Synergy between evolutionary optimization and induction graphs learning for simulated manufacturing systems

The performance of manufacturing systems is strongly dependent on a set of numerical and non-numerical decision variables, such as transport lot sizes and priority rules, for example. Configuring such systems requires determining a value for each decision variable to optimize a performance criterion. Recent works have shown that simulation optimization of manufacturing systems can be efficiently addressed using evolutionary algorithms. Nevertheless, these algorithms do not provide any understanding on the system's behaviour, which could be crucial regarding the investment amounts. Existing methods based on learning strategies try to provide knowledge on the system, but relevant information could be scattered. To optimize the system, characterize good solutions and highlight critical decision variables, an 'intelligent optimization' method is proposed based on the synergy between evolutionary optimization and induction graphs learning. These approaches work together to find good solutions with associated knowledge. The method is illustrated by the configuration of an assembly kanban system constituted of five machines that process three types of product.