Numerical Modelling of the Ulstein-Aquamaster Azipull Azimuth Thruster
Figure 1. Rolls-Royce Azipull pulling azimuth thruster.
|
MARINTEK has developed tools for the prediction of forces and moments of Azipull, Rolls-Royce’s innovative pulling azimuth thruster. Artificial neural network methodology is used for prediction.
The Ulstein-Aquamaster Azipull is an azimuth thruster with a pulling propeller. It combines the advantages of the pulling thruster with the flexibility inherent in a controllable-pitch propeller to match almost any type of drive to particular requirements. The Azipull is designed for a continuous service speed of 24 knots. It is currently offered in a propeller diameter range of 1.9 to 3.3 m, with a power rating of 1500 to 3500 kW.
Internally, the thruster has a purely mechanical drive system based on bevel gears at the top and bottom of the leg. Power is fed to the unit through a horizontal input shaft inside the hull, and the unit incorporates its own steering motors for azimuthing. The first four units were installed on a new concept double-ended catamaran car-passenger ferry (FerryCat) built in Norway by Fjellstrand AS. The service speed is 22 knots. Azipull will also be delivered to several offshore supply vessels and coastal tankers.
MARINTEK has specialized in numerical modelling of advanced hydrodynamic systems using artificial neural networks. MARINTEK is one the pioneers in the use of artificial intelligence methods in marine technology and has developed several software packages for the innovative propulsion systems of prominent propulsion manufacturers Rolls-Royce and Wärtsilä-Lips.
Model tests were performed with Azipull at different propeller pitch settings and different headings (propulsor angles). Propeller thrust TP and torque Q as well as thruster longitudinal X, transverse forces Y and steering moment MZ were measured. Measurement data were collected in a database in dimensionless form.
Figure 2a. Measurements and predictions for KY (for a high pitch ratio) at different headings (lines are predictions).
| 
Figure 2b. Measurements and predictions for KMZ (for a high pitch ratio) at different headings (lines are predictions).
|
The database was then analysed using an artificial neural networks method to develop equations capable of predicting the thruster characteristics (KX, KY, KTP, KQ and KMZ) as functions of input variables: advance co-efficient pitch ratio and heading angle. The input set does not need to be one of the tested configuration sets, although it should be within the boundaries of the test matrix.
Networks have a set of “training” rules whereby the coefficients are adjusted on the basis of data. In other words, artificial neural networks “learn” from examples (model test results) and show a capability to generalise beyond the training data.
After training sessions were completed, an equation with corresponding coefficients and activation functions was developed for each force and moment coefficient. Using these equations the characteristics of the Azipull could be calculated for all headings, advance coefficients and pitch ratios within the valid range. Figures 2a and 2b show examples of comparisons between model test data and predictions. There is good agreement between predictions and test results.
Figure 3. Main components of a feed forward recall network with a single hidden layer.
Contact at MARINTEK: Kourosh Koshan
(Article in MARINTEK Review No 2-2004)
