Modeling tools are widely used to create a performance map and decrease design cycle time in computer-aided centrifugal compressor impeller optimization procedures. However, a high-dimensional performance map is difficult to create and application of the approximate performance map brings errors into optimization procedures. This paper presents an online flow solver optimization procedure, in which a Quasi-three dimensional flow solver is directly used to evaluate impeller performances in the genetic algorithm (GA). Also, this procedure is compared with offline flow solver optimization procedure. In offline flow optimization procedure, the flow solver is employed to calculate performances in training database for creating a performance map trained by one type of artificial neural network (ANN), radial basis function network (RBFN). This performance map is further used to calculate the performances of impeller geometries. Results of these two optimization procedures under same GA parameters setting are compared and show that online flow solver optimization procedure can find better optima than offline flow solver optimization procedure. Moreover, influences of GA operators, parameters and local search algorithm on online flow solver optimization procedure are also investigated.

This content is only available via PDF.
You do not currently have access to this content.