Empowering knowledge graphs with hybrid retrieval-augmented generation for the intelligent mix scheme of mass concrete
To achieve precise multi-objective mix design of dam concrete-targeting strength, durability, and crack resistance-this study proposes a Dam Concrete Mix Proportioning intelligent generation System (DCMPS) based on a Retrieval-Augmented Generation (RAG) framework. Firstly, a total of 1723 mix propor...
Saved in:
| Main Authors: | , , , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Case Studies in Construction Materials |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2214509525007776 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | To achieve precise multi-objective mix design of dam concrete-targeting strength, durability, and crack resistance-this study proposes a Dam Concrete Mix Proportioning intelligent generation System (DCMPS) based on a Retrieval-Augmented Generation (RAG) framework. Firstly, a total of 1723 mix proportioning records collected from 15 hydropower stations were structurally extracted using a customized data transformation tool. A knowledge graph (KG) was then constructed on the Neo4j platform, encompassing entities such as environmental conditions, material compositions, and performance indicators. Secondly, the Moka Massive Mixed Embedding (M3E) model was employed to encode the KG into a vector database, enabling accurate retrieval of relevant mix proportioning cases via cosine similarity calculation. Finally, based on a hierarchical reasoning architecture driven by prompt engineering, DCMPS retrieves relevant cases from the vector database using a semantic search mechanism. A large language model (LLM) conducts comparative analysis on the retrieved cases to generate mix design schemes tailored to specific requirements. The DCMPS’s performance was validated using a test dataset, The results demonstrate that the system outperforms the larger-parameter native LLM in overall evaluation on the test dataset, confirming that knowledge augmentation significantly enhances the performance of smaller models. Case studies indicate that the recommended mix designs closely align with the target performance requirements, and the comparison of relevant cases along with the analysis of material influence mechanisms enhances the interpretability of the results. This DCMPS provides effective technical support for intelligent analysis and engineering decision-making in concrete mix proportioning. |
|---|---|
| ISSN: | 2214-5095 |