Métodos de solução para o problema de dimensionamento e sequenciamento de lotes com limpezas temporais.

Abstract

This paper presents the lot sizing and scheduling problem found in the production of fruit drinks. This is a two-stage production with some particularities: presence of a lung tank in the second stage and the need for mandatory cleanings at each predetermined time of continuous production of the same item. According to Associação Brasileira das Indústrias de Refrigerantes e de Bebidas Não Alcoólicas, the consumption of ready-to-drink fruit drinks is at its peak when compared to the consumption of soft drinks. From 2010 to 2017, the production of fruit drinks shows an increase of 48%, while the production of soft drinks market shows a decrease of 25%. In addition, on average, companies spend around 20% of the day cleaning line equipment. As far as the literature review is concerned, this problem with these characteristics is relatively new, difficult to solve for large real instances and has wide applicability. Based on a mathematical model found in the literature for the problem, analytical methods such as decomposition heuristic and hybrid meta-heuristic are proposed, in order to support decision making in a more efficient way in the planning and lot sizing seeking to minimize the costs associated with production. Computational tests were performed with instances found in the literature, based on real data, and with generated instances. These new instances were generated with the objective of having scenarios more adjusted to reality. Taking into consideration that a more in-depth investigation of the instances found in the literature will show, for the most part, excessive capacities and unbalanced costs. Decomposition and improvement heuristics (Fix-and-optimize) and Simulated Annealing heuristics were tested. The results found were compared with the best method found in the literature for the studied problem. Decomposition and improvement heuristics find better performances in instances with medium or low dimensions, however, for the larger dimensions the heuristic of the literature presents better computational performance.