An Enhanced Grouping Genetic Algorithm for Optimising the Formation of Design Teams and Manufacturing Cells

• Main Author: Tunnukij, Teerawut
• Published: University of Newcastle Upon Tyne 2008
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.489306

Capital goods are typical complex products and systems, which are highly customised, engineering-intensive goods that are likely to be produced in one-off projects or small batches for individual customers. The major business activities of capital goods companies are the design, manufaCture and construction of complex products such as turbine generators, power station boilers and oil rigs. These products have deep and complex product structures with many levels of assembly. Unlike mass-produced goods, which are based upon standardised components, they are often highly customised and produced in low volume to meet the specific requirements of individual customers. The highly customised nature of the products leads to a significant design content per order. The effective management of capital goods companies therefore requires both design and manufacturing to be well managed. In the design process, multifunctional design teams (MDTs) are frequently required to work together simultaneously to create a product that satisfies customers and market requirements and improve the design process. The implementation of MDTs has been reported to improve the performance of new product development in many companies. This includes reduction in product development time, engineering changes, scrap, rework and product time to market, improvement in product quality, and increase in return on assets and service life. However, all of the methods in the literature are likely to be suitable for relatively small or simple problems. In addition, additional methods are .required to complete the formation of MDTs - grouping individuals into MOTs and tasks into sets of tasks. None of the previous methods can group individuals into MOTs and tasks into sets of tasks simultaneously. There is a lack of effective methods for creating optimal MOTs, particularly in complex systems where hundreds of people may involve. In the manufacturing process, a well-designed manufacturing facility enhances manufacturing efficiency by reducing material flow, materials handling, work in progress and lead times. Scheduling and the control of operations may also be improved. Cellular manufacturing (CM) has been recognised by manufacturing industries as a major philosophy to achieve competitiveness through reduction in lead times and manufacturing costs as well as improvement in delivery performance and product quality. The implementation of CM requires parts with similar processing requirements to be grouped into part families. Maimfacturing cells are clusters of dissimilar machines placed in close proximity that are dedicated to the manufacture of part families. A large number of clustering methods have been developed for identifying potential manufacturing cells by solving the cell formation problem (CFP), which groups machines into cells and parts into part families. However, the clustering methods in the literature have been mostly applied to relatively small or simple problems. In addition, traditional clustering methods produce inconclusive results for some large complex manufacturing systems such as capital goods companies. There is a lack of effective clustering methods for solving the CFP in complex systems. The MDT formation problem and the CFP are typical grouping optimisation problems in complex systems. They have been shown to be non-deterministic polynomial (NP) complete problems which mean that the time taken to produce solutions increases exponentially with problem size. Effective clustering methods are therefore required for solving these problems in complex systems, which may involve hundreds of people that need to be organised into MOTs to work on sets of tasks and a large number of parts and machines that need to be grouped into manufacturing cells. However, most of clustering methods in the literature have been applied to relatively small or simple problems. Traditional clustering methods also produce inconclusive results for some large complex manufacturing systems. In addition, there are no generic clustering methods that have been applied to both the MDT formation problem and the CFP, particularly in complex systems. The aim of this research was to develop an improved way of solving both the MDT formation problem and the CFP in complex systems. Since MDT formation problems and CFPs are NP-complete problems, this research developed a clustering method based upon Grouping Genetic Algorithms, which are meta-heuristics (stochastic optimisation algorithms) that can find global or near-global optimal solutions within a reasonable amount of computation time. The developed clustering method is referred to as an Enhanced Grouping Genetic Algorithm (EnGGA). The EnGGA replaces the replacement heuristic in a standard Grouping Genetic Algorithm with a Greedy Heuristic and employs a rank-based roulette-elitist strategy, which is a new mechanism for creating successive generations developed in this research. A new approach was also proposed for clustering exceptional elements (EEs) into sub-cells. In the MDT formation problem, this clustering approach identifies engineering liaisons that facilitate information transfer between MOTs. In the CFP, it provides information on which exceptional machines should be placed near together in a cellular layout in order to reduce the inter-cell part distance travelled by exceptional parts. The EnGGA was used to form MDTs and sets of tasks simultaneously, and a local search heuristic was used to identify engineering liaisons when applied to the MDT formation problem. For solving the CFP, the EnGGA was used to cluster machines and parts simultaneously into independent manufacturing cells, and a local search heuristic was then used to identify exceptional machines that should be placed near together. The developed algorithm was tested with a wide range of CFPs in the literature as well as a complex CFP obtained from a collaborating capital goods company. It was then applied to MDT formation problems. The results showed that the developed algorithm was effective and outperformed all the other methods considered when applied to well-known data sets of CFPs in the literature. It produced the best solutions in all cases. The program required less than one minute computational time in all situations. It also effectively solved the large complex industrial CFP whilst traditional clustering methods produced inconclusive results in this case. It could provide information on which exceptional machines should be placed near together in a cellular layout. It was also effective when applied to MDT formation problems. It could complete the formation of MDTs by grouping individuals into MDTs and tasks 'into sets of tasks simultaneously with the identification of engineering liaisons whilst traditional clustering methods require additional methods to complete the formation of MDTs and they are likely to be suitable for relatively small or simple problems. It is therefore likely to be a promising tool for solving grouping optimisation problems in complex systems.