APPROPRIATE Life Cycle Assessment: A PROcess-Specific, PRedictive Impact AssessmenT Method for Emerging Chemical Processes
OPEN ACCESS
Loading...
Author / Producer
Date
2023-06-26
Publication Type
Journal Article
ETH Bibliography
yes
Citations
Altmetric
OPEN ACCESS
Data
Rights / License
Abstract
A sustainable chemical industry needs to quantify its emissions and resource consumption by life cycle assessment (LCA). However, LCA requires detailed mass and energy balances, which are usually not available at early process development stages. Here, we introduce a framework (A PROcess-specific, PRedictive impact AssessmenT method for Emerging chemical processes, APPROPRIATE) to provide a fully automated, predictive LCA framework for the early phases of process development. Based on Gaussian Process Regression, the framework is already applicable at Technology Readiness Level 2. To overcome the limited LCA data availability, we employ an encoder–decoder network in combination with transfer learning to achieve a latent representation as a condensed molecular descriptor. We further propose to integrate not only molecular but also process descriptors, e.g., the stoichiometric sum of the reactants’ impacts. Thereby, we can distinguish between process alternatives and incorporate changes in the background systems. The framework is compared to state-of-the-art predictive LCA approaches and shows increased prediction accuracy in terms of the coefficient of determination of R2 = 0.61 for the global warming impact compared to an R2 = 0.3 in former studies. Highly relevant features are the stoichiometric sum of the reactants’ impacts and the condensed molecular descriptors. APPROPRIATE supports decision making in early process development stages by allowing the distinction between process alternatives and quantifying predictions’ uncertainty.
Permanent link
Publication status
published
External links
Editor
Book title
Journal / series
Volume
11 (25)
Pages / Article No.
9303 - 9319
Publisher
American Chemical Society
Event
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Graph neural networks; Latent space; Autoencoder; Gaussian process regression; Automatizedflowsheeting
Organisational unit
09696 - Bardow, André / Bardow, André