| Data reconciliation is a well-documented approach for estimating values of measured process variables that are consistent with their constraining mass and energy balances. A great deal of research has been done on steady state data reconciliation but far less has focused on dynamic data reconciliation, especially nonlinear dynamic data reconciliation. In addition, the practical aspects of implementing data reconciliation strategies on-line in an industrial environment has not been addressed adequately.; The problem of data reconciliation and the detection and identification of gross errors, such as measurement bias, are closely related. In order to produce reliable estimates, standard formulations of the data reconciliation problem require data that are free of systematic errors. Many techniques designed to detect such irregularities include a data reconciliation step in the algorithm.; This close relationship prompted the development of a technique that combines these ideas, treating the presence of measurement bias as a discrete event and expressing it through the use of binary variables. This simultaneous method for data reconciliation and bias detection/identification requires the solution of a mixed integer optimization problem.; The principal goals of this dissertation are twofold; to improve on-line data reconciliation methods by presenting practical solutions to problems encountered in real-world applications and to integrate data reconciliation and bias detection and identification methods into a single coherent strategy. All the while keeping the feasibility of on-line implementation in mind. The combined data reconciliation and bias detection/identification method is developed on simple linear steady state systems and extended to be applicable to nonlinear and dynamic systems as well. This method has successfully been applied in simulation and is superior to other published methods. |
