Optimization of a Deck Structure due to Modularization of Cabins

Abstract

This thesis presents an optimization approach for design of a deck structure that enables modularization of cabins in superstructures of offshore support vessels. The methodology has been based on a superstructure designed by Vard, which has been simplified to a rectangular deck structure with no out-cuts. The deck structure has been organized as a grid system, where all crossing points between girders are pillar locations. Prediction of the pillar locations makes it possible to modularize the cabins. The simplified structure had 13 variables, where the 9 directly related to the grid have been selected as decision variables. Empirical weight and cost functions have been used to derive the objective functions. Class rules from DNV have been used to define a feasible region of the model. The methodology approach is based on genetic algorithm which is a heuristic search algorithm. The algorithm has crossovers and mutations as main action. Given the right input parameters, it will search widely within the feasible region to avoid getting caught in a local optimum. Evaluation of input parameters of the algorithm has been done to increase the searching capabilities. The major drawback of the algorithm is that there is no guarantee that the discovered solution is the global optimum. To increase the quality of the model, user-feedback if over-dimensioning is implemented. The model has been created by using a combination of self-made scripts and the global optimization toolbox in MATLAB.The program gives low spacing and small dimensions for weight minimization and the opposite for cost minimization this is as expected from practical experience. The program has solved the optimization task well, both considering expended run time and final results. The parameters of the algorithm should be reevaluated if changing the number of variables.