Parallel computing as a tool for efficient modeling of the high-pressure polymerization of ethylene in tubular reactors.

DIETRICH, MAIRA L.; ASTEASUAIN, MARIANO; SARMORIA, CLAUDIA; BRANDOLIN, ADRIANA

Buenos Aires

Congreso; 11th World Congress of Chemical Engineering; 2023

Asociación Argentina de Ingeniería Química

The high-pressure radical polymerization of ethylene in tubular reactors to produce low-density polyethylene (LDPE) is a well-known technology that has been used for decades. However, it is still a subject of theoretical study. Different grades of LDPE with different end-use properties may be obtained in a continuous process by changing the reactor operating conditions. These differences in end-use properties are directly related to the molecular structure of the polymer grade. This means that distinct LDPE grades may have different molecular weight distributions (MWD) and long- and short-chain-branching distributions (LCBD and SCBD, respectively). In particular, the amount and type of long-chain branches (LCB) have a direct effect on rheological behavior and consequently on the performance during the final processing of the material.1 Therefore, it is important for LDPE producers to find the relationships between operating conditions, molecular structure, and end-use properties. Finding these connections experimentally is usually costly, so mathematical models have become a practical and safe alternative.In this work, we present an update of a deterministic mathematical model of the high-pressure polymerization of ethylene in tubular reactors previously developed by the authors.2 The deterministic approach consists of derivating the population balance equations (PBEs) of the polymeric species in the polymerization system using the probability generating function (pgf) technique. This model can predict molecular properties such as the MWD, the bivariate molecular weight-short-chain branching distribution (MWD-SCBD), the bivariate molecular weight-long-chain branching distribution (MWD-LCBD), the number of branches per 1000 carbon atoms distribution (LCB/1000C), the branching index distribution (g), and rheological properties such as the complex viscosity () and the melt index (MI).The number of model equations is large: 4 620 108 differential equations. A computational tool that allows speeding up the execution of a large model like this is parallel computing. This tool involves dividing the set of equations into independent groups to solve them concurrently using the different CPU processors. Conveniently, the parallelization of a pgf model is straightforward. The model presented in this work was implemented in Julia, an open-source programming language with built-in functions for code parallelization. The model running time is short. Results are compared with experimental data of LDPE samples obtained in an industrial reactor.