Leveraging Large Language Models for Intelligent Manufacturing: Opportunities and Challenges
3rd International Conference on Artificial Intelligence, Computer, Data Sciences, and Applications, ACDSA 2026, Boracay Island, Filipinler, 5 - 07 Şubat 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/acdsa67686.2026.11467595
- Basıldığı Şehir: Boracay Island
- Basıldığı Ülke: Filipinler
- Anahtar Kelimeler: Artificial Intelligence, Deep Learning, Machine Learning
- Dokuz Eylül Üniversitesi Adresli: Evet
Özet
Manufacturing operations depend on technical documentation-such as manuals, sensor logs, and operating procedures - yet this content remains underused due to unstructured formats and limited retrieval and reasoning capabilities. We investigate how large language models (LLMs) combined with retrieval-augmented generation (RAG) can unlock this latent knowledge. We present a modular pipeline for semantic under-standing and knowledge extraction from real-world industrial texts and construct a domain-specific dataset of 29 documents (≈ 730 pages), including installation guides, technical manuals, and product catalogs. Using an open-source LLaMA3 model via the Ollama framework, we systematically compare raw versus preprocessed corpora with 500 - and 1,000-token chunking. Accuracy is measured by cosine similarity between generated and reference answers over 70 questions; responses with similarity ≥ 0.6 (sentence-embedding-based) are deemed correct. The best configuration - document cleaning with 1,000-token segmentation - achieves an accuracy of 89%. We also synthesize recent RAG applications in manufacturing (e.g., multimodal inputs, ontology-guided retrieval, and quality-focused QA) to contextualize our findings. Overall, the results indicate that RAG-enabled LLMs can enhance decision support, fault diagnosis, and knowledge reuse in intelligent manufacturing, while offering a reproducible baseline for future research.