A Hybrid Science-Guided Machine Learning Approach for Modeling Chemical and Polymer Processes

Abstract

This chapter provides a comprehensive examination of hybrid modeling in chemical and polymer processes, employing a science-guided machine learning (SGML) approach to fuse scientific knowledge with data analytics. We introduce the concept of hybrid SGML and outline our motivation for exploring this innovative approach. A critical review of the broad applications of SGML in chemical engineering highlights the growing complexity and diversity in methodologies, making it challenging for newcomers to navigate the field. To address this, we offer a systematic classification of hybrid SGML methodologies, distinguishing between models where machine learning complements scientific understanding and vice versa. We delve into various applications of machine learning to augment science-based models, discussing direct serial and parallel hybrid modeling, inverse modeling, reduced-order modeling, and the quantification of uncertainty in process models, including the discovery of process governing equations. Each category is explored in detail, evaluating their requirements, strengths, and limitations, and suggesting potential areas of application with specific focus on polyolefin manufacturing. Similarly, we examine how scientific principles can enhance machine learning models, discussing the design, learning, and refinement processes. The study discussing the challenges and opportunities that lie ahead for the hybrid SGML approach in the modeling of chemical and polymer processes, signaling a promising direction for future research and application in this interdisciplinary field.   

This is a preprint version of our chapter 11 [133] from the book - Integrated Process Modeling, Advanced Control and Data Analytics for Optimizing Polyolefin Manufacturing. Please cite the original work if referenced [144] and is also an extended version of the study [131].