Réduire les coûts de la simulation informatique grâce aux plans d'expériences : un exemple en calcul de procédé Reducing Computer Simulation Costs with Factorial Designs: an Example of Process Calculation

Abstract

Cet article est destiné à montrer que la méthode des Plans d'Expériences utilisée dans les laboratoires et les unités de fabrication est également applicable au calcul scientifique et en particulier, à la simulation informatique. Son emploi permet de réduire, dans une forte proportion, le nombre de passages informatiques. Il permet également d'écrire des modèles mathématiques empiriques qui orientent les recherches vers la bonne solution et qui fournissent une bonne image du phénomène étudié. The aim of this article is to show that Factorial Design, which is a commonly used method in laboratories and production units, can also be very successful for designing and computerized simulations. Computer runs can be reduced by a factor as great as four to achieve a comprehensive understanding of how a plant or a process runs. Simple models can then be constructed to provide a good image of the investigated phenomenom. The example given here is that of a plant processing raw Natural Gas whose outputs are a Sales Gas and an NGL which must meet simultaneously five specifications. The operator in charge of the simulations begins by defining the Experimental Range of Investigation (Table 1). Calculations (Table 1, Fig. 2) are set in a pattern defined by Factorial Design (Table 2). These correspond to the apices of the Experimental cube (Fig. 2). Results of the simulations are then reported on Table 3. These require analysis, using Factorial Design Theory, in conjunction with each specification. A graphical approach is used to define the regions for which each specification is met: Fig. 3 shows the zone authorized for the first specification, the Wobbe Index and Fig. 4 gives the results for the outlet pressure of the Turbo-Expander. Figs. 5, 6 and 7 show the zones allowed for the CO2/C2 ratio, the TVP and the C2/C3 ratio. A satisfactory zone is found, for this last ratio, outside of the investigated range. The results acquired so far enable us to define a whole new area where all specifications should simultaneously be met. This area has been found with the help of empirical mathematical models derived from the theory of Factorial Design. An instruction point is proposed (Fig. 8) and a confirmation run for these conditions has been made to show the validity of the model. With the use of Factorial Design, the computing budget was controlled with a number of simulations reduced by a factor of four compared to the more traditionnal approach. In addition to these savings, the engineer in charge now has a clear view of how the plant runs. He can now deal with various related problems without having to re-run simulations.