Application of System Learning to Precedence Graph Generation for Assembly Line Balancing

Manufacturing Process Planning is the systematic development of the detailed methods by which products can be manufactured in a cost-efficient manner, while achieving their functional requirements. An assembly line is a flow-oriented production system where the productive units performing the operations, referred to as stations, are aligned in a serial manner. The work pieces visit stations successively as they are moved along the line usually by some kind of transportation system, e.g., a conveyor belt. An important decision problem, called Assembly Line Balancing Problem (ALBP), arises and has to be solved when (re-) configuring an assembly line. It consists of distributing the work tasks among the work stations along the line due to changes in task requirements for planned production. The assignment of tasks to stations is constrained by task sequence restrictions which can be expressed in a precedence graph. However, most manufacturers usually do not have precedence graphs or if they do, the information on their precedence graphs is inadequate. As a consequence, the elaborate solution procedures for different versions of ALBP developed by more than 50 years of research are often not applicable in practice as not all constraint information is known. This is a common problem in automotive final assembly. In this work we describe a novel precedence generation technique that is based on system-learning from past feasible production sequences. This technique forms a sufficient precedence graph that guarantees feasible line balances. Experiments indicate that the proposed procedure is able to approximate a precedence graph generated by an expert sufficiently well to detect nearly-optimal solutions even for a real-world automotive assembly line segment. Thus, the application of system learning seems to provide a simple and practical way to implement Decision Support Systems to make assembly line planning more efficient.