Abstract
This paper aims at developing a fully automated hull form design technique employing an Neural Network and Genetic Algorithm methods resulting in accelerated convergence. For generating an input data that will be, by and large, a close relative of the desired hull, a linear relation has been assumed between the half breadth of different sections and principal dimensions (length, breadth, draft or (displacement)1/3) of a particular type of vessel. Compared to starting with a random value of the input, this technique resulted in faster convergence. The weight matrix for each of these parameters is produced from data obtained from the population. The half-breadth table for a new vessel can be obtained by multiplying the weight matrix with corresponding parameter. However, the half-breadth table obtained in such way may not provide the required displacement and speed of the vessel. Therefore, some readjustments of some of the principal dimensions are required. Neural Networks (Wasserman, 1989) has been used to find the required values of such improved design parameters (principal dimensions). The final design process consists of searching for the exact solution by examining several generations generated by the GA (Goldberg, 1989). The convergence criterion is the summed offset error, which is to be within the envelope defined by the tolerances. Since GA doesn’t guarantee fairness of the surface of the hull form, B-spline curve fitting method is used to obtain a fair hull. Thus, the hull form generated through this process is fully automated, accurate and having fair surface. The technique is also found to be an efficient one.
GAPWM: a genetic algorithm method for optimizing a position weight matrix
