Life cycle assessment (LCA) is a widely used method for quantifying the environmental impacts of products and processes, yet the inventory analysis remains the most resource-intensive phase, often consuming up to 80% of a project’s effort. This phase requires the listing of all foreground flows and their manual mapping to background datasets, demanding specialised expertise and domain knowledge. Recent advances suggest that artificial intelligence, particularly large language models (LLMs), offer opportunities to accelerate and standardise LCA workflows by automating key tasks. However, existing LCA software platforms have not yet integrated these capabilities into cohesive and accessible tools. This work introduces ARIA (Artificial Intelligence for Sustainability Assessment), an open-source Python package that embeds LLMs into the LCA workflow to automate inventory mapping and impact assessment. Built on the Brightway framework, ARIA accepts structured input data describing materials, energy, or transport flows and applies a multi-stage matching process. It filters database entries based on location and flow type, expands search terms using the OpenAI API to propose semantically related alternatives, and uses prompt-driven reasoning to select the most context appropriate datasets. Once mappings are complete, ARIA performs impact assessment using any Brightway compatible method and visualises results for interpretation. Benchmark case studies in battery production demonstrate that ARIA can map a 16-flow inventory and perform impact assessment in less than 7 minutes, producing results consistent with curated datasets and representative of manual practice, though no direct timing comparison with manual modelling is performed. A battery recycling case study highlights current limitations of the software common to end-of-life modelling such as terminology alignment, retrieval under ambiguity, and the use of proxy inventories for certain waste treatment flows, outlining opportunities for further development.
Kallitsis et al. (Thu,) studied this question.