC are used as training samples in the ANN. Depending on the faults, the features and patterns are extracted from the given data. These features are either represented in the frequency domain or in polar plots or by Bode diagrams. Furthermore, the open-structured with C++ programming is employed while friendly picture based man-machine interface is provided. Since the ANN expert system has an capability of self-learning, the change from different environments may be adequately adjusted to the optimal. Three patterns have been tested and completely furnished. They are (1) normal running, (2) mass unbalance, and (3) mis-alignment models. However, owing to the open program structure of the system, the user may add all other necessary fault patterns easily. |
本系統中英文摘要資訊取自各篇刊載內容。