Leveraging Machine Learning for Life Cycle Assessment: Predicting Environmental Impacts and Scaling Sustainable Chemical Production

Life cycle assessment (LCA) is a systematic analysis of potential environmental impacts of raw materials, products, or services during their entire life cycle. Initial design stages are inherently complex and often lack comprehensive information, posing challenges for effective LCA evaluations. Machine Learning (ML) emerges as a valuable solution to address these challenges. ML algorithms, particularly Artificial Neural Networks (ANN) and XGBoost, prove effective in predicting environmental impacts of new chemicals with limited data. This study focuses on comparing ML models trained on data sets consisting of a feature set (provided inputs) and a label set (predicted outputs). The feature set is comprised of both thermodynamic properties and molecular descriptors of the chemicals. The label dataset consisted of four endpoint and twelve midpoint impact assessment metrics. These assessment metrics are used in determining a chemical’s impact on human health, resource utilization, ecosystem quality, and climate change. The effectiveness of the ANN algorithm was tested using a dataset of 500 points, divided into training, testing, and validation sets.

Current and future work aims to expand this research towards technology scale-up analysis, focusing on the transition from lab-scale studies to industrial-scale production of greener chemicals. By examining case studies in sectors such as pharmaceuticals and specialty chemicals, we are collecting detailed data on costs, emissions, global warming potential, and energy consumption. Additional case studies will be simulated in programs such as ASPEN/SuperPro followed by Simapro to fill in missing data if necessary. These metrics will serve as inputs for algorithms to identify common scalability indices for environmental emissions.