Optimization Of Centrifugal Impeller Blades Based On Neural Network

Centrifugal compressor impeller is the main component of doing work, the shapeof the blade has an important influence on the flow properties. Therefore, the bladeshape optimization to improving the performance of the impeller flow and reducingenergy consumption is important.In this paper, optimization of forward problem based on neural network approachwas used. The impeller blades were used as the research object; isentropic efficiencyand the total pressure ratio of the impeller were used as target parameters. By changingthe shape of the blades, the optimization design method of the blade was studied.Firstly, the problem of blade parametric must be solved. Comparative study ofBezier and B-splines fitting and investigated the merits of the two methods, at lastBezier curves was used to fit the shape of the blade. So, blade shapes can be controlledby using the coordinates of the Bezier curve control point, to solve the problem of bladeparametric.Second, Comparative study of the three-dimensional CFD and two-dimensionalstreamline curvature method, at last the two-dimensional streamline curvature methodwas adopted as the flow field solver during the optimization in order to save time in thispaper. Flow solver is used to generate the initial sample point and test performance ofthe optimization blades.The computation time is too long and the relationship between the blade shapeparameter and the optimization goals is difficult to express analytically. The BP neuralnetwork which replaces the flow field solver is used as an approximate model predictsreal performance. Sample points generated by neural network are used to train thenetwork until its prediction accuracy increased to the goal. And then multi-objectivegenetic algorithms which embedded the BP network are used obtain Pareto optimalsolutions during global optimization within a given range of parameters. The BPnetwork is used to evaluate the fitness value. Bezier control points and the predictivevalues are input parameters and output parameters in the optimization system. Iterate isgoing on until the goal obtained.Finally, an example is used to verify the effectiveness of the method. The resultsshow the efficiency and pressure ratio of the optimized blade were increased0.126%and0.577%compared to the original blade, so that the method is reliable and effective during optimization of the blade design.