Conference Agenda

In this research project, we consider the hierarchical production planning problem of a configure-to-order company that manufactures modular capital goods. Each customer order is represented and processed as one project. The production planning problem is formulated as a two-stage problem with one tactical problem and multiple operational problems. The solution of the tactical problem defines the constraints for the operational problems. The operational problems are formulated individually for specific production areas, such as workshop assemblies or mixed-model assembly lines. The tactical problem determines the start and finish times of aggregated activities, which define the feasible production planning periods for the operational problems. Further, the tactical problem decides in which production area, and thus, in which operational problem, the activities are processed. We formalize the tactical problem using a multi-mode resource-constrained project scheduling problem (MMRCPSP) with deterministic and constant resource requirements and deterministic duration across the activity's modes. The tactical problem uses a time horizon of several months and aggregated information to solve the objective of minimizing total weighted lateness concerning the due dates of customer orders. The operational problems deal with operational planning and thus focus on shorter planning horizons, more granular planning periods, and more detailed information. In each operational problem, activities are scheduled, and workers are assigned to workstations such that the time windows from the tactical problem are respected and the total overtime is minimized. We will present MIPs for both levels of the hierarchical approach. Furthermore, we will report on a computational study using real-world instances.

This paper addresses integrated master scheduling and sequencing of mixed-model assembly in an engineer-to-order setting. Multiple orders containing models have to be assembled on multiple dedicated assembly lines. For each line, we must determine a model sequence, which must be maintained for all workstations. Depending on their skills, workers can be assigned to different workstations across lines. Workers can work undertime and overtime. For each type of workers, there is an aggregated working time account. The planning problem is to sequence the models and assign workers to the workstations. In a lexicographical approach, the primary goal is to start models as early as possible, while the secondary goal is to balance working time accounts. We model the problem as a mixed-integer optimization problem and undertake computational experiments on an instance set derived from a producer of packaging machines. Based on the computational study, we provide insight on the computation time and the value of working time accounts.

We consider a dynamic lot-sizing model with rework and different failure classes where a known demand must be met in each period. When producing a lot size, a given percentage of the products does not meet the requested quality. A subsequent inspection process assigns the product to exactly one of several failure classes. The product can then be restored to as-new condition during a perfect reworking process where reworking times and costs depend on the failure class. For production and rework, there is a joint resource and production and rework of the same product cannot take place in the same period. The objective is to minimize the total costs. As the problem is NP-hard, we propose a Fix&Optimize (F&O) heuristic. In an iterative way, the F&O heuristic fixes a subset of the binary variables and optimizes the remaining variables. We consider two decomposition approaches, by period and by product. In a computational study we assess the performance of the heuristic and the MIP when solved with CPLEX for a set of generated test instances. We show that the F&O heuristic solves complex test instances within a reasonable time with very good solution quality. Furthermore, we provide managerial insight.