| Summary: | 博士 === 中原大學 === 化學工程研究所 === 93 === A novel approach based on the controlled output performance for systematically assessing the current controller performance, diagnosing and removing the root cause resulting in performance deterioration is proposed. This fault diagnosis methodology is twofold: (1) assessing current control status relative to the benchmark; (2) detecting and diagnosing the root cause in performance deterioration. This work does not require the traditional complex physical model. Based on the current process operating data only, it can achieve accurate fault identification. According to the Diophanitine decompositions, the controlled output variances can be decomposed into controller independent term and controller dependent term. Furthermore, the achievable performance bound and the corresponding optimal parameters of the controller structure computed from the closed loop operating data and optimization procedure will be also derived. The fault detection and diagnosis approaches are ignited if and only if the current output variance significantly deviates from the initial benchmark. Based on the characteristics of the closed loop output response, a diagnosis tree structure that has the characters of hierarchy and integrated knowledge is established. This is achieved by a series of the statistical hypothesis procedure, testing an impulse response of the controlled output. As long as the value of the test statistics exceeds the threshold value, the effect of the fault is identified. The method starts with the comparison between the current operation system and the benchmark system and then the benchmark models are successively replaced whenever any fault is found out. The new benchmark model obtained from the estimated sum of square of impulse response coefficients is successively replaced. The stepwise diagnosis procedure is repeated till all possible faults are detected. It is actually practical and easy to do the operating process. The diagnosis tree structure is a stepwise search procedure from general to the special. Based on this diagnosis analysis procedure, the proposed research will be separately carried out for different control structures: feedback, feedforward/feedback, series and parallel cascade systems. The above selected control structures are widely applied to the industrial process. The capability of the proposed method will be illustrated through a series of simulation cases and pilot scaled experiments, including the single and the multiple fault problems.
|
|---|
