(371ab) Polyolefin Process Modeling and Monitoring

Polyolefins are one of the most widely used commodity polymers with wide applications in films, packaging and automotive industry. The polymerization process of producing polyolefins, including high-density polyethylene (HDPE), linear low-density polyethylene (LLDPE) and polypropylene (PP) using Ziegler-Natta catalysts with multiple active sites, is a complex and challenging task. In this study, we discuss two aspects of polyolefin process modeling. First, we present the methodology for kinetic parameter estimation of polyolefin processes from plant data in order to make a model that accurately predicts the plant performance. Next, we use the a dynamic polyolefin model to simulate plant data and use the data to demonstrate process monitoring techniques and to build data-based soft sensors to predict polyolefin quality.

Most of the studies on polyolefins process modeling over the years do not consider all of the commercially important production targets when quantifying the relevant polymerization reaction kinetic parameters based on measurable plant data. In our method, we present a general model to estimate kinetic parameters that have the most significant impacts on specific production targets, and to develop the kinetics using all commercially important production targets validated over many commercial polyolefin processes producing HDPE, PP and LLDPE using Ziegler-Natta catalyst. We then demonstrate an effective methodology for estimating the kinetic parameters to fit production targets in a computer-aided step-by-step procedure. We apply the efficient software tools within the Aspen Polymers, such as data fit, sensitivity analysis and design specification, for the kinetic estimation and modeling of the polyolefin processes. We use a validated dynamic polyolefin model to simulate plant data since we can vary many multivariate process parameters which is difficult in a real plant, to make a robust data-based model, which can be later validated on actual plant data.

The monitoring of polyolefin processes is an equally important aspect of process modeling. The process data provide an important basis for quality control since the process measurements are more frequent and precise, while the output measurements for polymer processes like molecular weight are sparse and noisy. Thus, data-based sensors derived from machine learning and multivariate statistical methods for polymer quality measurements are critical for process monitoring. The process variables we consider include the monomer feed flow rates, catalyst flow rate, hydrogen flow rates, operating conditions like temperature, pressure of reactors, flash, exchangers etc. The product quality variables we predict include polymer flow, molecular weight, polydispersity index, etc. We use both unsupervised and supervised learning methods for the analysis.

We apply the principal component analysis (PCA) to reduce the process data dimensions. We use anomaly detection techniques to detect outliers from process data. We also apply the partial least squares (PLS) technique to study the effect of input process variables on the product quality measures. We apply the traditional multivariate regression and multilayer neural networks for prediction. We also use ensemble learning methods like random forests and gradient boosting which combine multiple learners to improve accuracy. We compare the algorithms on the basis of accuracy of polymer quality predictions. We use the open source python and Aspen ProMV software for this data analysis.