(229d) Machine Learning Enabled Life Cycle Assessment for Early-Stage Sustainable Process Design

Holistic assessment during early-stage process design is important for developing environmentally sustainable processes. The complexity in assessing sustainability arises from the multifaceted nature of the design problem, where the initial stages are open to interpretation and lack sufficient information, making it challenging to evaluate sustainability. As the design progresses, the potential for enhancing sustainability is further constrained by preliminary decisions, requiring the consideration of interdisciplinary factors including environmental impacts, health and safety, and social considerations (Argoti et al., 2019). Furthermore, impact assessment requires collecting and analyzing a wide range of data, including physicochemical properties, molecular structure, toxicity, and fate of these chemicals in the environment (Finnveden et al., 2009). Machine Learning (ML) algorithms are proving to be a promising solution to address these challenges, as they can predict the environmental impacts of novel chemicals with high accuracy even with limited availability of data (Karka et al., 2019). Thus, early-stage ML assessments enable informed decisions for sustainable process design.

The proposed study uses ML to predict the Midpoint and Endpoint categories of the Life Cycle Inventory (LCI) for chemicals. The midpoint LCIs comprise directly observable and measurable emissions such as ozone layer depletion, PM10 concentration, aquatic eutrophication, respiratory effects, non-renewable energy use, etc., whereas endpoint LCIs are the final consequences or damage that includes a combination of midpoint categories that contribute towards the same endpoint LCI indicator. Endpoint LCIs are comprised of four summary indicators including human health, ecosystem quality, climate change, and resource utilization. Thus, depending on the data availability, we can either predict midpoint or endpoint LCIs and utilize them in conducting the entire Life Cycle Assessment (LCA) for a chemical in its manufacturing phase (cradle-to-gate), use phase (gate-to-gate) and end-of-life (gate-to-grave/gate-to-cradle) phases.

Thus, our work presents a novel approach to conducting the complete LCA for a given chemical and its associated processes in the early design stages. Many ML algorithms were tested, and their results were compared, for example, eXtreme gradient boosting (XGBoost), artificial neural networks (ANN), and advanced Transformer Models such as Graph Neural Networks (GNNs). The LCI data for known chemicals were acquired from the Ecoinvent database that is used in widely accepted LCA software such as SimaPro and GaBi. The feature set comprised over 200 molecular descriptors and 23 thermodynamic properties for each chemical. A stepwise feature selection criterion was used to reduce the number of features and hyperparameter tuning was employed in the final ML model. The dataset was split into training, testing, and validation sets, and the model performance was evaluated on the test set using the R-squared, Root-Mean-Squared-Error (RMSE), and loss functions.

These ML-predicted LCIs were used to conduct complete LCAs for case studies and the results followed similar trends as observed in published work. This provides an accuracy check on the ML predictions, and the anticipation is that these ML models will enable LCI predictions for novel chemicals, thus, allowing early-stage LCA of the entire process.

References:

Argoti, A., Orjuela, A., Narváez, P.C., 2019. Challenges and opportunities in assessing sustainability during chemical process design. Curr. Opin. Chem. Eng. 26, 96–103. https://doi.org/10.1016/j.coche.2019.09.003

Finnveden, G., Hauschild, M.Z., Ekvall, T., Guinée, J., Heijungs, R., Hellweg, S., Koehler, A., Pennington, D., Suh, S., 2009. Recent developments in Life Cycle Assessment. J. Environ. Manage. 91, 1–21. https://doi.org/10.1016/j.jenvman.2009.06.018

Karka, P., Papadokonstantakis, S., Kokossis, A., 2019. Predictive LCA - a systems approach to integrate LCA decisions ahead of design, in: Computer Aided Chemical Engineering. Elsevier, pp. 97–102. https://doi.org/10.1016/B978-0-12-818634-3.50017-5