Sensitivity Analysis as a Tool for Estimating Numerical Modeling Results

Abstract

Modeling of real physical processes by numerical methods is highly time-consuming and requires significant computational capacity. In some cases, tens or even hundreds of hours of high-power computing are needed to virtually model a real process that lasts one second. Processes that take many hours, such as drying, pose an even greater challenge. This problem can be solved in two ways: by using faster computers (such as computing clusters) or by significantly simplifying the modeled process (its geometry, physical phenomena, or the impact of individual factors). For this reason, all methods which speed up or minimize the number of simulations required to achieve the research objective should be analyzed. This article focuses on the latter approach, and it proposes a simple method for predicting the responses of a numerical model (values of any output parameter) to changes in input values (values of any input parameter). This method requires a base model, such as a numerical model which is qualitatively and quantitatively consistent with experimental observations, and a sensitivity analysis. This article discusses the mathematical and logical premises for the discussed model, and it proposes two methods for predicting numerical simulation results. Those methods are illustrated with examples which analyze the behavior of the Eulerian Multiphase Model and describe phase interactions based on Gidaspow's approach. The discussed example relies on data from a series of articles published by the authors in Drying Technology. This article was inspired by the observations made during a time-consuming process of modeling a spouted bed grain dryer, which was described in the above publications. The objective of this study was to discuss the advantages and possibilities created by sensitivity analyses of numerical models and to encourage their practical application.

Keywords:

• Eulerian multiphase model
• Gidaspow model
• Multiphase flows
• Results prediction
• Sensitivity analysis