Description
Surface texture plays a critical role in monitoring manufacturing processes and predicting the functional performance of workpiece surfaces, particularly in high value-added manufacturing. Over the past two decades, the language of surface texture specification—one of the key components of the ISO Geometrical Product Specifications (GPS)—has evolved significantly. This evolution has been driven by advancements in data acquisition technologies, data processing algorithms, and the introduction of new surface parameters. Furthermore, surface texture specifications are now being applied to surfaces generated by emerging manufacturing technologies. As a result, the development and application of surface texture specifications have become increasingly complex. Designers often face challenges in decision-making due to the rapid pace of knowledge updates and the intricate nature of the domain.To address these challenges, this project aims to develop an intelligent system for surface texture specification by leveraging generative artificial intelligence (AI). The proposed framework will incorporate large language models (LLMs), prompt engineering, retrieval-augmented generation (RAG), and graph-based RAG (see Fig. 1). Users will interact with the system through natural language, while prompt engineering will guide effective query formulation to improve the relevance and quality of LLM-generated responses. The system will be enhanced with a knowledge graph—constructed from expert input and relevant documentation—which will serve as a foundation for the graph RAG module. This will allow the system to retrieve context-rich information and generate more accurate and informed responses. Additionally, a continuous learning cycle will be implemented, enabling the knowledge graph to be updated dynamically based on user feedback and newly added information. The system is expected to significantly advance surface texture specification via enabling intuitive, natural language interaction and delivering context-aware, expert-informed responses through AI-driven retrieval and generation.
| Period | 12 Jun 2025 |
|---|---|
| Held at | Politecnico di Milano, Italy |
| Degree of Recognition | International |