Geçici durum sürecini dikkate alan hata algılama metodu ve uygulamaları

The purpose of Fault Detection (FD) is to determine the occurrence of an abnormal event in a process. The early detection of faults may help to avoid system breakdowns and product deterioration. Fault detection algorithms and their applications to a wide range of industrial processes have been the subject of intensive research over the past two decades (Isermann, 2006; Karami vd., 2010). The methods on Fault Detection and Diagnosis (FDD) can be divided into two main groups, the modelbased and data driven-based methods (Venkatasubramanian vd., 2003a; 2003b; 2003c). Model-based FD methods are based on comparing the behaviors of the actual plant and a mathematical model of the system (Hammouri vd., 2010). The method uses signal residuals, which indicate changes between the real process and the process model. However, obtaining a complete and robust mathematical model is difficult due to process complexity and dimension. The data-based FD methods can be used to solve these problems (Venkatasubramanian vd., 2003a). The advantage of these methods is that the model of the system is not necessary to know in order to make a conclusion on a fault appearance. This means that the method is appropriate for the systems that cannot be easily or ever modeled, or for which the model is nonlinear, hybrid, or structurally ill-posed. For the data-based methods, only the availability of large amount of historical process data is needed (Venkatasubramanian vd., 2003a). One of the most common multivariate statistical process control (MSPC) methods used for this purpose is principal component analysis (PCA) (Camacho vd., 2009). PCA method initially proposed by Pearson (1901) and later developed by Hotelling (1947). PCA method is used to extract a few independent components from highly correlated process data and use the components to monitor the process operations. Typically, two major monitoring indices are calculated, the squared prediction error (SPE) and the Hotteling T2 index. An abnormal situation will cause at least one of the two indices to exceed the control limit. Conventional PCA methods for the fault detection have largely focused on the steady-state operations and are not directly applicable during the transitions (Jia vd., 2010). Applying a PCA method to such a transient process can produce excessive number of false alarms or missed detection of process faults, that is significantly compromise the reliability of the monitoring system. Therefore, a novel PCA fault detection method is required that explicitly caters to the non-steady states and wide operating condition changes during transitions. In the present article, a new monitoring approach is proposed based on PCA method that covers both the steady-state and transient operating conditions for the stationary signals with the variance sensitive adaptive threshold (Tvsa). The method is implemented and validated experimentally on a process control system using on-line data. Experimental test confirms the fact that the proposed method is applicable and effective for both the steady-state and transient operations and gives early warning to operators.