(245a) Polyolefin Property Estimation using Process Modeling and Machine Learning in Industry

Polyolefins are one of the most widely used commodity polymers and the modeling of polyethylene, polypropylene etc. is challenging because of their complex thermodynamics and kinetics. The estimation of polyolefin properties is critical due to its wide applications in different domains like films, packaging and automotive industry. In this study, I showcase the application of process modeling and machine learning for accurate estimation of polyolefin properties in industry. I demonstrate the utility of Aspen Plus process modeling software, open source Machine Learning tools and other Aspen data-based software like Aspen AI Model Builder for this analysis.

For estimation of physical properties of polyolefins, it is important to use the right property methods to characterize the phase equilibrium. The property methods can be Activity Coefficient Models or based on Equation of State and there are specific guidelines for the selection of an appropriate polymer property method for modeling a specific industrial polyolefin process. The thermodynamic property parameters can be regressed from experimental vapor liquid equilibrium data which can then be used to directly estimate some thermophysical properties of polymers like density, viscosity, heat capacity, thermal conductivity etc. The thermodynamic models can then be combined with reaction kinetics to build first-principle process models of industrial polyolefin processes¹.

Machine Learning models have been used to develop data-based sensors for predicting some polymer properties like Melt Index since the polymer quality measurements are less frequent, compared to continuous process measurements. The combination of First-Principle-based models and Data-based models can be even more useful in prediction of these polymer properties since they are not only accurate but also give scientifically consistent predictions². The Hybrid models combining First-Principles and Data-based process models with a purely data-based Machine Learning model can be used to develop more accurate and physically consistent predictions of some Polyolefin properties. In this study I discuss in detail these Hybrid models and the new advancements in Machine Learning in process industry for estimation of polymer properties.

1. Sharma, N. and Liu, Y.A., 2019. 110th anniversary: an effective methodology for kinetic parameter estimation for modeling commercial polyolefin processes from plant data using efficient simulation software tools. Industrial & Engineering Chemistry Research, 58(31), pp.14209-14226.

2. Sharma, N. and Liu, Y.A., 2022. A hybrid science‐guided machine learning approach for modeling chemical processes: A review. AIChE Journal, 68(5), p.e17609.