ABSTRACT: Chemical engineering models may have many reactions, many kinetic parameters, and many mass transfer and thermodynamic constants. Consequently, models that can fully describe a chemical engineering process are usually nonlinear and complex and have many unknown parameters. Additionally, chemical engineering models are not perfect and there are random disturbances from the environment that should be estimated. A challenge in estimating parameters in models for chemical engineering systems is that experiments and measurements are often limited due to cost or difficulties in measuring certain variables.
As a result, the number of data values for parameter estimation may be limited and some of the states are often not measured. The purpose of the talk is 1) to illustrate statistical techniques to estimate parameters in nonlinear large-scale models with limited data and 2) to present an easy-to-use method for estimating parameters in nonlinear dynamic models with process disturbances and model mismatch. This presentation will show how to estimate parameters of a nylon 66 model using appreciate statistical techniques. The nylon model has too many parameters to estimate and not having enough data to reliably estimating all of the parameters. The second part of this seminar will describe my development of a computationally efficient Bayesian algorithm for estimating parameters and disturbances in nonlinear dynamic models with stochastic disturbances. The effectiveness of the Bayesian algorithm will be illustrated using a case study. The proposed parameter estimation techniques will help engineers who implement on-line state estimation, process monitoring, control and optimization schemes.
Dr. Hadiseh Karimi is an Industrial Postdoctoral Fellow at Nova Chemicals Corporation. She received her Ph.D. from the department of chemical engineering at Queen’s university in 2013 in the area of applied statistics. Her Master’s degree is in process control from Sharif University of Technology, Tehran, Iran. Dr. Karimi's research interests include applied statistics, mathematical modeling, advanced process control and optimization.