Issue link: https://www.e-digitaleditions.com/i/1475629
16 August 2022 Inhalation size to zero so that the curve may be accurately fol- lowed. In addition: • Some understanding of the expected flow is important. e mesh should be refined so that the cell size is small where the local gradient in velocity or another variable is large. High refinement facil- itates good resolution of changes in the parame- ter. is can provide good results in this vicinity and elsewhere because the fluid flows around the domain. However, the more cells used, the greater the computational time. • A mesh dependence study should be undertaken. Scale the cell size up and down to see the effect on the simulated results, and to find the size below which any decrease in cell size has no effect on results in the relevant part of the simulation. is supports the identification of the fastest possible accurate analysis. • e mesh may take several iterations to perfect; large gradients may appear in the CFD solution in unexpected places, requiring an updated mesh and a re-run of the solver. • Automatic mesh refinement (AMR) is the term used to describe changing of the mesh during the simulation. Cells are subdivided automatically when and where the solver detects a high gra- dient of any variable, to provide more accurate results, then returned to a larger cell size when the gradient reduces. is aims to create a com- promise between high simulation accuracy and short run time. Solver physics and sub-models Once a suitable mesh is created, the solver must be chosen and physics models set. ese can include models for turbulence; models for particle or drop- let behavior including temperature change and evaporation, collision/agglomeration/break-up, and surface interaction; and models for heat transfer at surfaces. Default choices in software may not be rel- evant for the situation studied. Due to these var- ied and complex physical sub-models, at different scales, CFD is referred to as a multi-physics simu- lation. Some choices that require particular care are discussed below: • Physical properties (such as density, viscosity and thermal conductivity) must be chosen and set. For multi-component mixtures in formulations, the properties set by the built-in packages found in CFD models may not be accurate enough to represent real, non-ideal behavior. If necessary, user-defined functions that implement tempera- ture-, pressure- and composition-dependent phys- ically-based models [22] can usually be included to define fluid properties. • Turbulent flow and appropriate turbulence models for CFD represent a complex field of study in any application area. For inhaled aerosol applications, the ambient gas flow could be laminar—but rapid expansion of formulation from an inhaler could induce turbulent flow. Correct modeling of turbu- lence within the aerosol will greatly influence the predicted penetration rate and spreading angle, consequently controlling the rate of break-up and agglomeration with knock-on effects on the parti- cle size distribution. Investigation and recommen- dations for the upper airways are made in reference 31, with the low Reynolds number k-ω turbulence model recommended, as also used in reference 19. • Turbulent dispersion models exist for RANS sim- ulations, which simulate the outward spread of particles because they are influenced by turbulent eddies generated within an aerosol. is effect would otherwise be missed in a RANS simulation, as only average flow is predicted. Initial and boundary conditions Quantitative initial conditions and boundary con- ditions must be set for all fluid parameters. Special wall functions are available for turbulence param- eters. Ideally, these should be based on measured data or otherwise be the output from other simula- tions or calculations. Simplifications for rapid tran- sient simulations might be appropriate, such as zero heat transfer through solid boundaries. Sensitivity to these assumptions should be tested before they are adopted wholesale. Solving and storing solutions Solutions should be stored at regular time intervals for transient simulations. During development, where storage capacity allows, each new simulation should be saved as a fresh solution "case" with a description. is means changes to the simulation set-up can be easily tracked and their effects understood, en route to the final version. Post-processing Generation of visualizations and extraction of data allow interpretation of CFD results and support deci- sion making. Often only the images are generated, which can help view the prediction qualitatively and make comparisons between cases. However, data for target parameters should be extracted at important points, along 1-D traverses through the domain and on 2-D surfaces, for quantitative or statistical anal- ysis. Of particular interest are particle size distribu- tions and deposition statistics, both of which require some post-processing to calculate. Alternatively, rele- vant metrics can be calculated in cells or on surfaces as part of the simulation, such as the Sauter mean diameter in the near-nozzle pMDI simulations in reference 14. Validation Where experimental or validation data exist, these must be properly compared with CFD results. is can build trust in the software, the set of results, the