Do you speak LCA? FAULDIER: A framework for large language model assisted Life Cycle Inventories in Life Cycle Assessment

Advances in Life Cycle Assessment (LCA) toward greater automation and methodological integration have intensified challenges in standardizing heterogeneous raw Life Cycle Inventory (LCI) data, which rarely aligns with LCI database nomenclature. Rule-based mapping approaches struggle with linguistic variations, typographical errors, unit inconsistencies, and location granularity mismatches. Furthermore, they fail to adapt automatically when data or terminology change. FAULDIER (Framework for lArge langUage modeL assisteD lIfe cyclE inventoRy) is proposed as a framework to bridge heterogeneities between raw LCI data and LCI database requirements. It aims to automate data transformation by resolving naming inconsistencies, classifying flow types, and harmonizing locations and units. By using LLMs, FAULDIER supports handling multilingual inputs, correcting typographical errors, resolving location granularity mismatches, and choosing proxies for missing processes. In a test scenario using the open LCI database FORWAST and a use case characterized by non-standardized multilingual entries, unit inconsistencies, and typographical errors, FAULDIER achieved approximately 57% process and elementary flow mapping accuracy (single-expert validated), with unit conversion error rates below 1%. Current limitations include LCI database constraints, LLM token limitations, performance variability of open-weight LLMs, mapping ability, and reproducibility across runs. Within these limitations, FAULDIER indicates the feasibility of LLM-assisted LCI construction for LCA modeling, particularly for non-standardized raw LCI data. Future work could focus on developing confidence metrics for mapped LCI data, optimizing LLM query efficiency, and expanding testing across additional LCI databases, use cases, and LLMs.