Combined Digitalization and Machine Learning Efforts: the α-Olefin Polymerization High-Throughput Experimentation Showcase

In the highly competitive world of α-olefin polymerization, innovation is driven by optimizing reaction processes, developing new polyolefin grades and enhancing the catalyst performance [1]. The emergence of single-site catalysts in the past decades have opened the route to specialty polymer grades due to the unique tuneability of these catalysts in terms of their chemical nature as well as their single center polymerization type behavior characterized by a narrow molecular weight distribution [2]. In the early 2000’s, high-throughput experimentation (HTE) was introduced to the field of polyolefin catalysis by a company called Symyx® [3]. Through a combined approach of high-throughput combinatorial chemistry, polymerization and analysis, it was now possible to vastly speed up catalyst and process conditions discovery as well as to immensely scale-down on materials use and costs. The University of Naples, Laboratory of Stereoselective Polymerization group has been operating such a HTE workflow for well over a decade [4a,b]. This includes a core module capable of combinatorial chemistry, the Parallel Pressurized Reactors (PPR) that can screen 48 different combinations of catalysts, co-catalysts, additives and process conditions per run and down-stream analytics including a high-throughput Gel Permeable Chromatography (GPC), Crystallization Elution Fractioning (CEF) and 1H/13C-Nuclear Magnetic Resonance (NMR). Whereas such a HTE workflow has proven to be highly successful in designing novel molecular catalyst structures for advanced polyolefin grades, there is still a wealth of opportunities available by combining such a HTE workflow that generates vast amounts of data with machine learning (ML) [5-7]. In this work, a digitalization strategy will be shown to both extract legacy raw experimental data as well as provide a path for to-be-generated data that can be directly used as input for ML efforts. Such ML efforts can be directed towards modelling kinetic parameters, Quantitative Structure Activity/Properties Relationships (QSAR & QSPR) as well as removing human bias in the downstream analytical processing. Finally, the idea is that such digitalization strategies could be employed to any type of high-throughput experimentation workflow.