This work presents the application of a full factorial design of experiments as a method of analysis of the simulation results of port operations in a port specialized in off-shore. As proposed by Montgomery (1991), this approach allows identifying the most representative input variables of the model and their respective interactions.
The development of simulation models has grown proportionally with the availability of simulation software and with the increase in computer performance. Within this context, it is observed that a representative portion of studies in the simulation area consume most of their time with computational work, that is, with model programming. With that, the analysis of the results is left aside, being limited in some cases to just one run that is responsible for the “response” of the model.
The design of experiments is a method that allows a better understanding of the input variables (factors) of the model. Its application assists in the construction of an adequate analysis structure that allows identifying the main factors, as well as the interaction between them.
MODELING AND THE DESIGN OF FACTORIAL EXPERIMENTS
The model was developed with the ARENA software and approached a specialized off-shore port. The modeling focused on port operations, where an adequate dimensioning of existing resources is fundamental. Strategic, tactical and operational issues were addressed. The complexity of the model can be translated by the large number of input variables. This made the design of full factorial experiments to be chosen for the structure of analysis of the results.
In the design of a complete factorial experiment, 5 factors were considered and 2 levels were established for each one (one high and one low). Thus, the number of experiments was 32 (25). In addition, the output variables that would be part of the analysis were chosen.
In the past, this type of experiment was unattractive, since the processing time of computers was limited for such an approach. However, nowadays, in addition to the evolution of computers, there are simulation packages that have facilitating mechanisms in the execution of this type of experiment. ARENA's scenario manager makes it possible to prepare all the experiments in folders (sub-directories) and run the program, which in a compacted form, manages to be agile and efficient.
RESULTS
The realization of the full factorial design of experiments allowed identifying the variables that most impact the main response variable of the model (time of permanence of the supply vessels), as can be seen in figure 1. In addition to the variables, it was also possible to identify the most representative interactions (side effects). The use of analysis of variance (ANOVA) helped to highlight factors and interactions that were statistically significant.
 |
CONCLUSION
This approach is extremely applicable for simulation models that have a large amount of input variables. However, it is important to point out that for queuing models, which are often characterized by high variability, this type of method must be used with care, since the autocorrelation between the entities in the model can invalidate the responses obtained.
BIBLIOGRAPHY
ARENA User's Guide – Version 0, Systems Modeling Corporation, 2.0.
BANKS, J., CARSON II, JS, BARRY, LN, Discret-Event System Simulation. 2nd ed. New Jersey, Prentice Hall, 1996.
FLEURY, PF, Production Structure and Operational Performance: Identification of Key Variables Through Simulation,
COPEAD Report No. 261, Rio de Janeiro, June, 1992.
LAW, AM, KELTON, WD, Smulation Modeling & Analysis. 2nd ed. Singapore, McGraw-Hill; 1991.
MONTGOMERY, DC, Design and Analysis of Experiments. 3rd ed. Singapore, Wiley & Sons, 1991.
