Page 381 - Practical Design Ships and Floating Structures

356

Additionally, the requirement is added that the ship in question should have a service speed of
minimum 18 knots. This does not change the distributions output by the Bayesian network for L and D.
By training a neural network with capacity, draught and velocity and inserting TEUs as above,
D = 8.5 m and V= 18 knots, the length and the draught shown as ‘neural network, triple input’ in
Figures 7(c) and 7(d) are obtained.

Although not shown by graphs, these steps cause the distributions of L, A, B and H from the Bayesian
network to be distributed with lower mean and variance. Thus, the range of the estimates of the main
dimensions becomes narrower given the new information.

The above illustrates how evidence can be inserted in Bayesian networks and how a dedicated neural
network can take multiple design requirements into account. This gives the designer the possibility of
finding the most appropriate main characteristics given certain restrictions and demands, the Bayesian
networks even quantify the uncertainties of the estimates.

5 CONCLUSIONS

New empirical formulae for the relation between main characteristics are derived and exemplified by
predictions for container vessels. It is demonstrated that power functions will adequately describe the
relation between TEU capacity and dimensions like displacement, length, draught, breadth, etc. The
versatility of neural and Bayesian networks to take account of multiple design requirements is shown.
Neural networks are simple to implement and yield smaller estimate errors, but as opposed to Bayesian
networks they must be trained for each combination of inputs separately, and in the current
configurations they do not yield information about the uncertainty of the given estimate.

References

Bertram, V. and Wobig, M.(1999), Simple Empirical Formulae to Estimate Main Form Parameter,
Schiff uprd Hafen, 11: 1 18-121.

Cheng, J., Bell, D. and Liu, W. (2000), Learning Bayesian Networks from Data: An Efficient
Approach Based on Information Theory, Technical report, University of Alberta.

Jensen, F.V. (1996), An Introduction to Bayesian Networks, Springer-Verlag, New York.
Lloyd’s Maritime Information Services (2000), Ship Characteristics.

Watson, D.G.M. and Gilfillan, A.W. (1977), Some Ship Design Methods, Transactions ofthe Royal
Institution ofNaval Architects, 119:279-303.