Automatic hull form generation using neural nets and genetic algorithms

Abstract

The rapid generation of tltired hull s,,,'bee is very important. Especially, early fairing and modcling of ship hull in preliminary design stage has more effective advantages in many sides like pcrformance analysis, production design, and process management etc. Hull form fairing process, however, still has been a very iterative and time-consuming job. For automatic fairing, hull form must be modeled using computer. In the field of surface modeling of hull form, geometric complexity of hull form gives many difficulties in adopting surface modeling technique that can describe the irregular topological characteristics precisely. Generation of faired hull sudace is also very important in ship production procedure. Ship drawings arc prepared in scales of 40: I to 100: 1. During the construction phase, full-scale drawings are prepared at the so-called "loll-floor". In this process, the minute discontil1l!ity in the shape of the vessel that is not apparent in the scaled drawings becomes apparent. This requires a modification of the. drawings that is both tedious and time consuming. Computer based techniques arc being evolved to measure the continuity of thc shape and thus to ensure that the scaled drawings are truly continuous avoiding the necessity of the corrections trom the lolling process. For designing shiP:s hull, four parameters arc supplied by the owner-length, displacement, speed and type of the ship. Neural Networks is a robust technique to tind the breadth and dratl hom the 11 2 given values. A linear relation between the parameters can generate a design from length, breadth and drat[, This design provides a guideline for generating similar and initial populations with small variations to be fed in a GA-based karner. Cross over and mutation operator along with add alternate openilor of GAs can be used to generate 'new populations (hull form data) from the existing populations to evolve into new generations and the fitness of the generations can be • computed including the new ones. After that weak population can be discarded, as they can not survive. Only strong population with good titness can compete and be able to improve their fitness and evolve themselves to generate new populations until an acceptable solution is found. This computer-based design makes the vessels hull torm design fully automated. The proposed system learns about the design from some previously designed ships and generales weight matrices. These weight matrices along with Neural Nets and Genetic Algorithms create a design, which satis~' all the hydrostatic properties within reasonable error (less than 2%). Moreover, the process has been able to design a ship by examining 60 generations with a population size of 50, i.e.. all together 3000 populations arc manipulated to get almost accurate design. The proposed system, for the first time, design a method lor modeling and fairing of hull 101'111 in fully aulomated manner.