Knowledge graph-augmented retrieval for crack propagation analysis and strength prediction in layered rock materials

Fu J., Li B., YAYLACI M.

STRENGTH OF MATERIALS, 2026 (SCI-Expanded, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1007/s11223-026-00952-z
  • Dergi Adı: STRENGTH OF MATERIALS
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
  • Recep Tayyip Erdoğan Üniversitesi Adresli: Evet

Özet

Finding effective knowledge in fracture and rock engineering is very challenging because the relevant data are distributed across different and specialized sources. This study investigates a framework for information retrieval that combines a knowledge graph (KG-RAG) with a domain-adapted 14 B-parameter large language model for mechanics-oriented problem solving. This framework is evaluated by applying it to a series of laboratory data involving multi-material specimens (hard-soft-hard layers with weak interlayers). In these specimens, interlayer thickness and pre-existing cracks with varying inclinations are considered critical variables in this evaluation. This system supports experimental design, numerical method selection, model construction, and micro-parameter calibration. Previously reported experimental and numerical results are used as references. In this study, artificial intelligence is used to interpret the mechanical behavior and support decision-making. The calibrated models reproduce the stress-strain curve behavior and failure mechanisms with high accuracy. The results indicate the potential of KG-RAG to provide good technical support for analyzing fracture mechanisms, crack propagation, and the strength of cracked specimens Before claims of higher accuracy or reduced hallucination can be made, controlled experiments are required to compare the proposed framework with conventional LLM and standard RAG baselines.