Degree Type


Date of Award


Degree Name

Doctor of Philosophy


Chemical and Biological Engineering

First Advisor

Derrick K. Rollins

Second Advisor

Stephen B. Vardeman


In this work three new methods are presented for improved identification of measurement biases in linear and nonlinear pseudo steady state processes. In addition to these methods, a new method is outlined for identification of biases in dynamic processes;The first method makes use of information contained in the relationship between individual measurements and the corresponding nodal balance. The performance of this method is demonstrated on a problem from the literature that has proved difficult for earlier methods. Additionally, this work discusses how the new technique can be used as a visual monitoring tool for identifying biased measured variables;The second and third methods examine each process variable individually and do not require the use of process physical constraints. Thus, neither of these methods is affected by the linearity or nonlinearity of process constraints. The second method involves obtaining the maximum likelihood estimate for the expected value of each measured variable. The decision rule is based on testing for a change in the expected value of each variable leading to an inference regarding the presence of a measurement bias. The third is a Bayesian approach that uses a priori information on the unknown parameters involved in the statistical measurement model for each process variable. The decision rule for identifying a bias is based on the mode of the conditional distribution of a change point parameter, given the data (and the prior information);Results of simulation studies are presented for all methods in terms of performance measures commonly used in literature. The studies involve varying the bias magnitude, time of occurrence, and probability of false identification. Using a process example previously employed by other researchers, the performance of the new methods is shown to be superior to current methods;While these methods are shown to be capable of accurately detecting mean shifts, a limitation that applies to all three methods is that it is not possible to ascertain whether the process variable has moved from a biased state to an unbiased state or from an unbiased state to a biased state. In other words, the methods are really concerned with detection of changes in bias, not directly with the detection of nonzero bias.



Digital Repository @ Iowa State University,

Copyright Owner

Sriram Devanathan



Proquest ID


File Format


File Size

78 pages