Research on the optimization of seamless steel pipe cooling process based on large language models

To address technical challenges such as non-uniform cooling and microstructural delamination during the cooling process of seamless steel pipes, this study proposes an intelligent process parameter optimization approach that integrates a large language model with a domain-specific knowledge base. Built upon the open-source DeepSeek-R1-Distill-Qwen-14B model, two parameter-efficient fine-tuning strategies, such as LoRA and P-Tuning V2, are applied. Fine-tuning is performed using a curated knowledge base tailored to the steel pipe cooling domain. The jointly fine-tuned model achieves an accuracy of 0.871—an improvement of 7.7% over the base model—substantially enhancing the reliability of process parameter recommendations and the overall precision of cooling process optimization. By integrating LangChain to enable natural language interaction and process parameter recommendation, the proposed system is experimentally validated through studies on the effects of different cooling strategies on the microstructural properties of steel tubes. Experimental results demonstrate that, in terms of cooling uniformity, the cut-circle spraying method markedly outperforms vertical spraying, reducing the maximum inner wall temperature difference to 49 °C under continuous spraying at 60 kPa. Notable strength enhancements were observed across various spray pressures, with tensile and yield strengths reaching 1088 and 634 MPa, respectively, at 150 kPa. These improvements highlight a favorable balance between strength and toughness. Furthermore, all performance indices under 60 kPa cooling met API standards, underscoring the robustness of the proposed reusable engineering workflow and technical framework for generating and optimizing process parameters.