Sparse data interpolation for selflearning cavitation control

Abstract

This paper describes methods for constructing and changing characteristic surfaces from sparse data. Particular emphasis is put on methods capable of locally modifying the surface whenever a new data point becomes available. A local radial-basis-function network (RBFN) is described and analysed in some depth and contrasted to two alternative methods which use iterative increment functions and a minimum-norm-network approach, respectively. The local RBFN requires the least computational effort while still providing a sufficiently high degree of accuracy for the current application. It can be implemented very memory efficiently on a programmable logic controller (PLC).<>