Emulsion polymerization is often conducted in batch and semibatch reactors. In a typical batch recipe, monomer and surfactant are mixed with water, the mixture is agitated and warmed to reaction temperature and, a free radical initiator is added to begin the polimerization process.

A batch operation is considered successful if process variables remain within acceptable limits during batch trajectory and the product has a desirable quality. As changes in the operating conditions during critical periods affect product quality, the implementation of on-line monitoring and fault diagnosis techniques for emulsion polymerization allows to detect faults and take correction actions prior the completion of the run.

Several Multivariate Statistical Control techniques have been developed to perform on-line monitoring for discontinuous processes. Multiway Principal Component Analysis (MPCA) was developed by Nomikos and MacGregor (1994) as an extension of the classic Principal Component Analysis (PCA) to batch processes data. Furthermore, the same authors (Nomikos and MacGregor, 1995) proposed Multiway Partial Least Square (MPLS) to be used if product quality data are available. For these techniques, the statistical tests are formulated assuming data follow a normal distribution. However data matrix may contain non-Gaussian observations.

Recently Independent Component Analysis (Hyvarinen, 2001) has been applied to obtain a set of latent variables, called Independent Components (ICs), which are assumed to be non-Gaussian and mutually independent. It is considered that observations are linear or non-linear mixtures of the ICs, where the mixing process is unknown. ICA extracts the ICs as well as the mixing matrix from the observed data. This technique provides more meaningful information for non-Gaussian data than PCA.

Batch process data contain non-Gaussian distributed data due to, for example, ramp and step changes. Consequently a better monitoring performance for batch processes is achieved using Multiway Independent Component Analysis (Yoo et. al., 2004), MICA. For some published examples, this technique provides better results for on-line monitoring than MPCA because MICA uses statistical tests which are formulated considering data are non-Gaussian distributed.

In this work a MICA model is implemented for monitoring a Styrene Polymerization semi-batch reactor. Data are obtained by simulation using a non-isothermal reactor model that includes the following equations: a) mass balances for initiator, surfactant, monomer, and radical and polymer molecules; b) population balances; c) energy balance d) expressions to calculate: the average radical number per particle, radical entry into the particles, radical entry into micelles, radical desorption from particles, monomer conversion, monomer concentration in particles, monomer concentration in the aqueous phase, particle growth rate, total reactor volume, etc. The model also includes molecular weight calculations. gPROMS code environment (Process System Enterprise, Ltd.) is employed for model implementation. Simulations are run assuming closed-loop control of temperature. Initiator, surfactant and monomer are added in open-loop mode. Results are validated using experimental data provided in the literature.

Comparative studies are performed to assess the monitoring performance of MPCA and MICA, using different unfolding modes of the data matrix, for a wide variety of faults sources.

References Nomitos, P.; MacGregor, J. F., 1994. Monitoring Batch Processes Using Multiway Principal Component Análisis, AIChe Journal, 40, 8, 1361-1375. Nomitos, P.; MacGregor, J. F., 1995. Multiway Partial Least Squares in Monitoring Batch Processes, Chemometrics & Intelligent Laboratory Systems, 30, 97-108. Hyvarinen, A., Karhunen, J., Oja, E., 2001. Independent Component Analysis, Wiley USA. Yoo, C., Lee, J., Vanrolleghem, P., Lee, I., 2004. On-line Monitoring of Batch Processes using Multiway Independent Component Analysis, Chemometrics & Intelligent Laboratory Systems, 71, 151-163.