Dynamic data reconciliation in mineral and metallurgical plants

Data reconciliation is a well-known technique to improve accuracy and reliability of plant measurements. It relies on process models that could range from simple mass and energy conservation equations to complete causal dynamic models. Generally, precise estimates imply detailed plant models that could be difficult to build and update in practice. The trade-off between modeling efforts and estimation performances has thus lead to various approaches to deal with plant dynamics. The objective of the paper is to review and compare most common observers used for dynamic data reconciliation in the mineral and metallurgical processing industries. Comparisons are carried out using a separation unit and a flotation circuit as simulated benchmark plants. Observer performances are evaluated in terms of variance reduction. Strengths and weaknesses of the different methods are highlighted. Aspects such as estimation of model parameters, detection of gross errors, and handling of bilinear equations and plant non-linearities are discussed.