AN ARTIFICIAL NEURAL NETWORK MODEL FOR PRELIMINARY SHIP DESIGN

A preliminary ship design framework is developed which consists of a set of interconnected neural networks (multilayer perceptrons) undergoing supervised training. The model proceeds to generate from the owners requirements, the set of principal dimensions, form coefficients with the estimates of steel, outfit and machinery weight, power requirement, capacity etc. The Modified Marquardt Levenberg Algorithm has been implemented for the minimization of the error in prediction which allows a faster training of the network. The neural network model in the preliminary design framework can thus be used for a quick appraisal of ship design. A containership design has been taken to illustrate the design principles and the behaviour of the neural network model in the ship design framework.