New method for predicting full-scale power performance of pumpjet propulsion system based on statistical learning

  Objectives  Aiming at the replacement of propellers behind surface ships with pumpjet propulsion systems, this paper introduces a novel method for predicting full-scale power performance based on statistical learning.

  Methods  Pump performance maps originating from the neural network learning of existing pumpjet thrust coefficient maps and matched to a ship's drag line from model tests are used to determine the pumpjet's full-scale power performance behind a large surface ship. To validate its precision and availability, traditional complete model tests including the ship model drag test, pump model open water test and ship-pumpjet self-propulsion test are completed to determine the full-scale benchmark power performance under different ship speeds.

  Results  The prediction errors of the pumpjet's rotation speed, thrust and power under different self-propulsion ship speeds from 18 knots to the design point of 30 knots are smaller than 5.4%, with no more than 2% from the design condition. As for the ship-propulsor interaction amplitude, the surface ship-pumpjet subsystem lies between ship-propeller interaction and ship-waterjet pump interaction with a thrust deduction coefficient approaching zero. From this point of view, the pumpjet propulsion system behind a surface ship can be recognized as a transitional stage from the propeller-shaft configuration to the waterjet propulsion system.

  Conclusions  The method proposed herein can predict the full-scale power performance of a pumpjet propulsion system behind a ship while advancing pumpjet propulsion system design and applications for new large-scale surface warships.