Anticipation in supply chain operations planning

Abstract

Complex production and distribution networks are governed by implicit or explicit hierarchical decision making processes. This thesis contributes to the understanding of such hierarchical decision making processes. Our starting point is the Eindhoven Framework for Production and Inventory Control. This generic hierarchical framework distinguishes between goods flow control and production unit control. We incorporate the concept of anticipation into this hierarchical framework. This implies that at the goods flow control level, the behavior of the production unit under the specific production unit control mechanism is taken into account. Based on explicit quantitative models for material and resource coordination at the goods flow control level, and for production scheduling at production unit level, we study the behavior of general production and distribution networks. Quintessential to our approach is that the quantitative models used for controlling at the goods flow control level are abstractions of the actual interactions between production units with respect to ordering and delivering materials, and the actual behavior of the resources within the production units. This mimics modeling and control of real-life processes and is fundamentally different from approaches that assume that the models are reality itself. Specifically, the abstraction introduces uncertainty in the output of the production unit over time. Our research methodology is a combination of formal (mathematical) modeling and analysis, and discrete event simulation. The model at the goods flow control level is a mathematical programming model supplemented with heuristics that account for the non-linear relationship between input and output in the production unit. Statistical analysis is applied to the simulation results to compare different anticipation concepts. Here we exploit the Safety Stock Adjustment Procedure which allows us to compare control concepts that yield identical service levels on the basis of costs associated with work-in-progress and inventory alone. We compare newly developed models to those developed by other researchers. The central trade-off in this thesis is the trade-off between effective use of resources and lead time reliability. A (bottleneck) resource in a production unit is used effectively when it is actively processing items that need to be produced. In this thesis, effectiveness translates to avoiding idle time on the resource due to waiting for the work to arrive or to be released. Such idle times can be avoided by ensuring a high enough workload in the production unit. Workload on the other hand, is directly related to the flow times of production orders in the production unit. Because proper goods flow coordination is dependent on reliable planned lead times, the workload in the production unit must be kept within bounds. In other words, workload must be high enough to meet the required output level but may not be so high that it can no longer be cleared from the production unit within the planned lead time at the corresponding level of output. The importance of the trade-off between effective resource use and lead time reliability is overlooked if the planning model is not considered outside the deterministic setting of the model itself. In Chapter 2 we use a queueing model to analyze the tradeoff in the simplest possible periodic release planning model for a single production unit with a single resource. We show how the maximum utilization level is affected by the workload constraint, how build-ahead stocks increase with tighter workload constraints and how lead time reliability is reduced by less tight workload constraints. The results demonstrate that it is often necessary to have an (explicitly) planned lead time of more than one planning period in order to combine a high resource utilization with reliable planned lead times. The models that we propose in Chapters 5 and 6 distinguish themselves from existing ones in this respect. They allow for explicitly planned lead times of multiple planning periods and corresponding higher workloads with high levels of lead time reliability.