#TITLE_ALTERNATIVE#

• Main Author: Utama Noor Permadi - NIM: 23415025 , Agri
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/20803
• Institution: Institut Teknologi Bandung

A large majority of facility location problems in industry aim at locating production plants or logistics facilities over a strategic time horizon. Sizing decisions determine the overall capacity installed at each capacity. One strong characteristic is that the installed capacity is generally not very flexible mainly due to space limitation. In this research, the problem of locating logistics platforms and planning a set of tasks is modelled. A characteristics feature is that some tasks require the use of mobile equipment, i.e. some machines that can be moved from one logistics platform to another. The problem includes two strategic decisions (select a subset of candidate platforms to be opened) and tactical decisions (locate the mobile equipment). The problem is modelled as a mixed integer linear program (MILP). We consider a time horizon of a year, decomposed into several time periods, typically month. The company has a large set of customers, with a deterministic but variable demand during a subset of time periods. Logistics platforms are used to store materials and process materials and allow consolidations for the supplies of customers. The objective function to be minimized is the sum of the fixed set-up cost of platforms, production, inventory and delivery costs and the cost of moving the mobile equipment from one platform to another. Instances is generated based on data collected from a public works company. To emphasize our experiments, two layouts are proposed: (i) cluster with significant correlations between facilities and customers and (ii) random locations for both facilities and customers. 60 instances are generated with different values for the number of facilities, customers, and mobile equipments. First, the model is solved with IBM Ilog Cplex 12.6.3 with standard tuning and a time limitation of 5 hour. Smallest instances have been solved to optimality and larger instances tend to point out the difficulty of the combinatorial behavior of this problem. Thus, we propose a Variable Neighborhood Descent (VND) metaheuristic. Neighborhoods are based on the destroy/rebuild principle: first relax the values of binary variables for some given periods and all continuous variables and then rebuild the MILP formulation. During a computational time of 10 minutes, our neighborhoods allow to iteratively increase the number of periods for which variables are relaxed. Results show better found solutions for some of the hardest instances.