MODIFIED INTEGRAL PROCEDURE (MIP) AS A RELIABLE SHORT-CUT METHOD FOR KINETIC-MODEL ESTIMATION - ISOTHERMAL, NONISOTHERMAL AND (SEMI-)BATCH PROCESS CASES

In both small and large-scale investigations, a reliable short-cut pro cedure to estimate the approximate parameters is very useful for the s uccessive rapid checking of different Kinetic Model (KM) structures fo r their adaptation to current process data. An improved quality of the initial parameter guess also improves the reliability and the converg ence rate for a subsequent exact Nonlinear Least Squares (NLS) regress ion technique applied for fitting the final model. The recently propos ed Modified Integral transformation Procedure (MIP) short-cut estimati on method of Maria and Rippin (1995) [Computers and Chemical Engineeri ng 19 (Supplement), S709-S714 (1995)] adds supplementary elements of s imilarity analysis and prior information about similar model structure s to the classical Integral transformation Procedure (IF) for kinetic parameter estimation. By exploiting the model structure and the intera ctive use of information stored in a kinetic databank, the MIP makes r apid adaptation of a KM and parameters, describing an already studied process, to a similar process under study with only the product distri bution known. The problem decomposition and the term-by-term sensitivi ty and estimation analysis of the model for various portions of experi mental data sets result in a very effective MIP. The generated initial parameter estimate is more reliable and of better quality compared wi th the classical direct techniques, especially for non-linear and ill- conditioned cases. Algebraic transfer of information functions are dev eloped in interaction with the kinetic databank, leading to a rapid ch eck of different kinetics, or the same kinetic model for different dat a sets, without time-consuming intermediate NLS steps. The MIP was int egrated in an expert system for kinetic identification and coupled wit h statistical data/estimate analysis (Maria, 1993 [Computers and Chemi cal Engineering 17 (Supplement), S435-S440 (1993)]; Maria and Rippin, 1996 [Computers and Chemical Engineering 20 (Supplement), S587-S592 (1 996)]). MIP implies any iterative search, it has no convergence proble ms and requires no tuning factor. The basic MIP, developed for isother mal data treatment, is also shown to be suitable for on-line kinetics identification in (semi-) batch processes. The interaction with the pr ior information allows on-line adaptations of the model structure and parameters, comparable with extended Kalman Filter (EKF)-based recursi ve estimators. In the present work these results are also extrapolated for linear kinetics estimation by using non-isothermal data. (C) 1997 Elsevier Science Ltd.