Assembly of Life Cycle Inventories using Physical Models and Supervised Learning to support Sustainable Process Development in the Pharmaceutical Industry

Abstract

A significant transformation is required in the pharmaceutical industry to contribute towards an environmentally sustainable global economy. This necessitates a fundamental shift in the feedstocks and processes that underpin current production. Here, a consistent approach to evaluating environmental impacts associated with novel synthesis pathways is essential. Life Cycle Assessment (LCA) is a methodology to holistically analyse the environmental impacts of goods and services. However, there are challenges in its application in early-stage process development: assembling Life Cycle Inventory (LCI) data is time consuming and prone to inaccuracies, mainly due to the lack of primary data or reliable proxy information. This thesis explores these challenges and demonstrates how utilisation of the vast amount of existing industrial process records can help overcome knowledge gaps. Part I investigates a case study of industrial-scale paracetamol manufacture from bio-waste derived β-pinene. An LCI of this alternative synthesis route is estimated based on detailed process models developed in Aspen Plus, and the associated environmental impacts are quantified. A “worst case” GWP of 58 kg CO2-eq./kg product was predicted, and comparative feedstock analysis showed that bio-waste derived β-pinene can be a cleaner feedstock than the benchmark feedstock benzene. However, the shortage of available tools to predict waste-specific impacts of treating industrial process effluents was identified as a key challenge for rigorous LCI assembly at the early design stage. Part II builds on this conclusion by applying supervised learning algorithms to historical process data from an industrial wastewater neutralisation system employed at a bio-pharmaceutical manufacturing site. A proof-of-concept predictive tool was developed to estimate gate-to-gate LCIs and calculate environmental impacts based on wastewater-specific characteristics and system parameters. Scenario analyses revealed significant variance in the environmental impacts associated treating different wastewater streams. The findings emphasise the importance of estimating waste-specific impacts during early-stage process development and support the hypothesis that utilising existing industrial records can address challenges associated with LCI data availability and reliability.