This study proposes a two-step method for utilizing trouble information in design: (1) extracting component–phenomenon relationships and (2) structuring causal relationships between states. By leveraging generative AI, the method enables flexible and accurate extraction from natural language failure reports. The structured information is visualized in matrix form to support validation and causal analysis. A case study using failure data from the nuclear domain confirmed the effectiveness of the proposed approach.
Shimizu et al. (Wed,) studied this question.