轉動機械之智慧型階次追蹤診斷系統

Abstract

本研究的主要目標是希望研究發展一套針對轉動機械的智慧型監測診斷系統，以更有效地來監測變動於各種轉速下之轉動機械。在傳統的階次分析法上，是以傅立葉分析法為主，並配合軸的轉速來達到轉動機械的階次追蹤(Order Tracking)。然而在轉軸轉速變化的情形下，再取樣過程(Resampling)常被用於取捨時、頻域上的解析度。此方法有許多缺點，尤其是相鄰近階次與相交越階次上，存在有頻率抹平(Frequency Smearing)的現象，而且傳統的階次分析法也無法處理多轉軸機械系統。在本文中將以兩種適應性階次追蹤技術(Adaptive Order Tracking Technique)，分別利用經過改良後的遞迴式最小平方法(Recursive Least-Squares method)與卡氏濾波器(Kalman Filter)，來解決傳統的階次分析法所遇到的困難。工作內容將包含訊號處理及狀態推論，訊號處理部分是將振動訊號經過改良後的遞迴式最小平方法與卡氏濾波器做階次追蹤而得到所需的特徵值；狀態推論的部分則是由模糊專家系統來實現。藉此以達到線上系統診斷的目的。此系統最後在轉子測試台上達到成功的診斷，我們並將此系統在引擎上做驗證，同樣地，對於各種狀況亦能達到辨識的目的。

The aim of this research is to develop an on-line monitoring and diagnostic system for rotating machinery. Conventional methods of order tracking are primarily based on Fourier analysis with reference to shaft speed. Resampling process is generally required in the fast Fourier transform (FFT)-based methods to compromise between time and frequency resolution for various shaft speeds. Conventional methods suffer from a number of shortcomings. In particular, smearing problem arises when closely spaced orders or crossing orders are present. Conventional methods also are ineffective for the applications involving multiple independent shaft speeds. In this proposal, we use adaptive order tracking techniques based on Recursive Least-Squares (RLS) filtering and Kalman filtering to overcome the problems encountered in conventional methods. The architecture of the system mainly comprises a signal processing module and a state inference module. In the signal processing module, we use RLS or Kalman filter method to track the order features of vibration signal. In state inference module, the fuzzy expert system is applied. Through these two modules, the on-line monitoring system is approached.