(630e) Augmented Life Cycle Assessment to Cover Chemical Inventory Data Gaps

Addressing current challenges caused by climate change and the impact of human activities requires tools to evaluate and identify strategies to mitigate such impacts. Especially in the chemistry sector, which emits approximately 5 Gt CO₂-eq of greenhouse gases [1] each year with a wide variety of products, it is of increasing importance to evaluate the sustainability level of current production processes. Life cycle assessment (LCA) has emerged as the prevalent tool to quantify the impact of industrial processes, providing valuable and detailed insights on how to improve their sustainability performance, based on so-called life cycle impacts. Despite the great potential of LCA, it still remains a time- and data-intensive procedure, primarily due to data requirements, often resulting in other simpler methods to quantify impacts being used instead. In this context, alternative approaches, such as green chemistry process-level metrics [2] are feasible in an early design to obtain preliminary trends in terms of sustainability, but fail to provide the detailed insights of a thoroughly designed LCA [3]. For instance, process-level metrics such as the process mass intensity cannot distinguish between high impact reagents and reagents with lower life cycle impacts [4].

Current life cycle assessment (LCA) databases, such as Ecoinvent [5], include data for approximately 1,000 chemicals, primarily focusing on high-production-volume substances manufactured via continuous processes under high temperature and pressure. In contrast, reaction databases like Reaxys contain data on over 280 million chemicals, including reaction pathways and reactants, solvents and catalysts involved therein. This highlights the significant data gaps in existing LCA resources. This disparity becomes especially critical when considering the assessment of fine chemicals. Fine chemicals, often complex molecules used in the pharmaceutical industry, are typically synthesized through multi-step reactions under complex conditions, usually in batch processes. As a result, conducting LCAs for fine chemicals requires time-intensive process simulations [6] and access to a broad range of detailed data. The situation is further complicated by intellectual property constraints, as patents often fail to disclose full process information such as the solvent choice, reagents, and their respective concentrations [7].

Current approaches to closing data gaps in LCA databases for (fine) chemicals largely rely on data-driven methods based on molecular descriptors. Moreover, these methods require large volumes of high-quality input data based on current fossil routes [8], which makes it challenging to predict the impact of emerging routes based on renewable carbon, particularly when dealing with alternative pathways to produce the same chemical with a given molecular structure.

In this work, we propose a novel approach to streamline LCA calculations based on first-principles data augmentation. Through the use of mass-based allocation, energy estimations, and reaction networks, we are able to automate the calculation of environmental impacts for fine chemicals. This approach begins with the collection of life cycle impacts, e.g., warming potential (GWP), for preselected chemicals already modelled in detail (for example, retrieved from the Ecoinvent database). This allows us to create a corpus of chemicals that will be used later to perform the data augmentation task. Subsequently, a reaction network is built from Reaxys by iteratively querying reactions that link the chemicals in the corpus with others with unknown footprints. This procedure is repeated to progressively expand and densify the network, ensuring broad chemical coverage and connectivity. In the next step, mass-based allocation in accordance with ISO 14044 [9] is employed to propagate life cycle impacts, such as the GWP, throughout the reaction network, enabling the estimation of impacts for chemicals lacking initial life cycle data.

In addition to the mass-based allocation, energy estimates are incorporated to provide gate-to-gate energy impact estimations. The energy assessment covers the estimation of reaction energy for endothermic reactions, as well as the energy required for product separations. The reaction enthalpies are computed from the enthalpies of formation of reactants and products.

In terms of the separation energy estimation, the heuristics by Gani et al. [10] are applied to identify suitable separation technologies. Following this approach, separation technologies such as distillation, liquid-liquid extraction, or recrystallization are selected based on differences in the thermodynamic properties, such as boiling point, melting point, and solubility parameter, among the products in the considered reaction. All thermodynamic data are estimated using the ADF COSMO-RS implementation by the Amsterdam Modeling Suite [11], [12]. Subsequently, the energy demand is estimated for each technology. Finally, these energy requirements based on selected separation technologies and the reaction energy are translated into life cycle impacts using existing databases and lastly combined with the mass-based allocation impacts obtained from the corpus of chemicals and the associated stoichiometric information.

Preliminary results from three selected case studies, aniline, ethyl acetate, and phenol, demonstrate that GWP predictions using the proposed approach deviate by less than ±20% from the Ecoinvent values as a reference. Our method also provides intervals within which impacts of chemicals missing in environmental databases might fall. The said level of accuracy might suffice to support design decisions in early stages of process development where the screening of technologies takes place. Overall, the developed data augmentation framework serves as a streamlined LCA method, enabling the quantitative assessment of fine chemicals' sustainability level in a high-throughput and scalable manner, without relying on data-driven algorithms.

Future work will focus on incorporating additional input parameters to enhance further the accuracy of impact estimations, particularly by accounting for solvent use, yield losses, and refining the shortcut methods employed for energy calculations. Moreover, we will investigate how to identify and address deviations in energy estimates to further improve the robustness and accuracy of the proposed approach.

References

[1] J. Huo, Z. Wang, C. Oberschelp, G. Guillén-Gosálbez, and S. Hellweg, “Net-zero transition of the global chemical industry with CO2-feedstock by 2050: feasible yet challenging,” Green Chemistry, vol. 25, no. 1, pp. 415–430, 2022, doi: 10.1039/d2gc03047k.

[2] R. A. Sheldon, “Metrics of Green Chemistry and Sustainability: Past, Present, and Future,” Jan. 02, 2018, American Chemical Society. doi: 10.1021/acssuschemeng.7b03505.

[3] E. Lucas, A. J. Martín, S. Mitchell, J. Pérez-ramírez, and G. Guillén-gosálbez, “The need to integrate mass- and energy-based metrics with life cycle impacts for sustainable chemicals manufacture”.

[4] M. U. Luescher and F. Gallou, “Interactions of multiple metrics and environmental indicators to assess processes, detect environmental hotspots, and guide future development,” Green Chemistry, vol. 26, no. 9, pp. 5239–5252, 2024, doi: 10.1039/D4GC00302K.

[5] G. Wernet, C. Bauer, B. Steubing, J. Reinhard, E. Moreno-Ruiz, and B. Weidema, “The ecoinvent database version 3 (part I): overview and methodology,” International Journal of Life Cycle Assessment, vol. 21, no. 9, pp. 1218–1230, Sep. 2016, doi: 10.1007/s11367-016-1087-8.

[6] X. Hai et al., “Geminal-atom catalysis for cross-coupling,” Nature, vol. 622, no. 7984, pp. 754–760, 2023, doi: 10.1038/s41586-023-06529-z.

[7] G. Wernet, S. Conradt, H. P. Isenring, C. Jiménez-González, and K. Hungerbühler, “Life cycle assessment of fine chemical production: A case study of pharmaceutical synthesis,” International Journal of Life Cycle Assessment, vol. 15, no. 3, pp. 294–303, 2010, doi: 10.1007/s11367-010-0151-z.

[8] D. Zhang, Z. Wang, C. Oberschelp, E. Bradford, and S. Hellweg, “Enhanced Deep-Learning Model for Carbon Footprints of Chemicals,” ACS Sustain Chem Eng, vol. 12, no. 7, pp. 2700–2708, 2024, doi: 10.1021/acssuschemeng.3c07038.

[9] H.-E. P. on L. ILCD., International Reference Life Cycle Data System (ILCD) Handbook - General guide for Life Cycle Assessment. 2010. doi: 10.2788/38479.

[10] R. Gani and K. M. Lien, “SEPARATION PROCESS DESIGN AND SYNTHESIS BASED ON THERMODYNAMIC INSIGHTS,” 1995.

[11] C. C. Pye, T. Ziegler, E. Van Lenthe, and J. N. Louwen, “An implementation of the conductor-like screening model of solvation within the amsterdam density functional package - Part II. COSMO for real solvents,” Can J Chem, vol. 87, no. 7, pp. 790–797, Jul. 2009, doi: 10.1139/V09-008.

[12] “AMS 2024.1 COSMO-RS, SCM, Theoretical Chemistry, Vrije Universiteit, Amsterdam, The Netherlands.” Accessed: Apr. 07, 2025. [Online]. Available: http://www.scm.com/