The units are members of different layers, which differ in their position within the network. There are three types of layers, where each of them can include several processing units.
Input layer Output layer Hidden layer
The information enters the network through the input layer and the output layer sends the processed data back to the external environment. Units of the hidden layer are only connected to units within the system and therefore are not ‘visible’ from the outside. A network can have one or more hidden layers. [6]
The outstanding advantage of ANNs compared to traditional computer techniques is their ability to learn solving selected problems. In analogy to that of humans the learning process of an ANN is based on experience, which means the network has to be trained until the desired accuracy is achieved. During the learning process the network stores data that can be reproduced during a recall process. [6]
This improvement is achieved by adjusting the weight of each incoming information. It can be said that the knowledge of the network achieved during its training is stored in the weight vector W. [6]
The widely used algorithm for training is called backpropagation. The backpropagation algorithm calculates the error between the output values derived by the network and the desired ones. The internal weights are adjusted in an iterative process with the effect that the error is reduced in the process of training. If the error reaches values beneath a specific limit the network is said to be trained.
To use this training method some information about the specific problem is required in advance to create the training data set, for example empirical data or data bases.
The application of Artificial Neural Networks has been of ever increasing importance during the last years. Their ability of dealing with large amounts of input data and their good performance in calculating non- linear equations make them an interesting tool in many different fields. Beside their most popular use in pattern recognition, such as hand writing or face recognition , they are also applied successfully to a variety of other problems. Their ability to map trends
complex to be noticed by either humans or other computer techniques makes them very
that are too useful
for
processing for example empirical data. They are best used for interpolation but in some cases also have been proven to compute valid extrapolations.[4, 5]
ANNs have also been used in a variety of different fields of the marine engineering sector.
B-34
Many applications are driven by the need for accurate design parameters in the preliminary design stage. Therefore ANNs have been applied to experimentally derived data such as propeller and bulbous bow design charts to map the non-linear relations between the influencing design parameters. [4, 7] Similarly ANNs have been used to derive functional relations between design parameters of cargo ships and tugs using existing designs and common design formulae. [5]
3. DESIRED ACHIEVEMENT
The idea behind the application of an ANN to the preliminary sailing yacht design described in this paper is to create a tool which derives accurate values of design parameters in an early design stage and gives the designer a better understanding of the
functional
relations between these parameters, all with the purpose of saving time needed for the design process.
The application of an ANN to this problem seems to be the optimal approach because this task requires the ability to deal with non-linear relations between multi- dimensional input and output spaces wherein ANNs have proven to be very powerful.
The parameters that have to be taken into account can be divided roughly into two groups:
Design Parameters Performance Targets
The design parameters are those parameters that describe the size and shape of the yacht, such as
Length Displacement Beam Draft Mast height Prismatic Coefficient Centre of Buoyancy ...
whereas the performance targets are parameters that characterise functionality, safety and sailing performance of the yacht, such as
Displacement ? Speed Motions Stability Sail Area Aspect Ratio of Sails Range …
An ANN is able to derive accurate values for the design parameters satisfying given performance targets and therefore eliminates the time consuming search for
©2007: Royal Institution of Naval Architects
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54