結合適應性IIR濾波器及人工智慧技術實現流量控制器之故障檢測與診斷

Abstract

在本論文中，我們結合適應性濾波器、模糊推論及Dempster-Shafer理論產生一線上即時故障檢測與診斷系統，且將此系統應用在半導體製造中之流量控制器(MFC)，並以一正規化格狀結構之時變IIR濾波器去找出MFC的模型，因為正規化格狀結構可防止找到不穩定的模型。而IIR濾波器係數的更新是使用遞迴式最小平方法(RLS)，因為RLS具有收斂速度快的優點。接著以IIR濾波器之係數萃取出具物理意義的參數，這些參數為阻尼因數、自然頻率及穩態誤差，並以這些參數為特徵進行故障之檢測與診斷。在故障檢測與診斷的方法上，是以模糊推論及Dempster-Shafer理論來實現，在本文中，我們也呈現了此MFC診斷系統的實驗結果。

In this thesis, we combine the techniques of adaptive IIR filtering, fuzzy inference, and Dempster-Shafer theory to create an on-line real-time fault detection and diagnosis system which is tested for mass flow controller (MFC) operated in the semiconductor manufacturing. The characteristic of the MFC is modelled as a time-varying IIR filter which is realized using the normalized lattice structure in order to prevent unstable modelling. The IIR coefficients are apdated using the well-known recurisive least square (RLS) algorithm for increasing the convergence speed.The resulting IIR coefficients are used to extract the physically meaningful parameters, the damping factor, the nature frequency, and the steady-state error, as the features or symptoms for fault detection and isolation. We then apply the fuzzy inference and Dempster-Shafer theory techniques to realize the functions of the fault detect and isolation. We also demonstrate the experimental results of the presented diagnosis system for the MFC.