Likelihood and Bayesian Methods for Accurate Identification of Measurement Biases in Pseudo Steady-State Processes

Abstract

<p>Two new approaches are presented for improved identification of measurement biases in linear pseudo steady-state processes. Both are designed to detect a change in the mean of a measured variable leading to an inference regarding the presence of a biased measurement. The first method is based on a likelihood ratio test for the presence of a mean shift. The second is based on a Bayesian decision rule (relying on prior distributions for unknown parameters) for the detection of a mean shift. The performance of these two methods is compared with that of a method given by Devanathan <em>et al</em>. (2000). For the process studied, both techniques were found to have higher identification power than the method of Devanathan et al. and appears to have excellent but sightly lower type I error performance than the Devanathan et al. method.</p>