Incorporating problem specific knowledge into a local search framework for the irregular shape packing problem

• Main Author: Bennell, J. A.
• Published: Swansea University 1998
• Subjects: 519
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.636077

This thesis investigates the use of problem specific information to enhance a local search approach to the irregular shape packing problem. It shows that such information can significantly enhance performance, but that the choice of information and the way it is included are important factors. The problem is that of arranging a given set of simple polygons in a rectangular stock sheet of fixed width so as to minimise length. A survey of the literature reveals room for further research into good generic approaches to the problem. Local search heuristics and Linear programming based local optimisers are identified as two promising areas. In particular it is conjectured that the two have features that would complement each other in a hybrid approach. The survey also highlights the benefit of using a concept known as the nofit polygon to handle the geometry. Further investigation suggested a lack of good computational procedures for finding nofit polygons. The research therefore starts by developing a new method of calculation. This is then used as the basis of a tabu thresholding approach to the irregular packing problem. Problem specific information is then incorporated into the algorithm in a variety of ways, some of which are designed to improve the search for local optima, while others, including those based on the LP optimisers, are designed to improve the solutions found at these local optima. As well as a theoretical discussion as to the underlying reasons for using each piece of information, extensive computational experiments are carried out on data from a variety of sources. These show that while the right information in the right place can be extremely useful, too much information can actually hinder the search. The most successful variant of our algorithm is shown to outperform other local search based generic approaches from the literature.