A model predictive control approach for real-time optimization of reentrant manufacturing lines

Abstract

A two layer hierarchical framework for optimization, control, and scheduling of semi-conductor reentrant lines is proposed. In this framework, model predictive control (MPC) is used at the top layer for real-time optimization (RTO). This layer acts as an interface between long-term planning (months) and scheduling (minutes). An ℓ 1-norm MPC, which uses a discrete linear model, addresses the long-term (shifts) inventory control problem while minimizing cycle time. It can also address the inventory and production control problems. The receding horizon feature of MPC allows the algorithm to simultaneously act as a long-term optimizer and as a controller. This algorithm is implemented as a linear programming (LP) problem, which is solved at the beginning of each shift. At the lower level, a variable priority policy (VPP) tracks the commands generated by the optimizer/controller providing the detailed operation of the discrete event fabrication line. The approach is illustrated with a case study of a five-machine, six-step line example developed by Intel.