Multivariate Statistical Process Monitoring On Structural Fault

In modern plants there are many operating variables measured by sensors and logged into the process system database. Thus the amount of available data needs to be analyzed is enormous and they are highly correlated. This creates a demand for a system to monitor, control and analyzes this complex processes data to ensure the monitored process stays within desired conditions, by recognising anomalies in the process behaviour and subsequently correcting it. Statistical Process Monitoring (SPM) meets the demands; it is a system capable of detecting fault occurrence. In this context, the anomalies or faults studied is specifically the fault in structure, which is a type of fault resulted from an alteration of the processes main characteristics. SPM can be broken down into two methods which are univariate and multivariate methods. Multivariate methods or multivariate statistical process monitoring (MSPM) method take into account the correlation among the process variables and measurements; and it is capable to accurately characterize the behaviour of the processes, subsequently detecting faults, for which univariate method unable to adequately perform. MSPM method studied in this research project is Dynamic Principal Component Analysis (DPCA) method specifically on its structural fault detection ability, along with Hotelling’s (T2-statistic) and Squared Prediction Error (Q-statistic) techniques. The accuracies of fault detection ability of DPCA method will be compared with Principal Component Analysis (PCA) method