UNBIASED ESTIMATION OF GROSS ERRORS IN PROCESS MEASUREMENTS

A new approach to gross error detection provides unbiased estimates and 100(1-alpha)% simultaneous confidence intervals of process variables when biased process measurements and process leaks exist. Presented in this article are estimation equations for process variables, as well as equations that help identify biased measurements and process leaks. These equations include the power function for a global test, and two types of alpha-level component tests and their power functions. Important strengths and weaknesses of this approach are compared to those of the serial compensation strategy, in particular, by varying the significance level (alpha), the variance-covariance matrix (SIGMA), the size of measurement bias (delta), the number of biased variables, and the sample size (N). Accuracy of delta estimation and performance in detecting the presence of process leaks (gamma) are also evaluated and compared. The proposed approach has unique features that can provide a basis for improving the reconciliation of variables in process operations.