Banca de DEFESA: FELIPE LIMA DE MEDEIROS

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : FELIPE LIMA DE MEDEIROS
DATE: 14/01/2026
TIME: 09:00
LOCAL: Meets: meet.google.com/gtj-hejt-jsq
TITLE:

Exploration of Structures in the Asset Administration Shell Pattern with a Conversational Agent Based on Large Language Models

KEY WORDS:

Asset Administration Shell; AASX; semantic search; embeddings; Retrieval-Augmented Generation; Industry 4.0.

 

PAGES: 70
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUBÁREA: Eletrônica Industrial, Sistemas e Controles Eletrônicos
SPECIALTY: Automação Eletrônica de Processos Elétricos e Industriais
SUMMARY:

The Asset Administration Shell (AAS) standard plays a relevant role in the representation of industrial assets. However, its practical use is still limited by the structural complexity of AASX packages and by the dispersion of information across structured content, such as XML files, and embedded unstructured artifacts, such as PDF documents. In addition, there is a noticeable lack of tools that enable AAS assets to be organized, explored, and queried in a simple and accessible manner, especially for users without specialized technical knowledge.

This work proposes a methodology and a reference architecture to simplify access to, exploration of, and querying of AAS data through natural language, combining structured extraction, vector indexing, and Retrieval-Augmented Generation (RAG) with explicit

evidence traceability. The approach integrates structured and unstructured content from AASX packages into a query-oriented representation designed for semantic information retrieval, performs semantic segmentation and indexing of the data in a vector database, and generates answers grounded in the retrieved context, accompanied by their respective sources and origin metadata.

Validation is conducted on a set of public multi-manufacturer samples, and answer quality is estimated through automated evaluation based on large language models (LLM-as-a-judge), using criteria focused on factual correctness with respect to reference answers. The results indicate that the proposed methodology enables natural-language queries with traceability and transparency, facilitates access to the information contained in AAS packages, and suggests potential to make their exploration more accessible, while preserving adherence to the original content and the principles of standardization and interoperability of the AAS.

 

COMMITTEE MEMBERS:
Presidente - 1153006 - LUIZ AFFONSO HENDERSON GUEDES DE OLIVEIRA
Externo ao Programa - 1775264 - GUSTAVO BEZERRA PAZ LEITAO - UFRNExterna à Instituição - JULLIANA CAROLINE GONÇALVES DE ARAÚJO SILVA MARQUES