Enhancement of gross error detection when data are serially correlated

Chemical process data are typically correlated over time (i.e., serially or autocorrelated) due quite often to recycle loops, large material inventories, sampling lag and dead time, process dynamics created by high order systems, feedback control, and transportation lag. However, many of the approaches that attempt to identify gross errors in measured process variables have not addressed serial correlation, which can lead to large inaccuracies in identifying biased measured variables. Hence, this work extends the unbiased estimation technique (UBET) (Rollins and Davis, 1992) to address serial correlation. The serially correlated gross error detection (GED) study of Kao, et al. (1990) is used as a basis for setting up the study and comparison. In their work, the type of autocorrelation was assumed known (ARMA(I,I)) and the measurement test (MT) was used for identification of the measurement bias. Kao, et nl. (1990) attempted to prewhiten the data and used variances of measured variables derived from the knowledge of the time correlation structure. This work presents a different and superior prewhitening method that is shown to truly transform the data to white noise. The UBET and MT are applied to the transformed data and compared in a simulation study.