Analysis of causes of welding defects in bridge weathering steel based on large language models

Abstract

The analysis of the causes of welding defects in bridge weathering steel necessitates a multifaceted approach integrating alloy element influences, crack control, and parameter optimization, as single-perspective methodologies inadequately address root causes and hinder effective solution development. To address these challenges, a large language model-based method for analyzing the causes of welding defects in bridge weathering steel is proposed. This method first integrates information from different welding perspectives through a multi-perspective associative memory mechanism and employs a hybrid retrieval strategy to retrieve factual memory and historical information, enabling precise recall of relevant content and providing comprehensive support for problem-solving. Second, a ”inhibition-cognition” task optimization strategy refines the problem-solving process by suppressing irrelevant information, decomposing tasks, and iteratively revising through cognitive simulation, thereby establishing a clear and efficient problem-solving pathway. Finally, the accuracy and consistency of sub-task processing are ensured by an expert-guided task verification meta-prompting method, where dynamic closed-loop validation is incorporated and expert knowledge is fused. Quantitative results demonstrate that the proposed method achieves consistent improvements in both ROUGE-L and BERTScore metrics across different models, while expert evaluations further confirm its exceptional performance in key dimensions such as rationality and comprehensiveness. This method provides a novel approach for analyzing the causes of welding defects in bridge weathering steel, playing a critical role in enhancing the accuracy and efficiency of defect analysis.