Projeto de parâmetros robustos para modelagem e otimização de sistemas kanban: uma abordagem baseada em experimentos simulados

Abstract

Inventory control impacts the flow of the production process, customer service and cost. To improve this management, many companies use the concept of kanban, which has the purpose of programming and controlling production visually. However, few studies in the literature make use of a systematic method for its optimization, and it is also possible to observe in industrial environments, the use of Toyota's traditional formula, considering the demand levelly. Therefore, this work was developed with the objective of proposing an approach to optimize kanban systems using the Robust Parameters Design, through simulated experiments. This strategy is based on the calculation of the safety stock to determine the experimental planning, seeking to achieve the desired service level. Then, the simulation is executed with the experimental planning data in the simulated model, considering the daily random variation of demand over 30 days, with the quantity of kanban not delivered and the number of kanban in stock recorded at the end. From these data mathematical models that take into account the mean and variance are constructed, so that the robust optimization for the dimensioning of the supermarket can be obtained through the Mean Square Error (MSE). To prove the applicability of the proposed method, the procedure was tested in two different cases in the literature, Tubino (2007) and Hurrion (1997). For the modeling of the optimization functions, the Response Surface Methodology (RSM) was used, being mathematically programmed using the MSE. The weighted approach (MSEP) was also applied as an additional technique to achieve better results in relation to robust optimization. Therefore, the proposed approach was developed and applied satisfactorily in both cases in the literature, leading to optimal results in relation to the objectives of robust optimization. It was also possible to make a comparison between the proposed approach and the results achieved with the optimizer of the Arena® simulation software, OptQuest®, thus demonstrating that the method proposed in this work obtained better results with regard to robustness.