Real-time kinetic hard-modelling for the optimisation of reaction conditions and the detection of process upset in semi-batch reactors

Process Analytical Technology (PAT) has greatly evolved in the last decades due to the development of multivariate online sensors that are able to monitor the properties of industrial processes in real time [1, 2]. The online monitoring of product quality and the detection of process upsets are important for the pharmaceutical and fine chemical industry in order to maintain their product specifications and their commitments regarding safety, health and environment. Most frequent sources of deviations from normal operating conditions in semi-batch processes are due to slightly imprecise initial conditions or impurities in the initial reactants causing unexpected side reactions [3]. <br><br> Online monitoring of industrial processes usually relies on calibration methods, such Principal Component Regression (PCR), Partial Least Squares (PLS) or Black Box models (e.g. Neural Networks) [4, 5]. A drawback of these calibration methods is their poor behaviour regarding extrapolation, requiring a constant effort for the operator to maintain the calibration conditions. Kinetic modelling techniques [6] do not suffer from this drawback as they are based on first principal models and can also be adapted for the monitoring of highly fluctuating processes, e.g. semi-batch processes. <br><br> In this contribution, we propose a method for the online monitoring of semi-batch processes based on kinetic hard-modelling. The proposed method assumes that the kinetic model and the associated rate constants have already been determined at an earlier stage in R&D. In a first phase, the algorithm corrects estimates for the initial concentrations from dosing a small amount of reagent and fitting the kinetic model to the measured signals, e.g. mid-IR, UV-vis or released/consumed heat. In a second phase, if no process upset is detected, the corrected initial concentrations are fed back into the kinetic model and the algorithm optimises the dosing rate of the reagent or the operating temperature by maximising a property of the process, e.g. yield, selectivity or conversion. When optimum operating conditions are found, the algorithm forces the reactor to work under these improved conditions and the process is continuously re-optimised to detect possible process upsets. <br><br> The method will be discussed based on simulated and experimental data taken from our high performance small scale reaction calorimeter coupled to in-situ mid-IR and UV-vis ATR-spectroscopy [7]. <br><br> [1] P. Gemperline, G. Puxty, M. Maeder, D. Walker, F. Tarczynski, M. Bosserman, Analytical Chemistry 76 (2004) 2575-2582. <br> [2] J. Workman, M. Koch, D. Veltkamp, Analytical Chemistry 77 (2005) 3789-3806. <br> [3] E.N.M. van Sprang, H.J. Ramaker, H.F.M. Boelens, J.A. Westerhuis, D. Whiteman, D. Baines, I. Weaver, Analyst 128 (2003) 98-102. <br> [4] M. Spear, Chemical Processing 70 (2007) 20-26. <br> [5] T.J. Thurston, R.G. Brereton, D.J. Foord, R.E.A. Escott, Journal of Chemometrics 17 (2003) 313-322. <br> [6] M. Maeder, Y.M. Neuhold, Practical Data Analysis in Chemistry, Elsevier, Amsterdam NL, 2007. <br> [7] F. Visentin, S.I. Gianoli, A. Zogg, O.M. Kut, K. Hungerbühler, Organic Process Research & Development 8 (2004) 725-737.

Presented at:
100th Annual Meeting of the American Institute of Chemical Engineers (AIChE), Philadelphia (USA), November 16-21, 2008
Presented as an Oral contribution

 Record created 2012-04-17, last modified 2018-03-17

Publisher's version:
Download fulltextPDF
External links:
Download fulltextURL
Download fulltextURL
Download fulltextURL
Rate this document:

Rate this document:
(Not yet reviewed)