Machine-learning-assisted joint optimization of impeller and volute for 6 MF-20 centrifugal fire extinguishers with three-dimension computational fluid dynamics and experiment

Centrifugal fire extinguishers are crucial for forest fire prevention. However, traditional design and optimization methods are often hindered by the high dimensionality of design variables. This study introduces a novel three-dimensional modeling and impeller-volute joint optimization strategy for the 6MF-20 centrifugal fire extinguisher, employing Machine Learning (ML) and Computational Fluid Dynamics (CFD) to overcome these limitations. We propose an AI-assisted framework, integrating a Back-Propagation Artificial Neural Network (BPANN) with the Non-dominated Sorting Genetic Algorithm II (NSGA-II), aimed at enhancing flow rate and efficiency. The framework is trained on a high-precision CFD dataset, enabling it to capture complex nonlinear relationships between design parameters and extinguisher performance. Hyper-parameters of the learning model are optimized using Bayesian Optimization algorithms. Comparative experiments compare the optimized impeller against the original counterpart. Numerical outcomes indicate that the optimization of 8 design parameters results in a 19.18% increase in flow rate and an 18.76% enhancement in efficiency within the impeller-volute joint optimization. Experimental performance enhancement of the optimized impellers is confirmed at 10.68%. The standard deviations of the experimental data for the optimized impeller are approximately 0.003 and 0.007, significantly lower than those of the prototype impeller. This research underscores the significant potential of AI in the optimization of mechanical systems, with determination coefficients R 2 for BPANN exceeding 0.95, indicating high model accuracy and reliability.