LLM-driven analysis of micro-pillar experiments for investigating fracture in Al-steel RSW joints

Analyzing the mechanical behavior of micro-pillars in resistance spot weld (RSW) joints between dissimilar materials, such as aluminum and steel, before and after corrosion is challenging due to their microstructural heterogeneity and its impact on fracture behavior. Conventional analyses are predominantly dependent on expert-driven interpretation and manual data processing, limiting scalability and consistency. Machine learning has been increasingly introduced into RSW to enhance prediction and processes optimization. However, its effectiveness remains constrained by data limitations and physical interpretability challenges. This study introduces a novel framework that integrates large language models (LLMs) to enable autonomous, high-fidelity analysis of micro-pillar images and associated mechanical data, such as engineering stress-strain curves. This framework combines image preprocessing pipelines, automated literature retrieval, and context-aware analysis to capture the critical features of crack initiation and propagation. By leveraging retrieval-augmented generation (RAG), this approach ensures that interpretative outputs remain grounded in validated physical phenomena and metallurgical principles, while uncovering insights into micro-pillar behavior.