Assembly Line Balancing : Addressing the Theory-Practice Gap

The efficiency of an assembly line depends on how the different tasks are distributed among the work stations that make up the assembly line. Assigning the tasks to the stations is called assembly line balancing (ALB) and is an NP-hard optimization problem. While automating line balancing has the potential to make manufacturing more efficient, and the problem has been studied for many decades, the research has not been widely adopted by industry, where planning is still largely done manually. This thesis discusses reasons for this theory-practice gap and suggests a direction that is more focused on real-world applications. This includes shifting the focus to a decision support system (DSS) rather than considering the ALB solver in isolation. Such a shift places the solver in a new context, where it plays a slightly different role and needs to fulfill slightly different requirements. A framework for building practical ALB solvers within a DSS context is proposed and then evaluated by implementing solvers for two different assembly line models. Increasingly refined versions of the ALB solvers are compared to each other, showing that the most specialized version performs substantially better than the simplest, which demonstrates that this framework allows an effective trade-off between development time and performance. One of the implemented solvers is also pitted against a state-of-the-art solver for the Simple Assembly Line Balancing Problem type 1 (SALBP-1) by comparing its results on a standard set of problem instances against the best published results. The comparison shows that the implemented solver is not just competitive but in some ways superior. On the largest instances, it produces improved solutions for 99 % of the instances that had not already been solved optimally, reducing the number of stations by more than 2 % on average. The solvers also demonstrated the potential to effectively trade off running time and solution quality, which is an important feature in an interactive system that is meant to support an iterative refinement process