A new framework uses large language models to improve the accuracy of maintenance records for wind turbines. The system extracts relevant information from unstructured text and assigns structured semantic fields to each record. This process was applied to a dataset of 16,316 maintenance records from 280 turbines, resulting in improved labels and evidence profiles that can be used for reliability analysis and failure mode identification. --- Why it matters: This framework matters because it enables more accurate data-driven decision-making in wind turbine maintenance, which is essential for optimizing lifecycle expenditure and extending asset life. It also demonstrates the potential of large language models to improve the quality of legacy datasets. Source: https://arxiv.org/abs/2605.31281