PSI - Issue 3

Alberto Lorenzon et al. / Procedia Structural Integrity 3 (2017) 370–379 A. Lorenzon et al. / Structural Integrity Procedia 00 (2017) 000–000

372

3

2. Methods 2.1. Turbulence models for CFD

Many commercial and open source programs are currently available to perform CFD analysis. In this review, only CFD methods currently implemented within these codes (or easily implementable) will be presented, in order to focus on a set of tools that are already available for the assessment of wind effects on large steel structures. The solution to the Navier-Stokes equations provides all information regarding every aspect of a turbulent flow. Considering a homogeneous incompressible fluid of constant viscosity, in general Navier-Stokes equations can be expressed as:

(  v v v ) t 

1

2

v

(1)

p        



0    v

(2)

As shown by many (see Speziale (1991), Hanjalić (2004), Argyropoulos and Markatos (2015)) limitations in computer capacity make it impossible –for now and the foreseeable future – to directly solve these equations in the complex turbulent flows of engineering interest. This is essentially due to the nature of turbulence, which is, by definition, an irregular condition of flow, characterized by strong non-linearity and large amplitude in the scales of length, time and velocity. However, for most engineering purposes, it is not necessary to identify every feature of the flow. In fact, in many relevant civil applications, as shown theoretically by POD and other similar decomposition (see, for example Romanowski (1996) and Dowell and Hall (2001)) that only eddies with the higher energy are responsible for dynamic effects on structures. Use of mathematical models to simulate the physics of turbulence is thus an obvious and reasonable choice. The various approaches to turbulence modeling differ essentially on the portion of turbulence, which is solved analytically against the portion that is instead simulated. The three major classes of models are: 1. DNS (Direct Numerical Simulation): Navier-Stokes equations are numerically simulated at all length and at all scales and there is therefore no need of any turbulence model. This class of models has provided significant contribution in turbulence research as it allows for the most accurate results. The need to model the largest scale of turbulence while also providing a fine enough grid resolution that allow to capture the dissipation length scale (Kolmogorov micro-scale) makes, however, its computational demand too high for any relevant industrial application. 2. RANS (Reynolds Averaged Navier-Stokes): the entire flow is averaged and the turbulence is modeled using various approaches. 3. LES (Large Eddy Simulation): only the major vortices are numerically solved while sub-grid eddies are modeled. RANS and LES models are further described below. Between these classes are also present many other models that aim to realize improvements in the description of the physics while obtaining reduction in computational cost. An ample review of hybrid RANS-LES models has been provided in Fröhlich and von Terzi (2008). Some of the most successful hybrid models are: - VLES (Very Large Eddy Simulation), see Johansen et al. (2004); - DES (Detached Eddy Simulation), see Spalart (2009); - PANS (Partially-Averaged Navier-Stokes), see Girimaji (2006). In recent years, many alternative approaches to “traditional” CFD implementation have been proposed. One of the most attracting for the industry may be represented by the Lattice Boltzmann method, which is a mesoscopic particle-based approach to CFD which eventually promises to remove the complexity introduced by the grid generation (see Holman et al. (2012)).