In this article, the design of a quality control network is considered in an entire supply chain, in the context of an international food and beverage company, denoted as ABC. For each possible control point of the network, two decisions have to be made: (1) establish a monitoring activity or not; (2) if a monitoring activity is placed, determine its acceptance level with a well-adjusted sampling plan. The objective function to minimize consists of a weighted sum of two conflicting components, namely risk (of having bad units) and processing time (to control the product batch itself). Indeed, for each control point, an accurate monitoring activity reduces risk but augments processing time. Each monitoring activity has a specific cost (if selected) and a budget constraint has to be satisfied for the entire network. Various solutions methods are proposed to tackle this NP-hard problem, ranging from a basic greedy heuristic to advanced metaheuristics involving tabu search as an internal procedure. CPLEX can be used for very small instances (with less than 20 control points), for which a basic tabu search can obtain similar results much quicker (in a fraction of second). The experiments for larger instances highlight the importance of having a good control (in terms of the involved search trajectory in the solution space) on the search process, as the methods with well-tuned diversification mechanisms perform best.

在本文中，在国际食品和饮料公司（称为ABC）的情况下，在整个供应链中考虑了质量控制网络的设计。对于网络的每个可能的控制点，必须做出两个决定：（1）是否建立监视活动；（2）如果进行了监视活动，请通过调整后的抽样计划确定其接受程度。最小化的目标函数由两个相互矛盾的成分的加权总和组成，即风险（具有不良单元）和处理时间（以控制产品批次本身）。实际上，对于每个控制点，准确的监视活动可以降低风险，但会增加处理时间。每个监视活动都有特定的成本（如果选择），并且整个网络都必须满足预算约束。提出了各种解决方法来解决这个NP难题，从基本的贪婪启发式方法到涉及禁忌搜索作为内部程序的高级元启发式方法。CPLEX可以用于非常小的实例（控制点少于20个），对于这些实例，基本禁忌搜索可以更快地（不到一秒钟的时间）获得相似的结果。对于大型实例的实验凸显了对搜索过程进行良好控制的重要性（就解决方案空间中涉及的搜索轨迹而言），因为具有良好调整的多元化机制的方法效果最佳。