Knowledge-Grounded Nut Inspection and Report Generation via Two-Stage Retrieval-Augmented Language Models
IEEE Access · 2026 · 14 · 94167-94175
DOI: 10.1109/access.2026.3703193 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Kuei-Jung Hung und 3 weiteren Autor:innen. Die Arbeit nennt 18 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Science and Technology Council, Taiwan.
Abstract (Originalsprache)
In smart manufacturing environments, generating standardized and traceable inspection reports remains a challenging task due to the gap between automated inspection systems and engineering-compliant documentation. This paper proposes a knowledge-grounded framework for nut inspection and structured report generation by integrating visual inspection, specification retrieval, and language-based reasoning. Multi-view internal thread images are first analyzed using a classification model to determine nut grade and material. A two-stage retrieval strategy is then employed to accurately identify the corresponding standard table and extract engineering specifications. In addition, visual analysis is incorporated to provide supplementary descriptions of surface conditions. The final inspection report is generated in a structured format by a language model based on the integrated information. Experimental results demonstrate that the proposed method achieves full Field-level Exact Match while also improving BLEU and ROUGE scores, indicating both high factual correctness and strong textual consistency. The results further show that retrieval strategy plays a more critical role than model scale in structured report generation. The proposed framework provides a practical solution for automated inspection reporting in manufacturing environments.
Förderung
- National Science and Technology Council, Taiwan
Zitieren
APA
Hung, K., Lai, P., Wu, B., & Perng, J. (2026). Knowledge-Grounded Nut Inspection and Report Generation via Two-Stage Retrieval-Augmented Language Models. IEEE Access, 14, 94167-94175. https://doi.org/10.1109/access.2026.3703193
Vancouver
Hung K, Lai P, Wu B, Perng J. Knowledge-Grounded Nut Inspection and Report Generation via Two-Stage Retrieval-Augmented Language Models. IEEE Access. 2026;14:94167-94175. doi:10.1109/access.2026.3703193
BibTeX
@article{Hung2026,
author = {Hung, Kuei-Jung and Lai, Peng-Wen and Wu, Bing-Fei and Perng, Jau-Woei},
title = {Knowledge-Grounded Nut Inspection and Report Generation via Two-Stage Retrieval-Augmented Language Models},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {94167--94175},
doi = {10.1109/access.2026.3703193}
}
EUVIMED