Cavitation intensity prediction and optimization for a Venturi cavitation reactor using deep learning

The Venturi reactor, widely used in process intensification through hydrodynamic cavitation technology, has proven highly effective in various chemical and environmental applications. The cavitation intensity of a Venturi is primarily influenced by shape parameters such as the convergent angle (β1), throat diameter (dth), throat length (lth), and divergent angle (β2). However, the impact of these parameters on cavitation intensity has not been sufficiently clarified. In this study, the structural optimization of a Venturi reactor was accomplished by integrating deep neural networks with particle swarm optimization. The Cavitation Intensity Prediction Network model, which combines artificial neural networks and numerical simulation, was used to establish the nonlinear relationship between shape parameters and cavitation intensity. Partial dependence plots and individual conditional expectation plots were utilized to clarify the influence of each parameter. The findings reveal that the cavitation intensity of the optimal Venturi is 2.76 times greater than that of the original design. Reducing β1 resulted in a swift conversion of static pressure into dynamic pressure, but it also caused an uneven distribution of fluid velocity. To reduce this unevenness and allow the dynamic pressure in the throat to reach its peak, which is advantageous for cavitation generation, lth should be extended. dth directly influenced the efficiency of converting static pressure into dynamic pressure and was a key factor in determining cavitation intensity. β2 indirectly impacted cavitation intensity by modulating the space available for cavitation development. The insights gained from this study may provide valuable guidance for designing Venturis in process intensification applications.