Two-vane pump and design method of two-vane pump for wastewater using machine learning

A design method of two-vane pump for wastewater using machine learning, the design method comprising:

• a) setting an objective function;
• b) setting design variables of an impeller and a volute for deriving the set objective function and a range of each design variable;
• c) deriving a plurality of experimental points including values of the design variables within a range of a design variable;
• d) generating an input value by calculating a value of the objective function through numerical analysis of each of the derived experimental points;
• e) constructing a surrogate model through machine learning for the input value;
• f) Deriving an optimal design of the two-vane pump for wastewater from the constructed surrogate model;
• and
• g) outputting at least one of a visual representation of the optimal design or a design value associated with the optimal design;
• wherein the design variable includes:
• a first variable which is a y-axis displacement of one end of the impeller located on an inlet side of a wastewater pump;
• a second variable that is a y-axis displacement of another end of the impeller;
• and
• a third variable that is a cross-sectional area according to a volute angle of the volute;
• wherein in the e), the surrogate model is constituted by a radial basis neural network (RBNN) artificial neural network;
• wherein the e) includes:
• e1) deriving a variable value for constructing the surrogate model through machine learning for the input value;
• and
• e2) constructing the surrogate model using the derived variable value;
• and
• wherein the f) includes:
• f1) predicting the objective function value by substituting a randomly selected design variable value in the surrogate model into a genetic algorithm;
• f2) deriving a pareto-optimal front surface based on a correlation between predicted objective function values;
• and
• f3) deriving an optimal design of the two-vane pump for wastewater that is a design variable value for a target design function value in the derived pareto-optimal front surface.