Multi-objective optimization of jet impact-negative pressure reactor based on computational fluid dynamics and machine learning

• Published in: Chemical engineering research & design Vol. 220; pp. 335 - 349
• Main Authors: Yang, Guangzhou, Yan, Hongying, Dong, Yingying, Hu, Lingxing, Hu, Dong, Qiu, Facheng
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.08.2025
• Subjects: BP neural network; Cavitation; Jet impact-negative pressure reactor; Multi-objective salp swarm algorithm; Jet impact-negative pressure reactor; BP neural network; Multi-objective salp swarm algorithm; Cavitation
• ISSN: 0263-8762
• DOI: 10.1016/j.cherd.2025.07.010

The jet impact-negative pressure reactor (JI-NPR) is a continuous and simple-structured high-efficiency deamination device. This study investigates a novel method for optimizing the JI-NPR through the integration of Computational Fluid Dynamics (CFD) and machine learning (ML) techniques. This method optimizes the jet orifices and operating parameters within the JI-NPR to maximize the cavitation volume fraction and the distribution of cavitation length. The objective is to achieve global optimization with reduced computational cost. Back propagation neural network (BPNN) model is developed using CFD simulation data, where jet orifice diameter, reactor flow rate and reactor pressure served as inputs, the cavitation volume fraction and the distribution of cavitation length within the jet orifice are the outputs. The BP neural network innovatively constructs a real-time surrogate model for jet parameters–cavitation response, which can replace the time-consuming CFD simulations. BP-MSSA (Multi-objective salp swarm algorithm) is used for extreme value optimization to obtain the optimal reactor process parameter combination and verify the results. The optimal parameter combination obtained is flow rate Q, 2.492 m3/h; jet orifice diameter d, 2.001 mm and pressure pout, −19.992 kPa. The deamination experiment shows that the ammonia nitrogen removal rate of the JI-NPR after optimization increased by 26.14 % compared to before optimization. [Display omitted] •The mapping relationship between the operating parameters, structural parameters and hydraulic cavitation of the jet impact negative pressure reactor was established.•The constructed BP neural network model was feasible for predicting cavitation behavior.•BP neural network was used as a surrogate model for MSSA global optimization.•The deamination experiment efficiency improved by 26.14 %.