Performance Optimization Design of Stirred Tanks Based on Deep Neural Networks

Stirring equipment is widely used in core fields such as energy and the chemical industry. Stirring operation is one of the important links in the process of industrial production. Its efficient mixing and green-energy-saving design have always been the focus of researchers. In this study, an optimization framework that integrates class/shape function transformation (CST) parametric design with a deep neural network was developed. The CST method was used to characterize the low-dimensional parametrization of the blade camber line, and the small sample blade space was generated by Latin hypercube sampling. A deep neural network surrogate model based on a multilayer perceptron (MLP) was established for the rapid prediction of impeller torque, particle dispersion uniformity, and other parameters, and the optimization design was carried out with a multiobjective genetic algorithm. The results show that the optimized blade can effectively reduce the energy dissipation in the wake region and strengthen the local turbulence near the tip region, thereby enhancing the off-bottom starting and axial transport capacities of the particles. Compared with the original blade, the solid cloud height in the stirred tank increases by 16.05%, reaching the criterion for complete suspension defined in previous studies, while the impeller power consumption decreases by 10.34%.This demonstrates the feasibility of this method for the energy-efficient design of stirring impellers and provides methodological support for the intelligent optimization and energy-efficiency enhancement of similar complex multiphase flow equipment.