Applying a Genetic Algorithm for Sequencing Optimization in a Mixed-Model Manufacturing Assembly Line

The article was presented in 37th International Manufacturing Conference and published originally with respective proceedings. Here, related program code is also included. Abstract In this study, we provide a genetic algorithm for the sequencing of made-to-order products in a mixed-model assembly line. Such assembly lines, in which similar products are assembled with customizations, are widely used in the manufacturing industry. An assembly line typically consists of several assembly stations that can be manual or automated. In each station, differences between consecutive products may require changing of tools or part pallets located next to the station, with an associated idle time cost away from actual assembly tasks. As a heuristic to improve production efficiency, we optimize the sequencing of incoming orders to minimize the total count of differences in the bills of materials (BOMs) between consecutively assembled products. The resulting sequencing problem is equivalent to a specific type of an open-loop traveling salesperson problem. The matrix of distances between cities corresponds to the matrix of BOM differences, and a fixed starting city corresponds to the last product that was assembled in the previous batch of orders. To test the optimizer, we generate incoming orders with a wide range of parameterizations of a BOM difference distribution. We find that optimizing the sequencing of a large batch of incoming orders gives a significant reduction in differences in BOMs of consecutively assembled products, and gives greater benefits than optimizing smaller batches of orders in succession. The genetic algorithm has a parallelizable design and, for a batch size of a few hundred orders, executes in seconds using a personal computer with a multi-core processor. The distance matrix approach allows also to include weights specific to different types of BOM items. The plain genetic algorithm can be directly adapted to cost functions that cannot be parameterized by a distance matrix alone, for example when constraints for order due dates are included