Luca Dadone
Luca Dadone has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Andrea Bioddo
Independent Researcher, Italy
1 shared publication - Gaia Falchi
University of Rome Tor Vergata, Italy
1 shared publication - Mirco Lecce
University of Rome Tor Vergata, Italy
1 shared publication - Vittorio Cesarotti
University of Rome Tor Vergata, Italy
1 shared publication
Latest Publications
- Open Access
- 3 November 2026
- Published by MDPI AG in Preprints.org
Enabling Smart Maintenance in SMEs via Domain-Specific RAG: An Industrial Field Study
Reactive maintenance management in Small and Medium-sized Enterprises (SMEs) is often hindered by technical knowledge fragmentation and reliance on senior experts. Although generative Artificial Intelligence (AI) offers new perspectives for decision support, adopting generalist models introduces operational risks linked to hallucinations and a lack of specific context. This study analyses the effectiveness of a domain-specific Retrieval-Augmented Generation (RAG) framework, designed to encode historical corporate know-how into reliable operating procedures. Through a field validation conducted on a significant sample of real assistance requests, the research compares the proposed system with a generalist Large Language Model. The methodology adopts an evaluation protocol based on the LLM-as-a-judge technique, validated through analyses conducted with a maintenance engineer, to ensure accuracy. The results highlight how targeted context integration reduces factual inaccuracies, ensuring superior technical relevance. Such accuracy emerges as a fundamental enabling factor to speed up fault diagnosis, laying the foundations for intervention time optimization and greater overall maintenance process effectiveness.
- 3 November 2026
- Published by MDPI AG in Preprints.org
Open AccessEnabling Smart Maintenance in SMEs via Domain-Specific RAG: An Industrial Field Study
Reactive maintenance management in Small and Medium-sized Enterprises (SMEs) is often hindered by technical knowledge fragmentation and reliance on senior experts. Although generative Artificial Intelligence (AI) offers new perspectives for decision support, adopting generalist models introduces operational risks linked to hallucinations and a lack of specific context. This study analyses the effectiveness of a domain-specific Retrieval-Augmented Generation (RAG) framework, designed to encode historical corporate know-how into reliable operating procedures. Through a field validation conducted on a significant sample of real assistance requests, the research compares the proposed system with a generalist Large Language Model. The methodology adopts an evaluation protocol based on the LLM-as-a-judge technique, validated through analyses conducted with a maintenance engineer, to ensure accuracy. The results highlight how targeted context integration reduces factual inaccuracies, ensuring superior technical relevance. Such accuracy emerges as a fundamental enabling factor to speed up fault diagnosis, laying the foundations for intervention time optimization and greater overall maintenance process effectiveness.