Preprint has been published in a journal as an article
DOI of the published article
Preprint / Version 1

Application of Multivariate Statistics to Optimizing Polyolefin Manufacturing


  • Niket Sharma Virginia Tech



Statistical Learning, Machine Learning


In this chapter, we delve into the sophisticated realm of multivariate statistical methods, focusing on Principal Component Analysis (PCA) and Projection to Latent Structures (PLS), as pivotal tools for unraveling the complexity of process data analytics. By anchoring these statistical techniques within the framework of polyethylene manufacturing processes, we aim to illuminate their exceptional utility and novelty in addressing the multifaceted challenges inherent in process optimization and quality control.

The discourse begins by introducing PCA, not merely as a statistical tool, but as a fundamental cornerstone for the analytical examination of process variables. Through a meticulously designed workshop, we demonstrate the application of PCA in dissecting the intricate web of variables influencing the quality and conversion rates of Low-Density Polyethylene (LDPE) production in a two-zone tubular reactor. The integration of Aspen ProMV as a practical tool for PCA applications exemplifies the seamless bridge between statistical theory and industrial application, emphasizing the method's accessibility and relevance to both academia and industry.

Transitioning to PLS, the chapter articulates its differentiation from PCA by its ability to simultaneously handle datasets comprising both process variables (X) and product quality variables (Y), offering a holistic view of the manufacturing process. Through pragmatic workshops, we showcase PLS's robustness in application to challenges such as melt index prediction and causal analysis in High-Density Polyethylene (HDPE) manufacturing, underscoring its adaptability to complex industrial datasets, including those with measurement time lags.

The exploration extends to the nuanced application of these multivariate statistical methods to batch polymer processes. Here, we introduce a novel batch-wise unfolding approach via multiway PCA and PLS, expanding the frontier of statistical applications in process data analytics.

This chapter transcends the conventional boundaries of statistical applications, highlighting the transformative impact of PCA and PLS in the domain of process data analytics. It aspires to foster a deeper understanding and appreciation of these statistical methods, encouraging their broader adoption and adaptation in optimizing manufacturing processes and enhancing product quality. This contribution not only reaffirms the critical role of advanced statistical techniques in the scientific community but also underscores their practical significance in improving industrial operations and outcomes.

This is a preprint version of a chapter from our book - Integrated Process Modeling, Advanced Control and Data Analytics for Optimizing Polyolefin Manufacturing. Please cite the original work if referenced [31,32].


Download data is not yet available.