Neural Networks in Comparing USN and Wageningen B-Series Marine Propellers

Abstract

The USN-series of experimental data on marine propeller performance (Denny et al, 1989) were compared with fitted Wageningen B-Series data. A Kohonen network has been used to attempt finding non-obvious similarities between the two data sets. The USN-Series has been tested under cavitating conditions, while the available B-Series not. A non-linear fit of the USN-Series, including information on cavitation number a, has been developed and compared with a neural network function approximation. Using the Kohonen classification results, the non-linear regression was re-applied with slightly improved results. In overall, the feedforward neural network architecture mapping gave the best fit both in a statistical correlation measure and in the maximum percentage deviation measure.