Support Vector Machine Classification Applied to the Parametric Design of Centrifugal Pumps

In this article the parametric design of centrifugal pumps is addressed. To deal with this
problem, an approach based on coupling expensive Computational Fluid Dynamics (CFD)
computations with Artificial Neural Networks (ANN) as a regression meta-model had been
proposed in Checcucci et al. (2015), "A Novel Approach to Parametric Design of Centrifugal Pumps for a Wide Range of Specific Speeds.",  ISAIF 12 paper n.121. Here, the previously proposed approach is improved by including also the use of Support Vector Machines (SVM) as a classification tool. The classification process is aimed at identifying parameters combinations corresponding to manufacturable machines among the much larger number of unfeasible ones. A binary classification
problem on an unbalanced dataset has to be faced. Numerical tests show that the addition
of this classification tool helps to considerably reduce the number of CFD computations
required for the design, providing large savings in computational time.