Comparison of finite element and finite volume methods for simulation of natural ventilation in greenhouses

Abstract

This article aims to contribute to the discussion on the efficiency of two different discretization methods used as computational fluid dynamics (CFD) solvers for the simulation of natural ventilation in greenhouses. The focus is not on a general use of CFD, but rather on its specific application to simulate airflow in naturally ventilated greenhouses. After a short review of the basic model and its extensions, we compare the accuracy and computational efficiency of two simulation codes based on the Finite Element Method (FEM) and the Finite Volume Method (FVM) for two-dimensional incompressible turbulent flow in naturally ventilated greenhouses. FVM software (ANSYS/FLUENT v 6.3.) is the most frequently used CFD code in ventilation research, but few papers using FEM software (ANSYS/FLOTRAN v. 11.0) have been published. CFD simulations have been compared to experimental data for 12 cases corresponding to three greenhouse types. The experimental greenhouses were chosen to represent a large range of ventilation situations: buoyancy effect in a mono-span greenhouse with adiabatic walls, buoyancy and wind effect in a multi-span greenhouse and ventilation in an Almeria-type greenhouse under conditions of large temperature gradient and high wind speeds. The data from simulations and field experiments were compared using different parameters to analyze the effectiveness of experimental data in the validations of CFD models. The possibility of repeating simulations with different discretization methods and commercial software has been tested, as well as the type of experimental data needed to ensure correct validation of CFD models for use in greenhouse ventilation studies. To this end, temperature distribution measurements are preferable to set-point measurements and the use of visualization techniques (laser sheets) or the measurement of velocity vectors (anemometer) are more indicative than ventilation rates. The computational capacity of these approaches has also been analyzed, comparing their performance in terms of the overall database space necessary to store the numerical models and the necessary CPU time to compute one step of the convergence process. On average, the FEM required twice as much computing time per cell and step as FVM, and the amount of required memory storage was approximately 10 times greater for the FEM.