Scheduling of a multi-product polymer batch plant

In this paper, a real-world scheduling problem from the polymer industries is investigated. Special nonstandard features of the problem are the high degree of coupled production where none of the different products can be produced separately but their relative proportion can be influenced by the choice of the recipes, and that the discontinuous and the continuous part of the plant are connected by a mixing stage which gives rise to nonlinear relationships between the batches. Two different mathematical models are presented: a continuous-time and a fixed-grid model. Both models give rise to large, nonconvex, mixed integer nonlinear problems (MINLP). The size of the problems makes it impossible to use general purpose algorithms. We present scheduling algorithms which take the specific properties of the problem into account and lead to good suboptimal solutions. The two problem formulations are compared wrt. the computational effort required to compute the schedules.