Advanced MILP-based strategies for logistics decisions in industrial gases supply chains

BONINO, SERGIO G.; ZEBALLOS, LUIS J.; MOOLYA, AKASH; LAÍNEZ AGUIRRE, JOSÉ; PINTO, JOSÉ; GROSSMANN, IGNACIO; MÉNDEZ, CARLOS A.

Buenos Aires

Conferencia; XXI Latin-Iberoamerican Conference on Operations Research; 2022

This paper introduces an efficient and realistic MILP-based heuristic procedure to deal with distribution scheduling of large size problems of industrial gases. The work proposes new strategies for generating efficient routes for product distribution in combination with an aggregated production scheduling. The problem statement takes into consideration a predefined time horizon, a set of production plants, a set of customers with daily product demands, a fleet of trucks to distribute gases, min/max production rates in plants as well as min/max tank levels in plants and customers. An efficient discrete time MILP-model was developed to consider the production routing problem (MPRP) with multiple production plants with different features. Initially, the logistic model was solved considering a problem involving 50 customers, 2 production plants, 6 vehicles and 2 products. The approach was successfully solved when a time horizon of three days and direct shipments to customers were considered. In order to consider industrial case studies, with at least a 30-day time horizon and several customers per route, an iterative procedure based on the efficient discrete time MILP-model was developed. Thus, the iterative procedure considers one day in each iteration and the whole problem is faced by subsequently solving and combining multiple discrete time route generation MILP models. The procedure was tested considering a real case study of an industrial gases supply chain with 5 production plants, approximately 500 monthly deliveries to 120 customers, 14 vehicles and a 30-day time horizon. The results obtained showed that efficient solutions can be generated with a robust solution strategy and stable CPU times. In this way, the problem is possible to be solved by considering multiple visits per route, routes with a duration of two or three days and multiple routes per day for every vehicle.