Monitoring and Process Parameter Optimization of Vanadium–Nitrogen Alloy Production Process by Integrating Deep Reinforcement Learning and Domain Knowledge
Processes · 11.11.2026 · 14 (22) · 3644
DOI: 10.3390/pr14223644 ↗Kurz gefasst
Fachartikel in der Zeitschrift Processes, online erschienen am 11.11.2026. Verfasst von Xiangyu Zhang und 4 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Department of Science and Technology of Shandong Province, Shandong Key Laboratory of Intelligent Manufacturing Technology for Advanced Power 789 Equipment.
Abstract (Originalsprache)
The continuous growth in market demand for vanadium–nitrogen alloys is driving vanadium–nitrogen alloy enterprises to accelerate their digital and intelligent upgrades. Currently, the production process of vanadium–nitrogen alloys lacks real-time and reliable information sensing methods and a reasonable dynamic optimization mechanism for process parameters, resulting in low production transparency and difficulty in ensuring product quality. To address these issues, this paper proposes a method for monitoring the production process and optimizing process parameters of vanadium–nitrogen alloys, combining deep reinforcement learning (DRL) with domain knowledge. Firstly, through nitridation reduction experiments and computational fluid dynamics numerical simulations, the correlation between process parameters and product quality is systematically analyzed, and domain knowledge is extracted to guide optimization. Secondly, the process parameter optimization problem is modeled as a Markov decision process (MDP), and a DRL framework based on the Soft Actor–Critic (SAC) algorithm is designed. Domain-knowledge-driven state feature selection, action space constraints, and composite reward function design are introduced to achieve collaborative improvement in training efficiency and optimization quality. Experimental results show that the SAC optimization scheme outperforms comparative methods such as PPO, TD3, and GA in terms of cumulative reward and convergence speed. The product nitrogen content compliance rate is increased to 98.3%, and the total energy consumption of the push plate kiln is reduced by approximately 12.6%. Finally, based on digital twin technology, a monitoring and process parameter optimization system for the vanadium–nitrogen alloy production process is established, and functional modules such as production line operation status monitoring, production data monitoring, multi-flow-field fusion evaluation of kilns, and intelligent optimization of process parameters are developed, achieving visibility, measurability, and controllability of the vanadium–nitrogen alloy production process.
Förderung
- Department of Science and Technology of Shandong Province
- Shandong Key Laboratory of Intelligent Manufacturing Technology for Advanced Power 789 Equipment
Zitieren
APA
Zhang, X., Shi, C., Zhang, Z., Gao, Z., & Sun, Y. (2026). Monitoring and Process Parameter Optimization of Vanadium–Nitrogen Alloy Production Process by Integrating Deep Reinforcement Learning and Domain Knowledge. Processes, 14(22), 3644. https://doi.org/10.3390/pr14223644
Vancouver
Zhang X, Shi C, Zhang Z, Gao Z, Sun Y. Monitoring and Process Parameter Optimization of Vanadium–Nitrogen Alloy Production Process by Integrating Deep Reinforcement Learning and Domain Knowledge. Processes. 2026;14(22):3644. doi:10.3390/pr14223644
BibTeX
@article{Zhang2026,
author = {Zhang, Xiangyu and Shi, Chunxiao and Zhang, Zhiming and Gao, Zenggui and Sun, Yanning},
title = {Monitoring and Process Parameter Optimization of Vanadium–Nitrogen Alloy Production Process by Integrating Deep Reinforcement Learning and Domain Knowledge},
journal = {Processes},
year = {2026},
volume = {14},
number = {22},
pages = {3644},
doi = {10.3390/pr14223644}
}
EUVIMED