Machine Learning and Reaction Chemistry for Efficient Thermal Conversion of Waste Plastic

Belden, Elizabeth

Etd

Machine Learning and Reaction Chemistry for Efficient Thermal Conversion of Waste Plastic

Public Deposited

Waste plastic is an ever-growing environmental concern with millions of tons of plastic entering the environment every year. The strongly bonded carbon backbone of plastics combined with their low recyclability results in an accumulation of this recalcitrant polymer with limited viable options for reuse. Thermal depolymerization technologies are one of the promising methods capable of chemically recycling waste plastics. Hydrothermal liquefaction (HTL) and pyrolysis use high temperatures, with or without the presence of oxygen, respectively, to break the strong covalently bonded polymer chain to create smaller molecules such a fuel grade oils and platform chemical monomers. Regardless of the thermal method, research has shown that thermal depolymerization performance is highly dependent on feedstock composition, which requires experimental studies of each new feedstock. In this work, computational and experimental studies are combined to provide a detailed understanding of feedstock compositional effects and thermodynamic and economic potential of thermal conversion of waste plastic to fuel and chemicals. The use of thermodynamic Monte Carlo modeling revealed the thermodynamic feasibility of self-powered cleanup of oceanic waste plastics via conversion into marine fuel, termed “Blue Diesel”. In addition to being thermodynamically feasible, it was shown the existing plastic present in the Great Pacific Garbage Patch could be entirely cleaned within 50 years, if new plastic input to the ocean can be eliminated. Stopping plastic waste in rivers before they can enter the ocean is a major step to turning off plastic input to the ocean. To understand the thermodynamic potential of river-based conversion systems, machine learning models were developed such that the oil yields from any new feedstocks can be predicted and incorporated into thermodynamic models. This work indicated that even in mildly polluted rivers, river-based conversion systems were thermodynamically feasible and that machine learned models could be used to analyze the impact of feedstock composition on reaction performance. Along with thermodynamic potential, machine learning models were also used to predict the oil yield potential of land-based pyrolysis of waste plastic to oil in all 50 states. These predictions were incorporated into a technoeconomic analysis, showing an average minimum selling price of $170/ton. These models allow for the rapid analysis of new feedstocks without the necessity of new experimental studies. Further improvement to the economic potential of thermal conversion of waste plastics was studied experimentally through the creation of high value oxygenated products from radical induced-HTL of polystyrene. These experiments combined with traditional HTL experiments began to shed light on the effect of water on the depolymerization mechanism of polystyrene, a mechanism that is not currently known in the literature. Economic potential was further analyzed computationally to understand depolymerization kinetics of model compounds to target high value products. The work done in this thesis expands the fields knowledge of the exceptional potential of thermal conversion technologies to change how the world handles end of life waste plastics.

Creator