Zhou, Tianyu, Zou, Lai, Wang, Wenxi and Huang, Yun
2026.
Towards intelligent knowledge assistance in abrasive belt grinding via a retrieval-augmented generation chatbot with reliability support.
Computers in Industry
178
, 104479.
10.1016/j.compind.2026.104479
|
Preview |
PDF
- Accepted Post-Print Version
Available under License Creative Commons Attribution. Download (2MB) | Preview |
Abstract
Abrasive belt grinding is a knowledge-intensive machining process characterized by complex parameter coupling and heavy reliance on tacit experience. With the advance of intelligent manufacturing, fragmented and implicit domain knowledge poses challenges for effective learning and task adaptation. To address these issues, this study develops an intelligent knowledge-assistance chatbot enhanced by retrieval-augmented generation (RAG) and lightweight conflict filtering. A domain-specific evaluation set covering five question types is constructed to benchmark the performance of seven mainstream large language models, among which DeepSeek-R1 demonstrates the highest accuracy and stability, with an average score approximately 15.13% higher than the others. Based on this, a locally deployable RAG-based system is implemented, integrating bilingual semantic retrieval, cross-encoder reranking, and structured prompting, with a semantic conflict filtering module to enhance reliability. Experimental results show that the proposed local system BelGrindGPT achieves 96.67% of the performance of full-parameter API models within the present evaluation setting while using about 2% of the parameters, and improves average response quality by 9.98% compared to the base model of DeepSeek-R1, with the most substantial gains found in analytical reasoning, industrial background, and research frontier questions.
| Item Type: | Article |
|---|---|
| Date Type: | Publication |
| Status: | Published |
| Schools: | Schools > Engineering |
| Additional Information: | RRS policy applied |
| Publisher: | Elsevier |
| ISSN: | 0166-3615 |
| Date of First Compliant Deposit: | 11 May 2026 |
| Date of Acceptance: | 4 April 2026 |
| Last Modified: | 11 May 2026 12:30 |
| URI: | https://orca.cardiff.ac.uk/id/eprint/186490 |
Actions (repository staff only)
 |
Edit Item |
Dimensions
Dimensions