betaFeedback

WorksPhysics and Astronomy

Turbulent flows over porous and rough substrates

Abstract

Turbulent flows over porous substrates are studied via a systematic exploration of the dependence of the flow properties on the substrate parameters, including permeability KK, grain pitch LL, and depth hh. The study uses direct numerical simulations mainly for staggered-cube substrates with L+≈10L^+\approx10 - 5050, K/L≈0.01\sqrt{K}/L\approx0.01 - 0.250.25, and depths from h=O(L)h=O(L) to h≫Lh\gg L, ranging from typical impermeable rough surfaces to deep porous substrates. The results indicate that the permeability has significantly greater relevance than the grain size and microscale topology for the properties of the overlying flow, including the mean-flow slip and the shear across the interface, the drag increase relative to smooth-wall flow, and the statistics and spectra of the overlying turbulence, whereas the direct effect of grain size is only noticeable near the interface as grain-coherent flow fluctuations. The substrate depth also has a significant effect, with shallower substrates suppressing the effective transpiration at the interface. Based on the direct-simulation results, we propose an empirical `equivalent permeability' KeqtK_{eq}^t, that incorporates this effect and scales well the overlying turbulence for substrates with different depths, permeabilities, etc. Based on this, we propose a conceptual h+h^+-K+\sqrt{K^+} regime diagram where, for any given substrate topology, turbulence transitions smoothly from that over impermeable rough surfaces with h=O(L)h=O(L) to that over deep porous substrates with h+≳50h^+\gtrsim50, with the latter limit determined by the typical lengthscale of the overlying pressure fluctuations.

Introduction

Turbulent flows over porous substrates are prevalent in both nature and engineering. They play a central role in diverse problems in environmental science (e.g. forest winds, soil evaporation, sediment transport in water, and riverbed/seabed erosion), aerospace engineering (e.g. surface treatment for drag and noise reduction and boundary layer control), chemical engineering (e.g. heat/mass transfer enhancement in catalyst layers, fluidized beds, and nuclear reactors), metallurgical engineering (e.g. industrial painting and metal foam processing), and light industry (e.g. food dehydration). This subject is characterised by a coupling of two systems originally with distinctively different nature – the turbulent boundary layer, which features vigorous fluctuations, inertia-dominated inter-scale energy transfer, and self-organised flow structures, and the porous medium, in which flow is relatively creeping and viscosity-dominated, and large flow structures spanning many pores are strongly impeded. Such contrast indicates an acute transition of flow behaviour across the interface between the overlying and the subsurface flow, which involves multiple mechanisms with a broad range of characteristic scales. In recent decades, this subject has been increasingly attracting experts from various traditional communities of fluid mechanics, including wall turbulence, flow instability, free shear turbulence, low-Reynolds-number flows, transport phenomena, and chaotic systems. Some typical effects of porous substrates, such as drag increase and mixing enhancement, have been extensively studied in diverse scenarios. However, due to the complicated interplay between different mechanisms, our understanding of the general dependence of the flow behaviour on different characteristics of a porous substrate is still vague.

The most distinctive characteristic of a porous substrate is permeability, i.e. the ability for the overlying flow to penetrate into the substrate. Jimenez et al. [29] represented permeability using a boundary condition where the transpiration, i.e. the wall-normal velocity at a notional interface plane, was proportional to the local instantaneous overlying pressure, with a ‘porosity coefficient’ of proportionality. Their direct numerical simulations (DNSs) showed that such boundary conditions cause the onset of large spanwise rollers associated to a Kelvin-Helmholtz-like (K-H) instability, resulting in an increase in mixing and drag. This boundary condition is a reasonable characterisation of substrates where the flow can travel freely through a plenum below the substrate ([32]).

A practical parameter to characterise substrate permeability is the bulk permeability KK of the porous medium, which is defined as K≡−νU/∂xPK \equiv-\nu U/\partial_x P, where UU is the mean velocity induced by a uniform mean pressure gradient ∂xP\partial_x P and ν\nu is the kinematic viscosity ([13]). Note that here, and elsewhere throughout the paper, any pressure we refer to is the kinematic pressure, that is, the actual, static pressure divided by the density. The above expression for KK can be obtained by volume-averaging the corresponding pore-resolved solution assuming Stokes flow, and K\sqrt{K} is a characteristic permeability lengthscale. Breugem et al. [6] systematically investigated the influence of KK on the overlying turbulence. They conducted DNSs for overlying flows while using a volume-averaged-Navier–Stokes (VANS) approach to model subsurface flows. Their results showed that turbulence differs little from smooth-wall flows for K+≲0.3\sqrt{K^+} \lesssim0.3, where the superscript ‘+’ denotes wall-unit scaling. As K+\sqrt{K^+} increases up to K+≈9\sqrt{K^+} \approx9, typical near-wall structures like low-speed streaks and quasi-streamwise vortices, are gradually destroyed. The significance of K+\sqrt{K^+} to near-wall structures was confirmed by the experiments of Suga et al. [62, 63] for K+≈1\sqrt{K^+} \approx1 - 11. They found that the increase of K+\sqrt{K^+} tends to intensify sweep events and weaken ejection ones near the interface, and presented a conceptual model for the associated destruction of structures. A broader range K+≈2\sqrt{K^+} \approx2 - 17 was experimentally investigated by Manes et al. [43]. Based on the evolution of near-wall vortical structures with K+\sqrt{K^+}, they proposed a theory to describe the competition between two types of eddies: the typical smooth-wall-like eddies, and the spanwise-elongated eddies induced by the K-H instability. The onset and development of this instability was theoretically modeled by Abderrahaman-Elena & García-Mayoral (2017), Sharma et al. [58], and Gómez-de-Segura et al. [24] in a more general situation with anisotropic permeability, establishing a criterion for its onset of Ky+≳0.3\sqrt{K_y^+} \gtrsim0.3 - 0.4, where KyK_y is wall-normal permeability. Habibi Khorasani et al. (2024) have recently corroborated this value in texture-resolving DNSs of mesh-like anisotropic permeable substrates. Focusing on low permeability in a range K+≈0.05\sqrt{K^+} \approx0.05 - 0.7, Rosti et al. [52] conducted VANS-based simulations and suggested a similar critical value, K+≈0.2\sqrt{K^+} \approx0.2, below which almost all the flow statistics are indistinguishable from those of a smooth wall. Voermans et al. [65] measured more flow details across the substrate interface for K+≈0.3−6\sqrt{K^+} \approx0.3 - 6. They could obtain detailed measurements of the flow near the interface and in the subsurface region, not usually accessible in experiments, by matching the refraction index between the flow and the solid inclusions of the substrate [53]. The results confirmed the strong dependence of various interfacial flow properties and penetration depths on K+\sqrt{K^+}. More recently, Wang et al. [66] investigated the transfer of information across the interface. They found strong asymmetry between top-down and bottom-up transfer in terms of both scale and strength for K+∼O(1)\sqrt{K^+} \sim O(1), which provided a novel perspective on understanding the turbulence-subsurface flow interaction.

Since real-world porous media are composed of grains or inclusions of finite size and pitch LL, the free-flow/substrate interface has an irregular topography, and thus a porous substrate also exhibits some features of surface roughness. Some research has aimed to understand the role of surface roughness in the problem of turbulence over porous substrates. A natural strategy to approach this issue is to compare between porous substrates and impermeable rough surfaces with analogous interfacial topography. Such discussions can be at least traced back to Zagni & Smith [70], Kong & Schetz [34], and Zippe & Graf [72]. In their experiments, the corresponding rough surface was obtained by placing a flat and smooth plate, which we term ‘floor’ in this paper, just below the first layer of grains that constitute a porous substrate, so the former has a depth h∼Lh \sim L in contrast to the latter where h≫Lh \gg L. They all observed that a porous substrate induces higher drag than its rough counterpart.

The role of surface roughness in turbulence over porous substrates has drawn more attention in recent decades. Manes et al. [44] explored this issue using particle image velocimetry (PIV). They considered porous substrates and rough surfaces consisting of regularly packed spheres with K+≈31−45\sqrt{K^+} \approx31 - 45, L+≈260−370L^+ \approx260 - 370, and h/L≈5h/L \approx5 and 1. Their measurements confirmed that porous substrates had higher drag coefficients, and they proposed that the intense downward transport of turbulent kinetic energy by pressure fluctuations is an important feature distinguishing porous substrates from impermeable rough surfaces. A similar argument has been proposed by Karra et al. [31] using fully-resolved DNSs for substrates constituted by randomly packed spheres with K+≈3\sqrt{K^+} \approx3-99, L+≈80L^+ \approx80-300300, and h/L≈1h/L \approx1-44. Also using fully-resolved DNSs, Kuwata & Suga [35] compared the flow structures over porous and rough surfaces consisting of staggered cubes with K+≈3\sqrt{K^+} \approx3, L+≈50L^+ \approx50, and h/L≈5h/L \approx5 and 1. They argued that the porous surface has stronger spanwise eddies originating from the K-H instability but weaker near-wall streaks. This is reminiscent of the ‘competing mechanism’ in Manes et al. [43], where they discussed the influence of K+\sqrt{K^+}. Cooper et al. [12] conducted experiments of turbulence over permeable substrates and impermeable surfaces with replicated interfacial topography, suggesting that momentum transfer was more efficient in the permeable case, and Reynolds stresses higher. Fang et al. [18] considered substrates composed of very large grains with K+≈1−100\sqrt{K^+} \approx1 - 100, L+≈250−3000L^+ \approx250 - 3000, and h/L≈3h/L \approx3 and 0.5 using large eddy simulations (LES). Their results suggested that the flow behaviour near the interface depends more on K+\sqrt{K^+} than L+L^+ even for such large grains. To separate the permeability and roughness effects, Esteban et al. [15] provided more experimental data in the ranges K+≈1−60\sqrt{K^+} \approx1 - 60, L+≈10−500L^+ \approx10 - 500, and h/L≈O(10)h/L \approx O(10), 3, and 1. A generic formulation to predict drag increase was proposed based on an analogy between the roles of K+\sqrt{K^+} and L+L^+ in the problem. This generic formulation was partially verified more recently by Wangswaiyjaya et al. (2023). They overlaid external roughness over the surface of a permeable substrate. The substrate had large permeability K+≈10\sqrt{K^+} \approx10–3030 but small grains, while the mesh-like roughness had very large mesh pitch L+≳5000L^+ \gtrsim5000. Their results suggested that the drag increase for the composite permeable-and-rough surface could be characterised by the scale K+/L+\sqrt{K^+}/L^+.

To understand the role of surface roughness in turbulence over porous substrates, Kim et al. [33] instead polished the substrate interface, in essence eliminating its rough character. For K+≈50\sqrt{K^+} \approx50 and L+≈1000L^+ \approx1000, the comparison between the original, unpolished substrate and the polished one showed that the latter induces higher drag than the former. They attributed the difference primarily to the roughness-coherent flow present in the former. A similar strategy was used in Shen et al. [59]. For K+≈3\sqrt{K^+} \approx3 and L+≈80L^+ \approx80. They compared two substrates with a regularly packed and a randomly packed surface layer, respectively, suggesting that the interfacial topographic details affect the flow dynamics.

Overall, the works cited above suggest that turbulence over porous substrates is affected by three characteristics of the substrate. The first is the permeability of the porous medium, characterised by the bulk parameter KK. This is a macroscale property, in the sense that it emerges from a volume-average over scales larger than the grain pitch, and therefore does not reflect the details of the geometry at the grain size, or microscale. The permeability controls the general degree of penetration of the overlying turbulent fluctuations into the substrate. The second characteristic is the granularity of the porous medium, which refers to those microscale features directly associated with the geometrical detail of individual grains. The granularity induces grain-coherent fluctuations in the flow, especially near the interface, with characteristic length scale LL. The third characteristic is the substrate depth, hh. A finite depth tends to suppress the penetration of the overlying flow, thus counteracting the effect of the bulk permeability. For the three characteristics, the literature generally suggests that the permeability and the granularity of a porous medium have some similar phenomenological effects on its overlying flow, such as intensifying the near-wall turbulence and increasing the drag, while these effects are attenuated if the substrate is not sufficiently deep.

In the present paper we aim to characterise the above general trends quantitatively. One difficulty lies in separating the effects of permeability and granularity. In previous studies, the grain topology, which determines the ratio K/L\sqrt{K}/L, has typically little variation. KK and LL are then varied in synchrony, making it difficult to separate their effects. In addition, transpiration is known to be important not only for porous substrates, but also for roughness ([49, 47, 48]). Although strictly speaking rough surfaces are impermeable, in the sense that they do not allow flow through, they allow transpiration in the sense of non-zero wall-normal velocity at a notional interface plane at top of the roughness crests. This transpiration does not only occur at the microscale LL, but also at the macroscale, for the typical sizes of the overlying turbulent eddies. The latter would be more intense for porous substrates, but the question arises of whether the nature of the transpiration effect is different for porous and rough walls, or whether the difference is only in intensity, and a smooth transition can be observed between a porous substrate with h≫Lh \gg L and a corresponding rough surface, with h∼Lh \sim L.

To address these questions, we systematically explore the parameter space of KK, LL, and hh using DNS, aiming to understand the effects of permeability, granularity, and substrate depth as independent parameters on the overlying turbulence. We limit the scope of this study to substrates composed of relatively small grains, 10≲L+≲5010 \lesssim L^+ \lesssim50, i.e. essentially in the transitionally rough regime. In this range, the overlying turbulence deviates from smooth-wall-like behaviour but the near-wall turbulent structures are not fully disrupted by the granularity of the substrate (Abderrahaman-Elena et al. 2019). At L+≈20L^+ \approx20–5050, the length scales of the overlying turbulence and the grain-coherent flow become comparable, and microscale and macroscale cannot be clearly separated [16, 69].

The paper is organised as follows. § 2 describes the substrate configurations, numerical methods and simulation setup, and techniques for post-processing. § 3 reports the general dependence of flow properties on the geometrical parameters of substrates. § 4 discusses the scaling of turbulence with substrate parameters. § 5 investigates the transition from porous substrates to typical rough surfaces. § 6 concludes this paper.

Methods

Configurations of substrates

The substrate configurations considered in this study are arrays of staggered solid cubes with grain pitch LL, inclusion width ℓ\ell, and gap size g=L−ℓg = L-\ell, as shown in figure 1, similar to those of Kuwata & Suga [35]. Compared with collocated arrays, such as those in Breugem & Boersma [7], the staggered arrays are more representative of randomly packed grains that are prevalent in realistic scenarios, where large gaps between grains would be occupied and blocked by other grains. The gap-to-pitch ratio g/Lg/L controls the connectivity of pores, which can be regarded as partially connected for g/L<1/2g/L < 1/2 and fully connected for g/L>1/2g/L > 1/2. The porosity is ε=1−2(ℓ/L)3+max⁡[(2ℓ/L−1)3,0]=1−2(1−g/L)3+max⁡[(1−2g/L)3,0]\varepsilon= 1 - 2(\ell/L)^3 + \max[(2\ell/L - 1)^3, 0] = 1 - 2(1 - g/L)^3 + \max[(1 - 2g/L)^3, 0]. Regular topologies as these have been used often in the literature to study porous substrates (Breugem & Boersma 2005; Breugem et al. 2006; Manes et al. 2009; Chandesris & Jamet 2009; Zhang & Prosperetti 2009; Liu & Prosperetti 2011; Jin et al. 2015; Kuwata & Suga 2016a,b; Fang et al. 2018; Kim et al. 2020; Wang et al. 2021; Rao & Jin 2022; Habibi Khorasani et al. 2024). They present the advantage that in macroscopically homogeneous flow only one pore unit of size LL is needed to obtain volume averaged quantities [7]. Concerns were raised by the reviewers of the paper about substrates with g/L>1/2g/L > 1/2 being made up of ‘frozen suspensions’ of inclusions, and thus not being realisable experimentally. This is indeed the case, but such substrates have nevertheless been used widely in the literature as idealised high-permeability topologies (Prinos et al. 2003; Breugem & Boersma 2005; Chandesris & Jamet 2009; Zhang & Prosperetti 2009; Liu & Prosperetti 2011; Jin et al. 2015; Lācis et al. 2020; Naqvi & Bottaro 2021; Sudhakar et al. 2021; Wang et al. 2021; Rao & Jin 2022; Aghaee-Jouybari et al. 2024). In appendix A, we show that there is no fundamental difference between the flow in these and in substrates made up of interconnected inclusions, and that there is a continuum in the flow properties as g/Lg/L increases above 1/21/2 and the inclusions cease to be interconnected. We have nevertheless included some additional simulations with mesh-like substrates, similar to those of Habibi Khorasani et al. (2024), which could also yield the desired high permeabilities, albeit at an increased computational cost. These substrates were designed so that they exhibited a rough interface with protrusions, like our cube topologies, as shown in figure 1(dd).

For each topology and value of g/Lg/L or ε\varepsilon, the bulk permeability K=−νU/∂xPK = -\nu U/\partial_x P is obtained via a simulation of Stokes flow driven by a uniform pressure gradient ∂xP\partial_x P and yielding a mean velocity UU, as in Sharma & García-Mayoral [57]. We refer to this as the a priori permeability. Alternatively, the value of KK can also be estimated from the Darcy region in a DNS. We refer to this as the a posteriori permeability. In the literature, the former is often denominated the intrinsic permeability, and the latter the effective permeability. For consistency, the Stokes computations were conducted with the same resolution per pitch as the DNSs – see Appendix § 8 for a discussion on spatial resolution.

Both Stokes-flow and DNS values of KK are displayed for our staggered-cube topologies against the porosity ε\varepsilon in Figure 2, showing no significant discrepancies between the two. The figure also shows that a change of g/Lg/L in the range 2/92/9 - 3/43/4 or ε\varepsilon in 0.230.23 - 0.970.97 changes K/L\sqrt{K}/L in 0.0130.013 - 0.2430.243, more than one order of magnitude.

Figure 2. Porosity-permeability relationship for the staggered-cube configuration considered in this study. The green markers represent the a posteriori values resulting from the DNSs for all deep porous (Pd) substrates, with those at Reτ≈360Re_\tau\approx360 and Reτ≈550Re_\tau\approx550 in magenta and purple. The black line represents the a priori values obtained from Stokes-flow simulations at the same resolution. The blue to red crosses are a priori results for increasing resolution and the black ones values kindly provided by one of the manuscript reviewers, as detailed in Appendix § 8. Symbols: ☆ for L+≈12L^+ \approx12; ◯ for L+≈24L^+ \approx24; ◇ for L+≈36L^+ \approx36; ◻ for L+≈48L^+ \approx48.

The interface of a staggered-cube configuration can be defined as the plane through the tips of the first layer of cubes and set to have y=0y = 0. The thickness DD of a grain layer, which is defined as the distance between the tips of two adjacent layers, is D=L/2D = L/2.

The bottom of the substrate is a smooth wall, termed ‘floor’, at y=−hy=-h, where hh is the substrate depth. In this paper, we primarily consider three categories of substrates based on the ratio of the depth hh to the grain-layer thickness DD: deep porous substrates (Pd), with h/D≥5h/D \ge5, shallow porous substrates (Ps), with h/D=2h/D = 2, and rough surfaces (Ro), with h/D=1h/D = 1. The floor of our rough surfaces is thus at y=−D=−L/2y=-D=-L/2, exactly through the tips of the second grain layer for the corresponding deep and shallow porous layouts, where the flow first perceives blockage after penetrating the interface from above. Note that this corresponding rough surface is slightly different from the one in [7] or [35], who set the floor at the bottom of the first layer of cubes. The a posteriori permeability was obtained from our deep-substrate DNSs, Pd, from the mean pressure gradient and the bulk velocity in the third layer of grains from the floor. Our aim is to investigate if and how the character of a given substrate topology changes gradually from being simply rough to being fully permeable, as its depth progressively increases. Although our interest is mainly academic, there are practical cases where substrates would exhibit an intermediate character, being permeable yet thin, e.g. for thin sediment beds or in porous surface coatings or paints. We note that, as detailed in § 1, porous substrates and rough walls with equivalent grain topology or superficial texture have been previously compared extensively, likewise for regular (Manes et al. 2009; Kuwata & Suga 2016a,b; Fang et al. 2018) and irregular topologies [70, 72, 12, 31], and both experimentally and numerically. One of the paper’s reviewers raised concerns, especially for our rough and shallow substrates, about regular topologies having a uniform superficial texture, while irregular ones would have unique, heterogeneous distributions that would make generalisations questionable. It is, however, well established for rough walls, which is what our substrates with h/D=1h/D=1–22 essentially are, that heterogeneities at the grain level do not result in significant differences in the flow over different samples of the same surface [28, 11]. Surface heterogeneity only becomes relevant when it occurs over significantly larger lengthscales, typically comparable to the boundary layer thickness, as studied recently in Li et al. (2021), Wangsawijaya et al. [68] and Stroh et al. [60] and reviewed in Bou-Zeid et al. [5] and García-Mayoral et al. [20].

Numerical methods and simulation setup

The direct numerical simulations in this paper are conducted in channels bounded by a pair of parallel substrates with identical configurations. The two substrates are symmetric about the channel central plane and the distance between their interfaces is 2δ2\delta, as portrayed in Figure 1(b). The size of the channel is 2πδ2\pi\delta in the streamwise (xx) direction and πδ\pi\delta in the spanwise (zz) direction. Periodic boundary conditions are applied in xx and zz. A constant mean pressure gradient ∂xP<0\partial_x P < 0 is imposed in xx in the entire domain to drive the channel flow. The substrate parameters in wall units are based on the kinematic viscosity ν\nu and the friction velocity uτu_\tau measured at the interface y=0y = 0, i.e. uτ=−∂xPδu_\tau= \sqrt{-\partial_x P\delta}. We note that this value of uτu_\tau is not strictly the one that is expected to scale the flow, which would be set at the virtual origin at y=−Δydy = -\Delta y_d perceived by the overlying turbulence, i.e. uτ=−∂xP(δ+Δyd)u_\tau= \sqrt{-\partial_x P(\delta+ \Delta y_d)} (Luchini [41]; Ibrahim et al. [27]). Nevertheless, as we will see in § 2.3, the difference between the two values of uτu_\tau is never larger than 1%.

Sketches of porous substrates and computational domain used in the DNSs

Figure 1. Sketches of porous substrates and computational domain used in the DNSs. LL and gg are the grain pitch and the gap size, and hh is the substrate depth. The wall-normal coordinate is set to y=0y = 0 at the interface of the bottom substrate with the free flow, the plane of the tips of the top layer of cubes. (a) Staggered-cube topology, where D≡L/2D \equiv L/2 is the thickness of one cube layer. (b) Computational domain, with dimensions 2πδ2\pi\delta, 2(δ+h)2(\delta+ h), and πδ\pi\delta in xx, yy, and zz, respectively. (c) Plan and side views of staggered-cube substrates of different depths. (d) Mesh topology. (e) Side view of a mesh substrate.

We use the computational code of Sharma & García-Mayoral [56], which is briefly summarised here. The code solves the incompressible Navier–Stokes equations using a pseudo-spectral discretisation in the xx and zz directions and a finite difference discretisation in the yy direction. A three-step Runge–Kutta scheme with a fractional-step method is used for the time discretisation. No-slip conditions are imposed on all solid surfaces of substrates through an immersed boundary method. The code features a multiblock structure (García-Mayoral & Jiménez [21]) that allows local refinement of near-wall grids in xx and zz to resolve small texture, while retaining a coarser resolution away from substrates, sufficient to fully resolve the turbulence. Full details of the numerical method and its validation can be found in Sharma [55].

In the substrate blocks, for the staggered cubes the resolution is set to 24-32 grid points in xx and zz per grain period LL. The number of grid points covering one gap size gg is 13 or greater, except for cases Pd-12-33, Ro/Pd-24-25, Pd-36-22, Ro/Pd-48-22, Pd-48-25, and Ro/Pd-48-28 (nomenclatures to be introduced below), for which it is 9-11. For these cases, both the porosity and the permeability are low, and thus the interfacial turbulent fluctuations are weak. In this situation, the grid-dependence investigation in Sharma & García-Mayoral [57] showed that further refinement had no significant influence. For the mesh substrates, the resolution is set to 36 grid points per grain period LL, with 8-16 points resolving the mesh liquid width ℓ\ell, depending on the ratio g/Lg/L. The fine-resolution blocks reach up to y≈2Ly \approx2L away from the substrate interface into the channel core, sufficient for the fine scales induced by the grain geometry to have vanished already. In the channel-core block, the resolution is set lower to Δx+≈6\Delta x^+ \approx6 and Δz+≈3\Delta z^+ \approx3. In the yy direction, the grid is finest with Δy+≈0.35\Delta y^+ \approx0.35 near the interface y=0y = 0, where the shear is higher, and is gradually stretched away from this plane, as in Chen & García-Mayoral [10]. For y>0y > 0, the grid spacing is stretched to Δy+≈3.5\Delta y^+ \approx3.5 near the channel centre; for y<0y < 0, the grid spacing is stretched to Δy≈Δx=Δz\Delta y \approx\Delta x = \Delta z at y=−2D=−Ly = -2D = -L and becomes uniform for y<−2Dy < -2D.

Table 1 lists the basic parameters of all 58 DNSs presented in this paper, among which 52 cases have friction Reynolds number Reτ≡uτδ/ν≈180Re_\tau\equiv u_\tau\delta/\nu\approx180, 4 cases with suffix ‘HR’ have Reτ≈360Re_\tau\approx360, and 2 cases with suffix ‘HHR’ have Reτ≈550Re_\tau\approx550. The cases at higher Reynolds numbers were run to verify that the effect of the texture is Reynolds-number independent in viscous scaling, with results compared in Appendix C. The two numbers in the label of a case indicate its pitch L+L^+ and gap-to-pitch ratio g/Lg/L, respectively, e.g. ‘Pd-24-56’ has L+≈24L^+ \approx24 and g/L≈0.56g/L \approx0.56. The prefix ‘Ro/Ps/Pd’, as introduced in §2.1 and Figure 1(c), indicates the depth h=1Dh = 1D, 2D2D, or ≥5D\ge5D; there is also one very deep case with suffix ‘VD’, with h=9Dh = 9D. A letter ‘M’ precedes the name for cases with interconnected-mesh substrates. The parameter space considered is portrayed in Figure 3. The porosity ε\varepsilon, pitch L+L^+, and permeability K+K^+ are in the ranges ε≈0.23\varepsilon\approx0.23-0.970.97, L+≈12L^+ \approx12-4848, and K+≈0.4\sqrt{K^+} \approx0.4-8.18.1, respectively.

Figure 3. DNS cases in this study represented in (a) K+\sqrt{K^+}-ε\varepsilon and (b) K+\sqrt{K^+}-L+L^+ parameter spaces. The isolines with embedded numbers represent constant values of the gap size g+g^+. Symbols: ☆ for L+≈12L^+ \approx12; ○ for L+≈24L^+ \approx24; ◇ for L+≈36L^+ \approx36; □ for L+≈48L^+ \approx48.

Techniques for post-processing

Table 2 lists the main properties resulting from all the DNSs. For each simulation, we obtain the flow statistics and spectra by averaging multiple instantaneous fields over a period of time at least 12δ/uτ12\delta/u_\tau after turbulence reaches a statistically steady state. Unless otherwise stated, the statistics at a yy location below the interface are spatially averaged over the entire xx-zz plane containing both fluid and solid areas, i.e. the ‘superficial average’.

We use the mean-velocity deficit in wall units, ΔU+\Delta U^+, commonly known as the roughness function, to quantify how a substrate increases drag compared to a smooth wall. Following classical turbulence theory, the mean velocity profile U+(y+)U^+(y^+) in the logarithmic layer over a complex surface is shifted relative to a smooth-wall one, USm+(y+)U^+_{\mathrm{Sm}}(y^+),

U+(y+)=κ−1ln⁡(y++Δyd+)+B−ΔU+,(1)U^+(y^+) = \kappa^{-1}\ln\left(y^+ + \Delta y_d^+\right) + B - \Delta U^+, \tag*{(1)}

where κ≈0.4\kappa\approx0.4 and B≈5B \approx5 are the constants characterising USm+(y+)U^+_{\mathrm{Sm}}(y^+) in the logarithmic layer and Δyd+\Delta y_d^+ is the zero-plane displacement, the yy-coordinate offset that results in a best collapse for the outer flow. We follow Chen & García-Mayoral [10] and extend this relation above the logarithmic layer to include the wake region, for a robust estimation of ΔU+\Delta U^+ by adjusting Δyd\Delta y_d to maximise the region sufficiently above the surface where U+(y+)U^+(y^+) and USm+(y++Δyd+)U^+_{\mathrm{Sm}}(y^+ + \Delta y_d^+) are parallel. Nevertheless, the resulting values of ΔU+\Delta U^+ are almost identical to the velocity difference ΔUSm+(δ)−U+(δ)\Delta U^+_{\mathrm{Sm}}(\delta) - U^+(\delta) measured at the channel centre y=δy = \delta without zero-plane offsetting, both listed in table 2. For all the cases studied, the distance between the interface and the zero-plane-displacement height was under 3 wall units, and the resulting difference in uτu_\tau under 1%, as mentioned above.

The mean pressure gradient ∂xP\partial_x P driving the flow induces a Darcy velocity UDaU_{\mathrm{Da}} within the substrate, which is not present in an external-flow application where ∂xP≃0\partial_x P \simeq0 (see [23]). In an internal flow, UDa+U^+_{\mathrm{Da}} scales with

CaseReτRe_{\tau}L+L^{+}g/Lg/Lh/Dh/Dϵ\epsilonK+\sqrt{K}^{+}NxN_xNzN_zNycN_{yc}NysN_{ys}
Pd-12-33182.712.01/350.440.37(0.37)2304115217664
Ro-12-50182.712.01/210.750.912304115217616
Ps-12-50182.712.01/220.750.912304115217628
Pd-12-50182.712.01/250.750.91(0.88)2304115217664
Pd-12-67182.712.02/370.932.03(1.97)2304115217688
Ro-24-25182.723.91/410.280.39153676817628
Pd-24-25182.723.91/470.280.39(0.39)1536768176126
Ro-24-38182.723.93/810.530.94153676817628
Ps-24-38182.723.93/820.530.94153676817646
Pd-24-38182.723.93/870.530.94(0.95)1536768176126
Ro-24-50182.723.91/210.751.82153676817628
Ps-24-50182.723.91/220.751.82153676817646
Pd-24-50182.723.91/250.751.82(1.76)153676817694
Pd-24-50-HR360.023.61/250.751.79(1.69)2304115235274
Pd-24-50-VD182.723.91/290.751.82(1.78)1536768176158
Pd-24-56182.723.99/1670.832.44(2.42)1536768176126
Ro-24-62182.723.95/810.893.33153676817628
Ps-24-62182.723.95/820.893.33153676817646
Pd-24-62182.723.95/870.893.33(3.29)1536768176126
Ro-24-67182.723.92/310.934.06172886417629
Ro-24-75182.723.93/410.975.80153676817628
Ps-24-75182.723.93/420.975.80153676817646
Pd-24-75182.723.93/470.975.80(5.73)1536768176126
Pd-36-22182.735.92/950.230.45(0.45)1152576176112
Ps-36-33182.735.91/320.441.10115257617651
Pd-36-33182.735.91/350.441.10(1.11)115257617696
Pd-36-39182.735.97/1870.551.55(1.50)1152576176133
Ro-36-50182.735.91/210.752.73115257617633
Ps-36-50182.735.91/220.752.73115257617649
Pd-36-50182.735.91/260.752.73(2.67)1152576176105
Pd-36-50-HR360.035.31/260.752.69(2.54)153676835291
Ro-36-67182.735.92/310.936.10115257617634
Ps-36-67182.735.92/320.936.10115257617651
Pd-36-67182.735.92/360.936.10(5.95)1152576176111
Ro-48-22182.747.82/910.230.6086443217644
Pd-48-22182.747.82/950.230.60(0.60)864432176117
Pd-48-25182.747.81/450.280.77(0.75)768384176106
Ro-48-28182.747.85/1810.330.9886443217644
Pd-48-28182.747.85/1850.330.98(0.97)864432176117
Ro-48-38182.747.83/810.531.8876838417640
Ps-48-38182.747.83/820.531.8876838417658
Pd-48-38182.747.83/870.531.88(1.89)768384176138
Pd-48-38-HR360.047.13/870.531.85(1.87)1536768352138
Ro-48-44182.747.87/1610.652.6676838417640
Pd-48-44182.747.87/1660.652.66(2.60)768384176122
Ro-48-50182.747.81/210.753.6476838417640
Ps-48-50182.747.81/220.753.6476838417658
Pd-48-50182.747.81/260.753.64(3.50)768384176122
Pd-48-50-HHR550.448.01/260.753.66(3.42)1728864484107
Ro-48-61182.747.811/1810.886.2886443217644
Ps-48-61182.747.811/1820.886.2886443217663
Pd-48-61182.747.811/1860.886.28(6.05)864432176135
Ro-48-62182.747.85/810.896.6676838417640
Ps-48-62182.747.85/820.896.6676838417658
Pd-48-62182.747.85/870.896.66(6.42)768384176138
Pd-48-62-HR360.047.15/870.896.56(6.35)1536768352138
Pd-48-62-HHR550.448.05/870.896.69(6.47)23041152484137
Ro-48-67182.747.82/310.938.1386443217644
MPd-36-78182.735.97/9100.875.34(5.16)1152576176202
MRo-48-56182.747.85/910.583.3086443217644
MPs-48-56182.747.85/920.583.3086443217663
MPd-48-56182.747.85/9100.583.30(3.18)864432176207
MPd-48-72182.747.813/18100.815.94(5.68)864432176207
MPd-48-78182.747.87/9100.877.12(6.71)864432176207

Table 1. Simulation parameters. LL is the grain pitch, gg the gap size, hh the substrate depth, D=L/2D = L/2 the thickness of one grain layer, ϵ\epsilon the porosity, and KK the permeability, with a posteriori values in parenthesis. The number of grid points is NxN_x in xx, NzN_z in zz, and NycN_{yc} and NysN_{ys} in yy in the free-flow region and for the substrates, respectively.

Reτ−1Re_\tau^{-1}, which is intrinsically different from other near-wall quantities that are essentially independent of ReτRe_\tau when normalised in wall units. To allow for direct application to external flows, we subtract UDa+U_{\mathrm{Da}}^+ from the mean velocity U+U^+ to evaluate the drag increase ΔU+\Delta U^+, thus redefined by

U+(y+)=κ−1ln⁡(y++Δyd+)+B+UDa+−ΔU+.(2)U^+(y^+) = \kappa^{-1}\ln\left(y^+ + \Delta y_{\mathrm{d}}^+\right) + B + U_{\mathrm{Da}}^+ - \Delta U^+. \tag*{(2)}

Similarly, we define the slip velocity Us+U_{\mathrm{s}}^+ as

Us+=U+∣0−UDa+,(3)U_{\mathrm{s}}^+ = U^+|_0 - U_{\mathrm{Da}}^+, \tag*{(3)}

where U+∣0U^+|_0 is the mean velocity at the interface. For a sufficiently deep substrate, the Darcy velocity is UDa≈−ν−1K ∂xPU_{\mathrm{Da}} \approx-\nu^{-1}K\,\partial_x P, while for a finite-depth substrate it can be approximated by equation (D 2a) evaluated at y=0y=0 – note that both results converge for h≫Kh \gg\sqrt{K}. In any event, Table 2 shows that UDa+U_{\mathrm{Da}}^+ is essentially negligible compared with ΔU+\Delta U^+ or Us+U_{\mathrm{s}}^+.

CaseUs+U_s^+ℓU+\ell_U^+rshr_{sh}rνr_\nuUDa+U_{\mathrm{Da}}^+Keqs\sqrt{K_{eq}^s}Keqs\sqrt{K_{eq}^s}Keqt\sqrt{K_{eq}^t}ΔUs+\Delta U_s^+ΔU+\Delta U^+
Pd-12-330.430.420.240.120.000.370.370.310.250.25
Ro-12-500.810.800.400.350.000.910.910.340.500.50
Ps-12-500.810.800.400.340.000.910.910.510.800.79
Pd-12-500.800.800.400.340.000.910.910.770.980.97
Pd-12-671.231.580.550.510.022.032.021.865.265.25
Ro-24-250.430.430.200.060.000.390.390.220.400.39
Pd-24-250.420.430.200.070.000.390.390.380.430.41
Ro-24-380.800.800.320.170.000.940.940.510.991.04
Ps-24-380.800.810.320.160.000.940.940.721.291.31
Pd-24-380.800.810.320.160.000.940.940.931.281.27
Ro-24-501.351.410.440.340.021.821.810.941.701.71
Ps-24-501.281.490.430.240.021.821.811.353.643.71
Pd-24-501.241.520.430.200.021.821.811.754.324.28
Pd-24-50-HR1.201.450.420.210.011.791.781.724.054.05
Pd-24-50-VD1.231.530.430.200.021.821.811.804.554.54
Pd-24-561.352.030.460.240.032.442.412.406.376.36
Ro-24-621.962.330.550.690.053.333.291.643.383.44
Ps-24-621.582.560.530.290.063.333.272.336.456.44
Pd-24-621.422.660.510.240.063.333.253.237.667.65
Ro-24-672.162.770.590.840.084.063.991.954.014.01
Ro-24-752.463.690.650.980.145.805.532.545.125.13
Ps-24-752.014.310.610.380.165.805.513.687.847.85
Pd-24-751.694.530.590.190.185.805.365.309.679.66
Pd-36-220.460.460.190.050.000.450.450.450.510.53
Ps-36-330.890.920.290.110.011.101.090.961.681.69
Pd-36-330.890.920.290.100.011.101.091.091.791.79
Pd-36-391.131.300.330.090.011.551.531.523.273.23
Ro-36-501.702.170.430.200.042.732.671.603.803.81
Ps-36-501.462.250.420.130.042.732.662.255.815.80
Pd-36-501.402.270.410.110.042.732.652.646.486.46
Pd-36-50-HR1.322.110.410.120.022.692.612.615.855.79
Ro-36-672.304.130.570.500.156.105.733.216.066.02
Ps-36-671.814.390.550.200.186.105.584.448.448.45
Pd-36-671.614.440.540.150.206.105.485.469.719.73
Ro-48-220.540.540.190.050.000.600.600.460.800.85
Pd-48-220.550.550.190.050.000.600.600.600.870.84
Pd-48-250.620.630.220.070.000.770.770.770.900.91
Ro-48-280.780.790.250.070.010.980.970.731.111.14
Pd-48-280.780.800.250.070.010.980.970.971.481.50
Ro-48-381.281.490.320.070.021.881.821.272.983.00
Ps-48-381.221.540.310.050.021.881.811.664.024.01
Pd-48-381.201.550.310.050.021.881.811.814.354.36
Pd-48-38-HR1.141.540.310.050.011.851.781.784.434.44
Ro-48-441.572.100.370.080.032.662.531.704.244.26
Pd-48-441.372.150.360.060.042.662.532.525.825.81
Ro-48-501.822.780.420.120.063.643.412.234.964.97
Ps-48-501.522.800.410.100.073.643.423.096.997.01
Pd-48-501.462.800.400.110.073.643.443.447.347.32
Pd-48-50-HHR1.392.760.400.100.023.663.433.437.167.10
Ro-48-612.164.370.500.240.156.285.623.556.716.71
Ps-48-611.724.460.490.140.196.285.514.858.808.81
Pd-48-611.574.440.490.110.226.285.425.429.949.98
Ro-48-622.194.530.520.260.176.665.943.746.776.75
Ps-48-621.724.640.500.140.216.665.785.078.968.97
Pd-48-621.574.590.500.120.246.665.685.689.939.97
Pd-48-62-HR1.584.790.500.110.126.665.565.569.599.60
Pd-48-62-HHR1.585.000.500.100.086.695.585.589.729.75
Ro-48-672.355.400.550.270.218.136.814.157.317.33
MPd-36-782.557.780.680.300.165.345.115.118.748.73
MRo-48-562.184.110.480.090.053.303.081.976.056.03
MPs-48-562.164.110.480.100.063.303.132.846.376.37
MPd-48-562.024.100.480.090.063.303.123.127.307.30
MPd-48-722.557.970.620.230.195.945.545.548.748.76
MPd-48-782.629.910.670.270.287.126.536.539.529.55

Table 2. Substrate properties obtained from DNS. UsU_s and ℓU\ell_U are the mean slip velocity and slip length, respectively; rshr_{sh} is the inner/outer shear ratio across the interface and rνr_\nu the effective viscosity ratio (see appendix D); KeqsK_{eq}^s and KeqtK_{eq}^t are the shear- and transpiration-based equivalent permeabilities; ΔUs+\Delta U_s^+ is the velocity deficit at the channel centre, and ΔU+\Delta U^+ the one obtained with a zero-plane displacement and optimal outer-layer matching.

Effect of substrate parameters on turbulence

In this section, we report the dependence of flow properties on the substrate geometry. Research exploring the parameter space of the substrate properties, e.g. permeability, porosity, depth, grain size, etc, is typically limited in which variations are possible. A sweep through Reynolds number for a fixed substrate, for instance, would vary K+\sqrt{K^+}, but in doing so would also vary L+L^+ proportionately. The present set of simulations has been designed so that the effect of parameter pairs such as K+\sqrt{K^+} and L+L^+ could be analysed separately. We therefore discuss the isolated effect of one parameter at a time – the grain spacing, or pitch, L+L^+, the porosity ε\varepsilon, or equivalently the gap-to-pitch ratio, g/Lg/L, and the relative depth, h/Dh/D.

Effect of porosity under fixed grain spacing

First, let us consider deep porous (Pd) substrates with fixed L+L^+ but varying K+\sqrt{K^+} by varying g/Lg/L. Figure 4 shows instantaneous velocity fields on the xx-yy plane for six deep porous substrates with identical L+≈24L^{+} \approx24 but varying g/Lg/L from 1/41/4 to 3/43/4, corresponding to K+\sqrt{K^{+}} from 0.4 to 5.8. For increasing g/Lg/L, the streamwise velocity u+u^{+} of the subsurface flow generally increases, while regions of relatively low speed become more prevalent immediately above the interface. The changes in u+u^{+} and v+v^{+} suggest a gradually intensified penetration of the overlying flow into the substrate. There is strong impedance to the overlying turbulent eddies, which typically span multiple grains, penetrating into the substrate, and their footprint below the interface is much attenuated, and dispersed lengthscale-wise, by the presence of the individual grains.

Instantaneous fields of velocity components $u^+$ (left) and $v^+$ (right) on an $x$-$y$ plane

Figure 4. Instantaneous fields of velocity components u+u^+ (left) and v+v^+ (right) on an xx-yy plane for deep-porous substrates with identical pitch L+≈24L^+ \approx24 but different gap-to-pitch ratio g/L=0.25g/L = 0.25–0.750.75, from top to bottom substrates Pd-24-25/38/50/56/62/75. Colours from blue to red correspond for u+u^+ to [0:5] and for v+v^+ to [−0.8:0.8-0.8:0.8].

The profiles of mean velocity U+U^{+}, Reynolds shear stress u′+v′+‾\overline{u^{\prime+}v^{\prime+}}, and root-mean-square (RMS) velocity fluctuations urms+u_{\mathrm{rms}}^{+}, vrms+v_{\mathrm{rms}}^{+}, and wrms+w_{\mathrm{rms}}^{+} for the above six substrates are portrayed in figure 5. For each case, the U+U^{+} profile below the interface in figure 5(b) shows a near-interface region with strong shear, i.e. the Brinkman layer. Further below is a plateau where U+U^{+} essentially results from the mean pressure gradient, i.e. the Darcy region. The Brinkman layer is thicker for larger g/Lg/L, indicating a deepened penetration of shear of the overlying flow. This also results in a slight increase of the slip velocity, Us+U_{s}^{+}, but generally decreases U+U^{+} above the interface, as shown in figure 5(a), which leads to an increase of the drag coefficient. Of the configurations discussed here, only those with smaller permeability, K+≲1\sqrt{K^{+}} \lesssim1, exhibit a smooth-wall-like character. Those with greater permeability experience significant departures from smooth-wall turbulence, exhibiting the usual decrease of u′u^{\prime} and increase in v′v^{\prime} and w′w^{\prime} near the wall, with all three converging towards similar peak values, together with an increase in near-wall Reynolds shear stress and the corresponding increase in ΔU+\Delta U^{+} and drag. Given that similar intense departures from the smooth-wall-like regime occur across the whole set of configurations studied, the virtual-origin framework proposed in Ibrahim et al. [27] will not be used here, as it only applies to smooth-wall-like turbulence.

The drag increase with increasing g/Lg/L is directly associated to the changes in the u+′v+′‾\overline{u^{+\prime}v^{+\prime}} profiles (see Gómez-de-Segura & García-Mayoral [23], §5.3), shown in Figure 5(c) for y+≈0y^+ \approx0–3030. Their magnitude increases significantly relative to smooth-wall values for g/L≳0.4g/L \gtrsim0.4 or K+≳1\sqrt{K^+} \gtrsim1. These profiles of u+′v+′‾\overline{u^{+\prime}v^{+\prime}}, together with urms+u_{\mathrm{rms}}^+, vrms+v_{\mathrm{rms}}^+, and wrms+w_{\mathrm{rms}}^+ in Figure 5(d, Figure 5e, Figure 5f), illustrate the gradually enhanced penetration of turbulence into the substrates as g/Lg/L increases. The penetration of urms+u_{\mathrm{rms}}^+ is accompanied by a drop of its peak value above the interface, while such a drop is not observed for vrms+v_{\mathrm{rms}}^+ or wrms+w_{\mathrm{rms}}^+. Similar trends have been observed not only for porous substrates ([6]) but also for rough surfaces ([39] (Abderrahaman-Elena et al. 2019)) and canopies ([57]). These trends have been interpreted by some authors ([28, 19]) as the flow losing some of the anisotropic characters of the near-wall cycle. It is also notable that for all cases studied, vrms+v_{\mathrm{rms}}^+ decays with the depth into the substrate more slowly than urms+u_{\mathrm{rms}}^+ and wrms+w_{\mathrm{rms}}^+, a feature consistent with the study for dense canopies by Sharma & García-Mayoral [57]. This implies that, for a finite-depth porous substrate, the wall-normal velocity fluctuations in the subsurface flow are more likely to perceive the presence of the substrate floor than the tangential (i.e. wall-parallel) fluctuations. This will be further discussed in §5.

Figure 5. (a–b) Mean velocity profile, (c) Reynolds shear stress, and (d–f) RMS velocity fluctuations for deep porous substrates with identical L+≈24L^{+} \approx24 but different g/L=0.25−0.75g/L = 0.25 - 0.75. Colours from blue to red are for cases Pd-24-25/38/50/56/62/75, and dash-dotted lines for smooth-wall data. The dotted lines in panel (b) are Darcy-Brinkman analytical solutions, equation (D 2), for the mean velocity within the substrate, and the dashed lines mark the location of the free-flow/substrate interface.

Effect of grain spacing under fixed porosity

Next, we consider the deep porous (Pd) substrates with the same gap-to-pitch ratio g/L=1/2g/L = 1/2 but different pitch L+L^+. Figure 6 displays the changes in flow statistics as the pitch L+L^+ increases from 12 to 48, which corresponds to K+\sqrt{K^+} increasing from 0.9 to 3.6. The changes in terms of drag, subsurface mean flow and turbulence penetration are qualitatively similar to those changes with g/Lg/L increasing illustrated in Figure 5. The dominant factor that underlies both the g/Lg/L-induced changes and the L+L^{+}-induced changes will be discussed in §4.2. In both instances, we note that the differences in the value of the mean velocity at the interface is small, while the differences far above the substrate are significant. This can be traced to the increase in the shear Reynolds stress profile, as discussed above. The latter shows a good correlation with the fluctuating transpiration vrms+v_{\mathrm{rms}}^{+} at the interface ([1]), which in turn has been shown to correlate well with ΔU+\Delta U^{+} for rough surfaces ([49, 47, 48]). This suggests that the increase in drag is more closely connected to the transpiration than to the tangential velocity at the interface.

Figure 6. (a(a–b)b) Mean velocity profile, (c)(c) Reynolds shear stress, and (d(d–f)f) RMS velocity fluctuations for deep porous substrates with identical g/L=0.50g/L = 0.50 but different L+=12L^+ = 12–4848. Colours from blue to red are for cases Pd-12/24/36/48-50, and dash-dotted lines for smooth-wall data. The dashed lines mark the location of the free-flow/substrate interface.

In contrast to the cases with identical L+L^{+} in Figure 5, the four cases with identical g/L=1/2g/L = 1/2 in Figure 6 share a similarity in substrate geometry, and thus exhibit some degree of similarity in the decay of subsurface flow properties. In Figure 7, where the flow statistics are normalised by the corresponding interfacial values and the wall-normal coordinate is normalised by DD, the four cases have a similar decaying trend for vrms+v_{\mathrm{rms}}^{+} in most of the subsurface region with y<0y < 0, while for urms+u_{\mathrm{rms}}^{+} and wrms+w_{\mathrm{rms}}^{+} the similarity occurs only for y≲−1Dy \lesssim-1D. This suggests that nonlinear inertial effects, which break the similarity of the subsurface flow, are largely limited to the near-interface region, and mainly influence tangential motions only. This would thus yield the difference in the decaying rate of near-interface UU among the four cases, shown in Figure 7(b).

Figure 7. (a) Shear-driven component of the mean velocity, (b) Reynolds shear stress, and (c-e) RMS velocity fluctuations for the flow within the substrate, normalised by the corresponding interfacial values and the thickness of one layer of cubes, DD, for the same cases of Figure 6. Colours are as in Figure 6.

Effect of substrate depth

Lastly, for a fixed pitch L+≈24L^{+} \approx24 and gap-to-pitch ratio g/L=1/2g/L = 1/2, we compare four cases with varying depths h/D=1,2,5h/D = 1, 2, 5, and 99, which are respectively labeled as Ro-24-50, Ps-24-50, Pd-24-50, and Pd-24-50-VD. As shown in Figure 8, as the depth hh increases from 1D1D to 5D5D, we observe an increase in drag and a deeper penetration of turbulence, which are qualitatively similar to the changes with increasing g/Lg/L in Figure 5 or with increasing L+L^+ in Figure 6. However, the present four cases demonstrate no substantial differences in U+U^+, urms+u_{\mathrm{rms}}^{+}, and wrms+w_{\mathrm{rms}}^{+} in the range y≈−1y \approx-1–0D0D, indicating that the change of hh has little influence on the penetration of tangential velocity components. Quantitative discussion on this phenomenon will be presented in §4.1.

Figure 8. (a-b) Mean velocity profile, (c) Reynolds shear stress, and (d-f) RMS velocity fluctuations for substrates with identical L+≈24L^+ \approx24 and g/L=0.50g/L = 0.50 but different depth h=1Dh = 1D–9D9D. Colours from blue to red are for cases Ro-24-50 (h/D=1h/D = 1), Ps-24-50 (h/D=2h/D = 2), Pd-24-50 (h/D=5h/D = 5), and Pd-24-50-VD (h/D=9h/D = 9), and dash-dotted lines for smooth-wall data. The dashed lines mark the location of the free-flow/substrate interface.

The changes in flow statistics caused by the increase in hh diminish gradually. Eventually, for h=5Dh = 5D and 9D9D, all the statistics become essentially indistinguishable except for vrms+v_{\mathrm{rms}}^{+} deep inside the substrate, where the wall-normal fluctuations seem always able to penetrate to the floor, as discussed in Sharma & García-Mayoral [57]. Nevertheless, vrms+v_{\mathrm{rms}}^{+} becomes ultimately negligible below y≈−4Dy \approx-4D even for the substrate with depth h=9Dh = 9D. Above y≈−4Dy \approx-4D, an asymptotic state is already reached for h≥5Dh \ge5D. This suggests that h=5Dh = 5D is a depth sufficient for the overlying turbulence to essentially no longer perceive the floor. This concept of ‘sufficient depth’ will be further investigated in §5.

Scaling of turbulence with substrate parameters

This section discusses the scaling of the overlying flow properties with substrate parameters. We compare the raw collapse of properties such as slip velocity, slip length and, especially, roughness function, with different substrate parameters, without making any a priori assumptions on which parameters will produce a better collapse. This is used to provide insight into the separate roles in the problem of the macroscale permeability, the microscale granularity, and the substrate depth.

Slip and shear across the interface

The values of the mean slip velocity Us+U_s^+ for all the cases in this study are portrayed versus different substrate parameters in Figure 9. No apparent correlations are found between Us+U_s^+ and any of the pitch L+L^+, the gap size g+g^+, the inclusion size ℓ+\ell^+, the porosity ε\varepsilon, or the depth h+h^+. The values of Us+U_s^+ for deep porous (Pd) substrates, however, tend to correlate well with K+\sqrt{K^+}, as shown in Figure 9(f). This suggests that the slip velocity for deep substrates is essentially determined by their macroscale permeability, and is not directly associated with the microscale details of the individual grains. Similar results were also observed by Efstathiou & Luhar [14].

Six-panel plots of mean slip velocity versus substrate parameters

Figure 9. Mean slip velocity, Us+U_s^+, for all the substrates studied versus (a) pitch L+L^+, (b) gap size g+g^+, (c) inclusion size ℓ+\ell^+, (d) porosity ε\varepsilon, (e) depth h+h^+, (f) permeability K+\sqrt{K^+}. ☆, L+≈12L^+ \approx12; ◯, L+≈24L^+ \approx24; ◇, L+≈36L^+ \approx36; □, L+≈48L^+ \approx48; ●, very deep substrate Pd-24-50-VD; ◁, mesh substrates. Blue, yellow and red colours are for deep porous (Pd), shallow porous (Ps), and rough (Ro) substrates at Reτ≈180Re_\tau\approx180, respectively; magenta for Reτ≈360Re_\tau\approx360; purple for Reτ≈550Re_\tau\approx550. Symbols connected by dashed lines have the same gap-to-pitch ratio g/Lg/L and porosity ε\varepsilon.

For K+≲2\sqrt{K^+} \lesssim2, the values of Us+U_s^+ for rough surfaces in Figure 9(f) agree roughly with Us+U_s^+ for their corresponding deep substrates. For K+≳2\sqrt{K^+} \gtrsim2, however, the former are higher and more scattered than the latter. This is in spite of the slip length having a strong correlation with the permeability, as shown in Figure 10, with the slip length defined as ℓU=Us/∂yU∣0\ell_U = U_s / \left.\partial_y U\right|_0, where ∂yU∣0\left.\partial_y U\right|_0 is the mean shear in the free flow at y=0y = 0. Figure 10(f) shows that ℓU+\ell_U^+ is roughly proportional to K+\sqrt{K^+}, with a constant of proportionality of order 0.7–1. This is consistent with the analysis [2, 23] based on a homogenised model for the subsurface flow, which leads to ℓU+∼K+\ell_U^+ \sim\sqrt{K^+}. The results in 10(f) indicate that under a fixed value of shear at the interface, the slip length is essentially independent of the substrate depth, even for depths as shallow as h=1D=L/2h = 1D = L/2. The different behaviours of the slip length Us+U_s^+ and the slip velocity ℓU+\ell_U^+ are caused by different non-zero Reynolds shear stresses u+′v+′u^{+\prime}v^{+\prime} at the interface plane. In their absence, the shear in viscous units would be ∂y+U+=1\partial_y^+ U^+ = 1, and both quantities would have equal value. The comparison of Figure 9(f) and Figure 10(f) suggests that the substrate depth plays a key role in this difference.

Six panels showing mean slip length versus pitch, gap size, inclusion size, porosity, depth, and permeability

Figure 10. Mean slip length, ℓU+\ell_U^+, versus (a) pitch L+L^+, (b) gap size g+g^+, (c) inclusion size ℓ+\ell^+, (d) porosity ε\varepsilon, (e) depth h+h^+, and (f) permeability K+\sqrt{K^+}. Symbols and colours are as in Figure 9. The two dotted lines in (f) are for ℓU+=0.7K+\ell_U^+ = 0.7\sqrt{K^+} and ℓU+=1.0K+\ell_U^+ = 1.0\sqrt{K^+}.

Another feature of our porous and rough substrates in terms of tangential velocities is the discontinuity of shear ∂yU\partial_y U across the interface, shown in Figure 11. A force balance in a thin volume containing the interface shows that the mean shear stress just above and below are different, as the shear stress above is partly balanced by the shear force between the fluid and the flat top surface of the substrate elements. This effect is in our case concentrated at the element tips, but can be expected to be more diffuse in substrates composed of rounder grains and with less even interfaces. Figure 12 portrays the ratio of inner to outer shear, rsh=∂yU∣0−/∂yU∣0r_{\rm sh} = \left.\partial_y U\right|_{0^-}/\left.\partial_y U\right|_0, where ∂yU∣0−\left.\partial_y U\right|_{0^-} is the mean shear approaching y=0y = 0 from the substrate side, for all the cases. None of the length scales L+L^+, g+g^+, L+−g+L^+-g^+, h+h^+, and K+\sqrt{K^+} scale the ratio rshr_{\rm sh}. Instead, rshr_{\rm sh} appears to correlate with the porosity ε\varepsilon, suggesting that a denser substrate with lower ε\varepsilon tends to have a stronger jump of shear, i.e. smaller rshr_{\rm sh}. This is consistent with the above observation that the discontinuity is caused by the shear absorbed at the exposed flat faces of the grains, as their surface area is a larger fraction of the interface plane for lower ε\varepsilon. The ratio rshr_{\rm sh} is roughly linear with ε\varepsilon except for highly porous cases (ε≳0.9\varepsilon\gtrsim0.9), of which rshr_{\rm sh} adjusts to approach the no-jump asymptotic limit, lim⁡ε→1rsh=1\lim_{\varepsilon\to1} r_{\rm sh} = 1.

Image not extracted

Figure 11. Mean velocity profiles near the interface. (a), (b), and (c) are for the cases in figures 5, 6, and 8, respectively, with line styles as in the respective figure.

Image not extracted

Figure 12. Ratio of inner to outer shear across the substrate interface, rshr_{sh}, versus (a) pitch L+L^+, (b) gap size g+g^+, (c) inclusion size ℓ+\ell^+, (d) porosity ε\varepsilon, (e) depth h+h^+, and (f) permeability K+\sqrt{K^+}. Symbols and colours are as in figure 9. The dotted line in (d) is for rsh=0.42ε+0.10r_{sh}=0.42\varepsilon+0.10.

The observations in this subsection suggest that, for both porous and rough substrates, the interfacial shear jump and the slip length are mainly influenced by the porosity and the permeability, respectively, with no significant direct influence of the substrate microscale details or depth. The scaling of slip length with permeability applies also to slip velocities only for small permeabilities, K+≲2\sqrt{K^+} \lesssim2. For larger ones, the slip velocity also depends on the substrate depth, as the shear Reynolds stress at the interface becomes increasingly significant for the deeper substrates. In any event, we note that the values of Us+U_s^+ are significantly smaller than those of ΔU+\Delta U^+, which implies that the slip plays only a small role in determining the drag. The effect of substrate granularity and depth are further investigated in §4.2 and §5.

Drag increase and near-interface flow for deep substrates

Let us now focus on the drag increase of a substrate, given by ΔU+\Delta U^+ as defined in §2.3. Figure 13 portrays ΔU+\Delta U^+ for all the cases simulated versus different substrate parameters. The values of ΔU+\Delta U^+ do not correlate well with any of L+L^+, g+g^+, L+−g+L^+-g^+, ε\varepsilon, and h+h^+, individually. However, ΔU+\Delta U^+ for deep porous substrates shows a good correlation with K+\sqrt{K^+}, extending to shallow porous (Ps) and rough (Ro) substrates for small permeability, K+≲1\sqrt{K^{+}} \lesssim1. Beyond this, ΔU+\Delta U^{+} for shallow and rough substrates is lower and exhibits more scatter than that for deep ones. In general, the discrepancies between deep and shallow substrates are considerably smaller than those between shallow and rough ones, implying an asymptotic behaviour of ΔU+\Delta U^{+} as the depth hh increases, similar to the observations in §3.3. These results suggest that, in essence, permeability alone determines the drag for sufficiently deep porous substrates.

Focusing for now on deep substrates, we observe that the turbulent statistics in general also depend essentially on permeability alone. Figure 14 shows that the substrates with different L+L^{+} and ϵ\epsilon but similar K+\sqrt{K^{+}} have fairly similar mean velocity profiles, turbulent shear stress, and RMS fluctuations in the overlying flow. Differences in the turbulent stress and RMS occur below the interface, where cases with larger L+L^{+} or lower ϵ\epsilon tend to have larger magnitudes. These differences are likely attributable to the different substrate geometries causing different dispersive or grain-coherent stresses, although this would require more in-depth analysis.

Image not extracted

Figure 14. (a-b) Mean velocity profile, (c) Reynolds shear stress, and (d-f) RMS velocity fluctuations for deep porous substrates. Blue to red dotted lines are for substrates with different L+L^{+} and ε\varepsilon but similar K+≈1\sqrt{K^{+}} \approx1, cases Pd-12-50, Pd-24-38, Pd-36-33, and Pd-48-28; dashed for similar K+≈2.5\sqrt{K^{+}} \approx2.5, cases Pd-24-56, Pd-36-50, and Pd-48-44; and solid for similar K+≈6\sqrt{K^{+}} \approx6, Pd-24-75, Pd-36-67, and Pd-48-61. The dash-dotted lines are for smooth-wall data, and the vertical dashed lines mark the location of the free-flow/substrate interface.

Some more details of the structure of turbulence near the interface can be illustrated by instantaneous flow fields and energy spectral densities. The flow fields portrayed in Figure 15 exhibit a signature of the grain-coherent flow with a characteristic length scale L+L^{+}. Superimposed with this signature, we can observe the grain-incoherent features of the background turbulence. Just as the flow statistics in Figure 14, the background turbulence is visually similar for cases with different L+L^{+} and ϵ\epsilon but similar K+\sqrt{K^{+}}, while the grain-coherent flow varies greatly with L+L^{+}. The four cases with K+≈1\sqrt{K^{+}} \approx1 exhibit the typical features of smooth-wall turbulence in the streamwise elongated shapes in u′u', shown in Figure 15(a-Figure 15d), and v′v', shown in Figure 15(k-Figure 15n). The streamwise elongation of these structures is disrupted for K+≈2.5\sqrt{K^{+}} \approx2.5, as shown in Figure 15(e-Figure 15g, Figure 15o-Figure 15q), and entirely lost for K+≈6\sqrt{K^{+}} \approx6, Figure 15(h-Figure 15j,Figure 15r-Figure 15t), for which eddies have an xx-zz aspect ratio closer to unity.

Image not extracted

Figure 15. Instantaneous fields of (a–j) u′u' and (k–t) v′v' at y+≈3y^+ \approx3 for the same deep porous substrates of Figure 14. Columns from left to right correspond to substrates with L+≈12L^+ \approx12, 24, 36, and 48, respectively. (a, b, c, d) and (k, l, m, n), substrates with K+≈1\sqrt{K^+} \approx1; (e, f, g) and (o, p, q), with K+≈2.5\sqrt{K^+} \approx2.5; (h, i, j) and (r, s, t), with K+≈6\sqrt{K^+} \approx6. Colours from dark to clear are for the value range [−2:2][-2:2] relative to the RMS value of the variable at that plane.

The above discussion is also consistent with the statistical information displayed in the spectral density maps of Figure 16. At y+≈3y^{+} \approx3, these maps present two distinct features: a main spectral region attributable to the background turbulence, and smaller lobes centred about the grain-spacing wavelengths, caused by the grain-coherent flow. These lobes extend beyond the mere harmonics of the texture because of the amplitude modulation of the grain-coherent flow by the background turbulence (Abderrahaman-Elena et al. 2019; Habibi Khorasani et al. 2024). The cases with similar K+\sqrt{K^{+}} show good agreement for the background turbulence, but differ in the regions produced by the grain-coherent flow due to their different L+L^{+}. The grain-coherent flow quickly decays away from the interface, as evidenced in the maps at y+≈11y^{+} \approx11. For the background turbulence, as K+\sqrt{K^{+}} increases, the spectral densities become lower in xx-elongated wavelengths but higher in wider zz-wavelengths.

Image not extracted

Figure 16. Pre-multiplied spectra αxαzΦ∗∗\alpha_x\alpha_z\Phi_{**} at (a, c, e) y+≈3y^+ \approx3 and (b, d, f) y+≈11y^+ \approx11 for the same deep porous substrates of Figure 14. (a, b), substrates with K+≈1\sqrt{K^+} \approx1; (c, d), with K+≈2.5\sqrt{K^+} \approx2.5; (e, f), with K+≈6\sqrt{K^+} \approx6. Dashed lines are for L+≈12L^+ \approx12, shaded contours for L+≈24L^+ \approx24, solid lines for L+≈36L^+ \approx36, and dotted lines for L+≈48L^+ \approx48. The contours mark values [0.044:0.044:0.264] relative to the corresponding variance or covariance.

The discussion in this subsection suggests that the effect of deep porous substrates on the overlying turbulence is essentially governed by the permeability, a characteristic not directly associated with the geometric microscale detail of individual grains in a porous medium. The grain-coherent flow near the interface, in turn, manifests the effect of the granularity, but decays quickly away from the substrate, at least for the grain pitches here considered, L+≲50L^{+} \lesssim50 and g+≲30g^{+} \lesssim30.

A unified characterisation from porous to rough substrates

The mean velocity deficit ΔU+\Delta U^{+} in Figure 13(d) correlates well with permeability K+K^{+} only for deep porous substrates, but not for rough surfaces. Unfortunately, a general approach to determine the scaling of drag for rough surfaces remains elusive (see [11] for a review). The substrates in this study have been designed to transition continuously from deep-porous to rough-but-impermeable geometry as the depth hh decreases, while retaining the same grain and interface topology. In this section, our aim is to identify a scaling law for ΔU+\Delta U^+ that captures this continuous transition. With this aim, we now focus on the effect of substrate depth.

Image not extracted

Figure 13. Velocity deficit ΔU+\Delta U^{+} versus (a) pitch L+L^{+}, (b) gap size g+g^{+}, (c) inclusion size ℓ+\ell^{+}, (d) porosity ϵ\epsilon, (e) depth h+h^{+}, and (f) permeability K+\sqrt{K^{+}}. Symbols and colours are as in Figure 9.

An equivalent permeability incorporating the effect of depth

In § 3.3 and § 4.1, we have discussed the relatively small influence of depth on the tangential velocity and interfacial shear and slip, i.e. a porous substrate and a typical rough surface with identical grain geometry have fairly similar subsurface decay of the tangential mean and fluctuating velocities, as shown in Figure 8(b, Figure 8d, Figure 8f), and also similar interfacial slip and shear properties, as shown in Figure 10(f) and 12(d). Therefore, we can infer that the apparent differences in drag increase between porous and rough surfaces in Figure 13(f) principally originate from their differences in interfacial transpiration. This is further supported by Figure 17(a) and (Figure 17b), which show that, for all the substrates studied, the drag increase is highly correlated with the intensity of the interfacial wallnormal velocity fluctuation, rather than with the tangential one, as is the case also for rough surfaces [49, 47, 48].

Image not extracted

Figure 17. Velocity deficit ΔU+\Delta U^+ for all the cases studied as a function of (a) the RMS of the interfacial u′u'; (b) the RMS of the interfacial v′v'; (c) the slip-based equivalent permeability Keqs+K_{eq}^{s+}; and (d) the transpiration-based equivalent permeability Keqt+K_{eq}^{t+} (right). Lines and symbols are as in Figure 9. In (d), Keqt+K_{eq}^{t+} has been calculated for a characteristic near-wall pressure lengthscale λp+=200\lambda_p^+=200, and the error bars represent the range λp+=150\lambda_p^+=150–250250. The values of ΔU+\Delta U^+ versus K+\sqrt{K^+} from Figure 13(f) are displayed in grey for comparison.

The effect of substrate depth on the interfacial slip and transpiration can be observed in the analytical solution of the homogenised Darcy-Brinkman model for the flow within the substrate (Gómez-de-Segura et al. [24]; Gómez-de-Segura & García-Mayoral [23]). Appendix § 10 presents this model for the case of isotropic substrates and discusses the effect of depth as deduced from its solution. The model ultimately results in an admittance, linear relationship, equation (19), which gives the interfacial slip and transpiration velocities in response to the interfacial shear and normal stresses. This relationship involves five coefficients, Lslip\mathcal{L}_{slip}, Kslip\mathcal{K}_{slip}, Ntrsp\mathcal{N}_{trsp}, Ktrsp\mathcal{K}_{trsp}, and LslipL\mathcal{L}_{slip}^{L}, which characterise the effect of the substrate on the overlying free flow. Each coefficient can be written as a product of two parts: a component that is a function of permeability KK, and a dimensionless attenuating function f∗(α,h)f_*(\alpha,h) that depends on the wavenumber α\alpha of the exciting stress and the substrate depth hh.

The function f∗(α,h)f_*(\alpha,h) tends to unity for deep substrates and long exciting waves, and to vanish for shallow substrates or for small waves, making the substrates effectively impermeable and smooth in the latter cases. The first component can be thus identified as the admittance coefficient relating interfacial velocity and stress for large-scale flows over sufficiently deep substrates. One exception to this, though, is the coefficient Ktrsp=KfKt(α,h)\mathcal{K}_{trsp}=Kf_{Kt}(\alpha,h) relating the transpiration velocity v′v' and the pressure fluctuation p′p', which also vanishes for very long waves, and is maximum for an intermediate wavelength comparable to the depth. The discussion in D.3 highlights that, as hh increases, Ktrsp\mathcal{K}_{trsp} reaches its asymptotic value significantly more slowly than the other four coefficients for the typical wavelengths in wall turbulence. This indicates that the principal effect of a finite depth is to decrease Ktrsp\mathcal{K}_{trsp}, i.e. to suppress the pressure-excited transpiration at the interface. Given the strong correlation of this transpiration with ΔU+\Delta U^+ discussed above, we would expect this suppression effect to play a leading role in determining the effect of substrate depth on drag.

The above discussion is based on Darcy-Brinkman solutions, which are after all a mere model for the homogenised subsurface flow – they fail to capture, for instance, the discontinuity in macroscopic velocity shear across the interface, or the direct effect of granularity, which grows with the texture size L+L^+. We thus simply use these solutions to guide us in proposing an empirical ‘equivalent permeability’ KeqtK^t_{eq} to incorporate the effect of substrate depth on transpiration,

Keqt=KfKt(α~,h~),(4)K^t_{eq}=Kf_{Kt}(\widetilde{\alpha},\widetilde{h}), \tag*{(4)}

where the attenuating function fKt(α~,h~)f_{Kt}(\widetilde{\alpha},\widetilde{h}) is calculated from equation (D 11d), and where the dimensionless wavenumber α~\widetilde{\alpha} and depth h~\widetilde{h} are defined by (D 12) and (D 3). The value of α\alpha in (4) should be chosen to represent the characteristic scale of the typical near-wall pressure fluctuations that excite transpiration at the interface. Informed by the spectra in Figure 16, we assume a characteristic wavelength λp=150\lambda_p=150 - 250ν/uτ250\nu/u_\tau and thus α=αp=2π/λp\alpha=\alpha_p=2\pi/\lambda_p. In addition to KeqtK^t_{eq} defined by (4), for comparison we also define another ‘equivalent permeability’ that incorporates the effect of substrate depth on slip,

Keqs=KfLs(α~,h~),(5)K^s_{eq}=Kf_{Ls}(\widetilde{\alpha},\widetilde{h}), \tag*{(5)}

where the attenuating function fLs(α~,h~)f_{Ls}(\widetilde{\alpha},\widetilde{h}) is calculated from equation (D 11a). This KeqsK^s_{eq} gives a measure of the response of the substrate to an overlying shear, and is therefore an inherently interfacial quantity, while KeqtK^t_{eq} gives a measure of the penetration of wall-normal flow due to pressure fluctuations, and therefore accounts for the properties of the substrate not just near the interface but deeper within. An alternative set of admittance relationships to equation (D 9) can be derived from homogenisation ([4, 46]). This gives a set of upscaled coefficients that play the same role of KeqtK^t_{eq}, KeqsK^s_{eq}, and the other coefficients derived from a Darcy-Brinkman model. The scaling of the roughness function with those upscaled coefficients is portrayed in Appendix § 11, and the collapse is generally not as good as the one discussed below with Keqt\sqrt{K^t_{eq}}. In the authors’ view, the reason is likely the lack of separation of scales between turbulence and surface texture, which is a central assumption in homogenisation.

The values of ΔU+\Delta U^+ against the newly defined Keqs+\sqrt{K^{s+}_{eq}} and Keqt+\sqrt{K^{t+}_{eq}} are portrayed in 17 for all porous and rough cases. Figure 17(c) shows that the values of Keqs+\sqrt{K^{s+}_{eq}} are close to those of K+\sqrt{K^+}. This is consistent with the observations in § 4.1 and the analysis in §D.3, which suggests that the effect of depth on interfacial slip is small. As a result, Keqs+\sqrt{K^{s+}_{eq}} and K+\sqrt{K^+} produce a similarly poor collapse for ΔU+\Delta U^+ across all the substrates studied. Meanwhile, the values of ΔU+\Delta U^+ across all substrates collapse well with Keqt+\sqrt{K^{t+}_{eq}}, as shown in Figure 17(d). For rough surfaces, Keqt+\sqrt{K^{t+}_{eq}} is significantly smaller than the original K+\sqrt{K^+}, indicating that their small depths suppress significantly the interfacial transpiration. For deep porous substrates, such suppression tends to vanish and the differences between Keqt+\sqrt{K_{eq}^{t+}} and K+\sqrt{K^+} are small.

The transpiration-based equivalent permeability, Keqt+\sqrt{K_{eq}^{t+}}, also characterises well the turbulence for substrates with different depths. In Figure 18 to Figure 20, we consider three groups of substrates, each including a deep porous (Pd), a shallow porous (Ps), and a rough (Ro) case with different depth h/Dh/D but similar Keqt+\sqrt{K_{eq}^{t+}}. Figure 18 shows that substrates with similar Keqt+\sqrt{K_{eq}^{t+}} also share similarity in their free-flow mean velocity profiles, RMS velocity fluctuations and Reynolds stress, in spite of their differences in subsurface flow, to be expected given the different depths and granularities. This similarity can also be observed in instantaneous realisations for the flow just above the interface, as illustrated in Figure 19. The flows with similar Keqt+\sqrt{K_{eq}^{t+}} exhibit similar features for the background turbulence, but differences in the grain-coherent flow, directly attributable to the different grain topologies. The same effects can be observed in the spectral density maps of Figure 20, which quantify statistically the similarities in the fluctuations at different lengthscales for the background turbulence, and also display the different intensities and lengthscales of the grain-coherent flow for different topologies.

Image not extracted

Figure 18. (a–b) Mean velocity profile, (c) Reynolds shear stress, and (d–f) RMS velocity fluctuations for cases with similar Keqt+\sqrt{K_{eq}^{t+}}. Blue, yellow, and red are for deep porous (Pd) substrates (h/D≥5h/D \ge5), shallow porous (Ps) substrates (h/D=2h/D = 2), and rough surfaces (h/D=1h/D = 1), respectively. Dotted lines are for the cases with different L+L^+, ϵ\epsilon, and h/Dh/D but similar Keqt+≈1\sqrt{K_{eq}^{t+}} \approx1: Pd-48-28 (K+=0.98\sqrt{K^+} = 0.98, Keqt+=0.97\sqrt{K_{eq}^{t+}} = 0.97), Ps-36-33 (K+=1.10\sqrt{K^+} = 1.10, Keqt+=0.96\sqrt{K_{eq}^{t+}} = 0.96), and Ro-24-50 (K+=1.82\sqrt{K^+} = 1.82, Keqt+=0.94\sqrt{K_{eq}^{t+}} = 0.94). Dashed lines are for the cases with similar Keqt+≈1.7\sqrt{K_{eq}^{t+}} \approx1.7: Pd-24-50 (K+=1.82\sqrt{K^+} = 1.82, Keqt+=1.75\sqrt{K_{eq}^{t+}} = 1.75), Ps-48-38 (K+=1.88\sqrt{K^+} = 1.88, Keqt+=1.66\sqrt{K_{eq}^{t+}} = 1.66), and Ro-48-44 (K+=2.66\sqrt{K^+} = 2.66, Keqt+=1.70\sqrt{K_{eq}^{t+}} = 1.70). Solid lines are for the cases with similar Keqt+≈3.5\sqrt{K_{eq}^{t+}} \approx3.5: Pd-48-50 (K+=3.64\sqrt{K^+} = 3.64, Keqt+=3.44\sqrt{K_{eq}^{t+}} = 3.44), Ps-24-75 (K+=5.80\sqrt{K^+} = 5.80, Keqt+=3.68\sqrt{K_{eq}^{t+}} = 3.68), and Ro-48-61 (K+=6.28\sqrt{K^+} = 6.28, Keqt+=3.55\sqrt{K_{eq}^{t+}} = 3.55). The dash-dotted lines are for smooth-wall data, and the vertical dashed lines mark the location of the free-flow/substrate interface.

Image not extracted

Figure 19. Instantaneous fields of (a-i) u′u' and (j-r) v′v' at y+≈3y^+ \approx3 for the same substrates of figure 18. Columns from left to right correspond to deep porous (Pd), shallow porous (Ps), and rough (Ro) substrates, respectively. (a, b, c) and (j, k, l), substrates with Keqt+≈1\sqrt{K_{eq}^{t+}} \approx1; (d, e, f) and (m, n, o), with Keqt+≈1.7\sqrt{K_{eq}^{t+}} \approx1.7; (g, h, i) and (p, q, r), with Keqt+≈3.5\sqrt{K_{eq}^{t+}} \approx3.5. Colours from dark to clear are for the value range [−2:2][-2:2] relative to the RMS value of the variable at that plane.

Image not extracted

Figure 20. Pre-multiplied spectra αxαzΦ∗∗\alpha_x\alpha_z\Phi_{**} at (a, c, e) y+≈3y^+ \approx3 and (b, d, f) y+≈11y^+ \approx11 for the same deep porous substrates of Figure 18. (a, b), substrates with Keqt+≈1\sqrt{K_{eq}^{t+}} \approx1; (c, d), with Keqt+≈1.7\sqrt{K_{eq}^{t+}} \approx1.7; (e, f), with Keqt+≈3.5\sqrt{K_{eq}^{t+}} \approx3.5. Shaded contours are for deep porous (Pd) substrates, yellow dotted lines for shallow porous (Ps) substrates, and red solid lines for rough (Ro) surfaces. The contours mark values [0.044:0.044:0.264] relative to the corresponding variance or covariance.

The collapse observed for Keqt+\sqrt{K_{eq}^{t+}}, and the comparatively lack of collapse observed for Keqs+\sqrt{K_{eq}^{s+}}, suggest that effect of the substrate on the overlying turbulence is mainly governed by the wall-normal transpiration triggered by overlying pressure fluctuations. The effect of the interfacial slip, which is characterised by Keqt+\sqrt{K_{eq}^{t+}}, plays in contrast a minor role. The good collapse with Keqt+\sqrt{K_{eq}^{t+}} may also suggest that the Darcy-Brinkman model captures reasonably well the fluctuating transpiration induced by overlying pressure waves, even if it fails to capture other features of the flow such as the tangential shear and velocity near the interface ([54]). It is somewhat surprising that this is the case even for substrates with one or two layers of grains, for which the lack of sufficient depth would in principle preclude the type of homogenisation in yy implicit

in Darcy–Brinkman. An answer to this is however out of the scope of the present paper, where we limit ourselves, as stated above, to using the Darcy–Brinkman solutions to guide us in proposing an otherwise empirical coefficient to collapse the DNS results. We note that the present analysis extends to substrates that exhibit a rough interface due to their intrinsic granularity, and leaves out surfaces that are rough but would not have an obvious permeable substrate counterpart, e.g. wavy walls or smoothly curved surfaces. For those surfaces, we could expect the slip dynamics to play a more dominant role, as occurs also for drag-reducing textures [27, 25]. For porous substrates like the present ones, the interfacial roughness can potentially be eliminated by making the surface artificially flat, which would partially suppress the resulting increase in drag (see Kim et al. 2020, or Figure 1 and the corresponding discussion in Rosti et al. 2015). It is also possible to generate an interfacial roughness completely unrelated to the substrate topology and decouple their effects, as in Wangsawijaya et al. [67]. Both cases are beyond the scope of this study. Likewise, we have also only considered isotropic substrates. Anisotropic substrates such as those of Habibi Khorasani et al. [26] would be beyond the present scope. In the latter, the permeability is tensorial, and the expression in 5.1 for an effective Keq+\sqrt{K_{\mathrm{eq}}^{+}} would not be valid. It is possible, however, to find an analogous solution for the anisotropic Darcy–Brinkman equations. We have done so in the past for the case of principal directions aligned with xx, yy and zz [24, 23, 58]; the solution presented in Appendix § 10 is actually a particularisation of such solutions. A simple explicit expression is not possible in that case, but from our previous work we would expect that the role of KK in equation 5.1 was played by KyK_y for streamwise preferential substrates [23], and by KxKy\sqrt{K_xK_y} for wall normal preferential ones [58]. Here, KxK_x and KyK_y are the streamwise and the wall-normal permeability, respectively.

The reader may wonder if the present findings apply only to the regular staggered-cube and mesh topologies for which DNSs have been conducted in this paper, with relatively small grain pitch L+≲50L^{+} \lesssim50. In Appendix F we compare our results with data compiled for different substrate layouts and from previous studies, namely the randomly packed spheres of Zippe & Graf [72] and Karra et al. [31], the staggered cubes of Kuwata & Suga [35], the collocated spheres of Kim et al. [33], the reticulated foams of Esteban et al. [15], and the isotropic mesh lattices of Habibi Khorasani et al. [26]. The substrates in those studies have generally larger grain and therefore larger K+K^{+}, and data for the roughness function was sometimes obtained indirectly, especially for experimental works, but the agreement is generally good. We note that some of those studies were for irregular substrate topologies. We expect our conclusions to hold beyond the homogeneous substrates of our DNSs, so long as the heterogeneity occurs on a scale small enough not to interact with individual turbulent eddies. If heterogeneities were present over lengthscales of ∼100\sim100 viscous units or larger, they would need to be accounted for and would probably require a unique ad hoc analysis for each instance.

A conceptual regime diagram for finite-depth substrates

The above results and discussion suggest that, for any given substrate topology, its effect on the overlying turbulence can be essentially characterised by Keq+K_{\mathrm{eq}}^{+} alone. This coefficient is proportional to the substrate permeability K+K^{+}, which encodes the relevant information on the grain topology, but also depends on the substrate depth through fKtf_{Kt}. Based on this, we propose a regime diagram to conceptually illustrate the relationship between porous and rough surfaces for a given grain topology. As an example, in figure 21 we consider our staggered-cube topology with gap-to-pitch ratio g/L=1/2g/L = 1/2, corresponding to a constant K≈0.076L\sqrt{K} \approx0.076L. For other topologies, the regime diagram would be qualitatively similar.

Image not extracted

Figure 21. Substrate regime diagram for a staggered-cube topology with g/L=1/2g/L = 1/2. The shaded contours are for the transpiration-based equivalent permeability Keqt+\sqrt{K_{\mathrm{eq}}^{t+}}, taken as a surrogate for the drag increase. The dashed and solid contour lines are for fLsf_{Ls} and fKtf_{Kt}, respectively, for values [0.1:0.1:0.9] from red to blue. The results have been obtained for a characteristic wavelength for the overlying turbulence λp+=200\lambda_p^+ = 200. The dotted straight line represents impermeable rough surfaces with h=L/2h = L/2. The dashed lines represent boundaries between regimes. The markers represent the present twelve DNSs with this topology, g/L=1/2g/L=1/2.

The diagram portrays, as a function of K+\sqrt{K^+} and h+h^+, the equivalent permeability Keqt+K_{\mathrm{eq}}^{t+}, taken as a surrogate for the roughness function ΔU+\Delta U^+. To illustrate the effect of substrate depth on slip and transpiration, we also portray the corresponding values of the attenuation coefficients fLsf_{Ls} and fKtf_{Kt}. To construct the diagram, these have been calculated for a characteristic exciting wavelength λp+≈200\lambda_p^+ \approx200, as discussed in § 5.1. Note that this would be valid so long as the signature of the overlying pressure fluctuations at the interface plane retained roughly its characteristic size for smooth-wall flows. This is the case for the texture sizes studied here, L+≲50L^+ \lesssim50, as shown in Figure 16 and Figure 20, but would eventually break down for larger sizes. Therefore, in the following, we only consider substrates with K+≲10\sqrt{K^+} \lesssim10 or L+≲100L^+ \lesssim100, which are the scope of the present study and for which we would expect the diagram to be representative.

In the diagram, a dotted line marks rough surfaces with cubic roughness elements of pitch LL and height h=L/2h = L/2, which includes our four rough DNSs with the topology under consideration, g/L=1/2g/L = 1/2. Following this line for increasing L+L^+ is equivalent to increasing the Reynolds number for a fixed surface, thus increasing its dimensions in viscous units, and its roughness function accordingly, as if following a Moody chart [45]. Substrates to the right of this line would also be rough, featuring roughness elements with the same pitch and element width and gap but smaller depth, h<L/2h < L/2. The isocontours for fLsf_{Ls} indicate that, in terms of depth, the slip properties of the surfaces with h=L/2h = L/2 would already be maxed out, i.e. beyond the line labeled ‘slip saturated’ – no further increases would take place when increasing hh further. Meanwhile, the shallower rough substrates with h<L/2h < L/2 would experience comparatively reduced slip.

As discussed in §5.1, however, the effect of slip through Keqs+K_{eq}^{s+} is secondary, and the drag is mainly governed by Keqt+K_{eq}^{t+}, for which the relevant attenuating function is the transpiration one, fKtf_{Kt}. The latter is a continuous, smooth function of the substrate depth, which tends to zero for vanishing depth and to unity for deep substrates. Its isocontours show that, while fLsf_{Ls} saturates at much shallower depths hh, i.e. already for simply rough, impermeable surfaces, fKtf_{Kt} saturates only for deeper substrates. Its isocontours indicate that, for the present textures with K+≲10\sqrt{K^+} \lesssim10, the saturation depth is essentially independent of K+K^+ or L+L^+, and is roughly h+≈50h^+ \approx50. This can be traced to the fact that hh is scaled by λp\lambda_p in the expression for fKtf_{Kt}, and that λp+≈200\lambda_p^+ \approx200 independently of the substrate, as discussed above. Whether a substrate is shallow or sufficiently deep is governed by whether pressure fluctuations of this wavelength would penetrate deeper in the absence of an impermeable floor at a finite depth y+=−h+y^+ = -h^+. Perhaps counterintuitively, this depends on the depth in viscous units, rather than in permeability lengths or number of inclusion layers.

From this, we can distinguish a series of regions in the diagram which are characterised by different flow regimes. For substrates deeper than h+≈50h^+ \approx50, the flow would perceive the substrate as deep enough to exhibit its permeable character fully. We refer to these as ‘sufficiently deep porous substrates’, for which any further increase in depth would not result in any significant changes to the flow. For substrates shallower than this but deeper than the rough surfaces with cubic inclusions, h=L/2h = L/2, the substrate would exhibit an intermediate character, where the transpiration is partially suppressed compared to a deeper substrate with the same K+K^+. We refer to these as ‘shallow porous substrates’. Substrates with h=L/2h = L/2 would be so shallow that they are conventional rough, impermeable surfaces. Substrates even shallower than h=L/2h = L/2 would likewise be rough, but with shorter-height roughness elements, further suppression of transpiration, and additionally suppression of slip.

Conclusions

This study has focused on understanding the effect of different characteristics of a porous substrate on the properties of the overlying turbulent flow. Specifically, we have aimed to investigate the effects of three characteristics: i) the permeability of a porous medium, characterised by the bulk permeability KK, ii) the granularity of the porous medium, whose length scale is characterised by the grain pitch LL, and iii) the depth of the substrate, hh. To this object, we have systematically explored the parameter space of KK, LL, and hh for porous substrates using DNSs. We have used staggered-cube arrays with various porosities ϵ=0.23\epsilon= 0.23 - 0.97 and permeabilities K/L=0.013\sqrt{K}/L = 0.013 - 0.243, and studied substrate depths ranging from h=1D=L/2h = 1D = L/2 to h≥5D=5L/2h \geq5D = 5L/2, corresponding to the transition from typical impermeable rough surfaces to deep porous substrates. The values of grain pitch we have considered are in the range 10≲L+≲5010 \lesssim L^+ \lesssim50. For this range, the overlying turbulence soon deviates from smooth-wall-like behaviour, but the grains remain small enough that the grain-coherent flow is not sufficiently intense to directly interact with the background, grain-incoherent turbulence, other than perhaps for the largest L+≈50L^+ \approx50.

Through the discussion on the scaling of flow properties with substrate parameters, two major insights have been provided on the effect of the above three. First, the permeability of a porous medium has significantly greater relevance than its granularity to most of the main properties of the overlying turbulence. For all substrates with h≥1Dh \geq1D, the mean-flow slip length ℓU+\ell_U^+ is essentially proportional to K+\sqrt{K^+}, and the inner/outer shear ratio across the interface rshr_{sh} correlates well with ϵ\epsilon. For deep substrates with h≥5Dh \geq5D, the mean velocity deficit ΔU+\Delta U^{+} and the statistics and energy spectra of the background turbulence are essentially determined by K+\sqrt{K^{+}} alone. In contrast, the effect of L+L^{+} on the overlying turbulence is essentially not significant, and can only be observed in flow realisations and spectral density maps very near the interface. This implies that, at least for grain pitch values L+≲50L^{+} \lesssim50, a porous substrate can be reasonably approximated as a continuum represented only by macroscale parameters like KK, hh, and ε\varepsilon.

Second, the principal mechanism that distinguishes porous substrates from surfaces with analogous topology that are rough, but are otherwise impermeable, is the effect of the substrate depth on the interfacial transpiration excited by the overlying pressure fluctuations. We propose an empirical ‘equivalent permeability’, KeqtK_{\mathrm{eq}}^{t}, with some theoretical support based on Darcy-Brinkman models, that incorporates this effect of depth on transpiration. For substrates with different depths, the values of Keqt+\sqrt{K_{\mathrm{eq}}^{t+}} alone collapse ΔU+\Delta U^{+} well, and different substrates with similar Keqt+\sqrt{K_{\mathrm{eq}}^{t+}} exhibit similar properties for the overlying turbulence in general. A conceptual h+h^{+}–K+\sqrt{K^{+}} diagram illustrates the regime transition between sufficiently deep porous substrates, with h+≳50h^{+} \gtrsim50, and typical impermeable rough surfaces with h=O(L)h = O(L).

The scope of this study is substrates composed of relatively small grains, L+≲50L^{+} \lesssim50. In the fully-rough regime, where L+≳70L^{+} \gtrsim70, the flow coherent with individual grains could be energetic enough to generate grain-coherent eddies that interacted strongly with and modify directly the overlying turbulence. In that situation, the effects of substrate permeability and granularity may be intrinsically indistinguishable. Understanding the relationship between the substrate parameters and the overlying turbulence in the fully rough regime may thus require a different framework. The present work covers sizes just up to the onset of that regime.

Acknowledgements. The authors thank Dr. Garazi Gómez-de-Segura for her advice on flow physics and simulations.

Funding. This work is supported by the UK Engineering and Physical Sciences Research Council (EPSRC) under grant EP/S013083/1, and by the Air Force Office of Scientific Research – European Office of Aerospace Research and Development (AFOSR–EOARD), under grant FA8655-22-1-7062. Computational resources were provided by the University of Cambridge Research Computing Service under EPSRC Tier-2 grant EP/P020259/1 (projects cs066 and cs155), and by the UK ’ARCHER2’ system under PRACE project pru1702 and EPSRC project e776.

For the purpose of open access, the authors have applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising from this submission.

Declaration of interests. The authors report no conflict of interest.

Author ORCIDs.

R. García-Mayoral, https://orcid.org/0000-0001-5572-2607;

Z. Hao, https://orcid.org/0000-0003-2199-6938.

Appendix A

Here we present evidence supporting the use of idealised substrates made up of ‘frozen-suspensions’ of inclusions to study the effect of substrates with higher permeability. These would not be realisable as solid permeable surfaces in the real world. Such is the case of our staggered-cube arrangements with gap size g>L/2g > L/2. The reader may wonder if this would not give rise to fundamental differences in the interstitial flow, compared to interconnected topologies. Figure 22 shows that this is not the case. The figure portrays the detail of the microscale flow as the gap size changes and the topology shifts from the substrate being just interconnected to just disconnected. The figure shows that the flow evolves smoothly from one case to the other, with no intrinsic difference or abrupt changes in its structure.

Image not extracted

Figure 22. Detail of the interstitial microscale flow in substrates made up of interconnected and disconnected staggered cubes. Results are shown for the velocity magnitude, normalised by the Darcy velocity, in xx-zz sections; through the middle of a row of cubes in (a, c, e), and through a plane at the half-height between centres of consecutive rows in (b, d, f). The location of the cubes is marked in white, with dashed lines for cubes not intersected by the section. In (a, c, e), streamlines of the locally 2D flow are also portrayed in black. (a, b), overlapping cubes with g/L=7/16g/L = 7/16; (c, d) just-touching cubes with g/L=1/2g/L = 1/2; (e, f), suspended cubes with g/L=9/16g/L = 9/16. For each panel pair, results on the left are for Stokes-flow simulations, and results on the right for ensemble averages from DNSs with L+≈24L^+ \approx24, when available (cases Pd-24-50 and Pd-24-56).

The results presented in this paper suggest that the overlying turbulence depends essentially on the macroscopic properties of substrates, and not on their microscopic detail, regardless of whether substrate inclusions are interconnected or not. Figure 23 shows the same data of Figure 17(d), but highlighting which substrates are interconnected and which are frozen suspensions. Data from both groups spans with substantial overlap the full range of permeabilities studied, showing that there is indeed no essential difference.

Image not extracted

Figure 23. Results from figure 17(d) for the roughness function ΔU+\Delta U^+ as a function of the transpiration equivalent permeability Keqt+K_{\mathrm{eq}}^{t+}. Black symbols, interconnected substrates of staggered cubes or mesh-like lattices. White symbols, frozen suspensions of staggered cubes.

Integral quantities like the roughness function can conceal more subtle differences in the flow, but the similarity across the two topology groups extends also to more detailed flow properties. Figure 24 compares turbulent statistics for interconnected and frozen-suspension substrates for similar values of Keqt+K_{\mathrm{eq}}^{t+}, and shows again that there is good agreement in the results across different substrates, regardless of their microstructure topology, for similar values of the macroscopic effective permeability.

Image not extracted

Figure 24. Comparison of turbulent statistics for interconnected and frozen-suspension substrate topologies. (aa-bb) Mean velocity profile, (cc) Reynolds shear stress, and (dd-ff) RMS velocity fluctuations. Dashed lines, substrates with Keqt+≈2\sqrt{K_{\mathrm{eq}}^{t+}} \approx2; blue, Pd-12-67 (K+=2.03\sqrt{K^+}=2.03, Keqt+=1.86\sqrt{K_{\mathrm{eq}}^{t+}}=1.86); yellow, Ps-36-50 (K+=2.73\sqrt{K^+}=2.73, Keqt+=2.25\sqrt{K_{\mathrm{eq}}^{t+}}=2.25); red, MRo-48-56 (K+=3.30\sqrt{K^+}=3.30, Keqt+=1.98\sqrt{K_{\mathrm{eq}}^{t+}}=1.98). Solid lines, substrates with Keqt+≈5\sqrt{K_{\mathrm{eq}}^{t+}} \approx5; blue, Pd-24-75 (K+=5.80\sqrt{K^+}=5.80, Keqt+=5.30\sqrt{K_{\mathrm{eq}}^{t+}}=5.30); magenta, MPd-36-78 (K+=5.34\sqrt{K^+}=5.34, Keqt+=5.11\sqrt{K_{\mathrm{eq}}^{t+}}=5.11); purple, MPd-48-72 (K+=5.94\sqrt{K^+}=5.94, Keqt+=5.54\sqrt{K_{\mathrm{eq}}^{t+}}=5.54).

Appendix B

Section 2.2 details the grid resolution used for the various DNSs conducted in this work. The grid convergence study in Sharma & García-Mayoral (2020b) concluded that, for the present DNS code and discretisation, 13-15 fluid points were sufficient to resolve the gaps in-between obstacles through which the fluid would flow, and that when a lower resolution of 7-8 fluid points per gap was used, errors of up to ∼20%\sim20\% were observed in the fluctuating velocity within the substrate, mainly in the wall-normal velocity, although the error in the fluctuating flow above was reduced to ∼4%\sim4\%. Sharma & García-Mayoral (2020b) argued that the latter marginal resolutions were acceptable for small gaps g+≲5g^+ \lesssim5, where the flow would be predominantly viscous and instantaneously unidirectional. The present DNSs follow these general criteria, with 13 points or more per gap except for simulations Pd-12-33, Ro/Pd-24-25, Pd-36-22, Ro/Pd-48-22, Pd-48-25, and Ro/Pd-48-28, with 9-11 points per gap. Of these, the most critical case is Pd-48-25, since it has the largest gap g+≈12g^+ \approx12 with the lowest resolution, 9 points per gap and Δz+≈1.5\Delta z^+ \approx1.5.

To quantify the effect of this marginal resolution, we have run an additional simulation doubling the resolution in xx and zz, i.e. 64 points per pitch LL and 17 points per gap.

The flow statistics from the DNSs with the coarser and finer resolutions are compared in Figure 25, and are consistent with the resolution analysis of Sharma & García-Mayoral (2020b). The results are very close in the free flow region, with somewhat larger deviations for the mean velocity U+U^+ and the RMS velocity fluctuations urms+u_{\mathrm{rms}}^+, vrms+v_{\mathrm{rms}}^+, and wrms+w_{\mathrm{rms}}^+ below the interface, with the lower resolution resulting in slightly underpredicted values. Among them, the differences for U+U^+ and vrms+v_{\mathrm{rms}}^+ seem more apparent than the others, and can be directly linked to the difference in a posteriori K+\sqrt{K^+} for the two simulations, of order ∼8%\sim8\%.

Image not extracted

Figure 25. (a-b) Mean velocity profile, (c) Reynolds shear stress, and (d-f) RMS velocity fluctuations for case Pd-48-25 with different resolutions. Blue and red lines are for the coarser and finer resolutions, 32 and 64 points per pitch LL, respectively. The dashed lines mark the location of the free-flow/substrate interface.

In order to avoid grid-related discrepancies between the a posteriori and a priori values of K+\sqrt{K^+}, as mentioned in § 2.2 we have computed both with the same resolution. This is the same strategy we followed in [21], [17] and [57] to compute macroscopic surface parameters such as protrusion heights, slip lengths and permeabilities consistent with the DNSs. The deviation from the grid-converged values of K+\sqrt{K^+} is in any event small, with underpredictions ≲10%\lesssim10\% for the topologies with narrowest gaps, g≤L/4g \le L/4, and ≲5%\lesssim5\% for the topologies with g≥L/3g \ge L/3. These deviations are estimated from the values kindly provided by a reviewer and the grid-convergence results portrayed in Figure 2. For the grid-convergence study, we have computed the a priori K+\sqrt{K^+} for the staggered-cube configuration with g=L/4g = L/4, the topology of Pd-48-25, for our baseline resolution of 36 points per LL plus for 64, 96, 144 and 192 points per LL. The results are shown in figure 26, and are consistent with a ground-truth value K+≈0.0177\sqrt{K^+} \approx0.0177, to which they tend asymptotically – we note that this entails an overprediction of ∼2%\sim2\% in the reviewer’s value of 0.0181. Relative to the asymptotic value, the values of K\sqrt{K} corresponding to 32 and 64 points per pitch have errors of ∼9%\sim9\% and ∼5%\sim5\%, respectively.

Figure 26. *A priori values of K\sqrt{K} for the staggered-cube configuration with g/L=1/4g/L = 1/4, obtained from Stokes-flow simulations with different grid resolutions. The blue and red makers represent the resolutions compared in DNSs for case Pd-48-25. The dashed line marks the asymptotic value K+≈0.0177\sqrt{K^+} \approx0.0177, used in (b) to estimate the relative error.

Effect of the Reynolds number

Most of the analysis on the scaling of flow properties in this paper is for the data nondimensionalised in wall units. This is based on the assumption that these flow properties are dominated by the dynamics of near-wall turbulence in the buffer layer [22]. In this section, to check the dependence of the analysis on Reynolds number, we compare some results of simulations run at different Reτ≈180Re_\tau\approx180, 360, and 550, while matching the substrate dimensions in wall units. Five sets of cases are considered, in each set the cases sharing the same L+L^+, g/Lg/L, and h/Dh/D: Pd-24-50(-HR), Pd-36-50(-HR), Pd-48-38(-HR), Pd-48-50(-HHR), and Pd-48-62(-HR,-HHR). Near the interface, the visual similarity between the cases in a set can be observed in the instantaneous flow fields in figure 27.

Figure 28 shows the profiles of the mean velocity U+U^+ and the RMS fluctuations of the wall-normal velocity vrms+v_{\mathrm{rms}}^+ for the five sets of cases, together with the corresponding smooth-wall results for reference, with figure 29 showing the corresponding spectral energy densities. For each set, above the interface, the profiles of U+U^+ for substrates show good agreement, with small discrepancies in the wake region similar to those observed for smooth walls. Such an agreement extends below the interface until the Darcy regions are reached, below where the U+U^+ ratio is proportional to 1/Reτ1/Re_\tau. This is a result of the mean pressure gradient being different in wall units. The cases in a set have similar profiles of vrms+v_{\mathrm{rms}}^+ for y+≲15y^+ \lesssim15, above which the values gradually approach the corresponding smooth-wall results at each given ReτRe_\tau. Although not shown, the profiles of the RMS fluctuations urms+u_{\mathrm{rms}}^+ and wrms+w_{\mathrm{rms}}^+ and the Reynolds shear stress u+′v+′u^{+\prime}v^{+\prime} exhibit similar trends to the ones of vrms+v_{\mathrm{rms}}^+, as in [56 b]. The spectral energy densities of the velocity fluctuations and the Reynolds shear stress show the same agreement in Figure 29, with further information at larger wavelengths available for the higher-ReτRe_\tau cases due to their larger domain in viscous units. In summary, the comparison in this

Mean velocity profile and wall-normal velocity RMS fluctuation at different $Re_{\tau}$

Figure 28. Mean velocity profile and wall-normal velocity RMS fluctuation at different ReτRe_{\tau}. From top to bottom, cases Pd-24-50(-HR), Pd-36-50(-HR), Pd-48-38(-HR), Pd-48-50(-HHR), and Pd-48-62(-HR,-HHR). Blue lines are for Reτ≈180Re_{\tau} \approx180, magenta lines for Reτ≈360Re_{\tau} \approx360, and purple lines for Reτ≈550Re_{\tau} \approx550.

Figure 29. Pre-multiplied spectra αxαzΦ∗∗\alpha_x\alpha_z\Phi_{**} at y+≈3y^+ \approx3 for the cases with different ReτRe_\tau. Blue shades are for Reτ≈180Re_\tau\approx180, magenta lines for Reτ≈360Re_\tau\approx360, and purple lines for Reτ≈550Re_\tau\approx550. From top to bottom: Pd-24-50(-HR), Pd-36-50(-HR), Pd-48-38(-HR), Pd-48-50(-HHR), and Pd-48-62(-HR,-HHR). Contours represent six equidistant levels (0.044, 0.088, 0.132, 0.176, 0.220, and 0.264) relative to the corresponding variance or covariance.

Instantaneous fields $u'$ (left), $v'$ (middle), and $p'$ (right)

Figure 27. Instantaneous fields u′u' (left), v′v' (middle), and p′p' (right) at y+≈3y^+ \approx3 for cases with different ReτRe_\tau. From top to bottom: Pd-24-50(-HR), Pd-36-50(-HR), Pd-48-38(-HR), Pd-48-50(-HHR), and Pd-48-62(-HR,-HHR). The subplot at the bottom-left corner of each panel is for the case with lower ReτRe_\tau than the corresponding case with higher ReτRe_\tau in the remainder of the panel. Colour range black-red-yellow-white corresponds to [−2:2][-2:2] times the RMS of the variable on the plane.

section indicates that the scaling of flow properties presented in this study is essentially Reynolds number-independent.

Appendix D: Analytical solutions of the Darcy-Brinkman equation for isotropic permeable substrates

As in Taylor [64], we interpret the Darcy-Brinkman equation as a constitutive model for the volume-averaged (i.e. homogenised) flow in permeable media, rather than an empirical remedy to the Darcy equation near interfaces to satisfy boundary conditions. Compared to the classic Darcy model (Darcy [13]), which accounts for the effect of pressure gradients, the Darcy-Brinkman model (Brinkman [8]) also incorporates the stresses that are induced by gradients of the volume-averaged velocity over length scales larger than the pore size. A formal derivation of the Darcy-Brinkman equation for the case of isotropic highly porous media can be found in Sanchez-Palencia [54] and Lévy [42]. Gómez-de-Segura et al. [24] and Gómez-de-Segura & García-Mayoral [23] derived analytical solutions to the Darcy-Brinkman equation for permeable substrates with anisotropic permeability tensors whose principal orientations coincide with xx, yy, and zz. For isotropic substrates, these solutions are considerably simplified and provide direct insight into how the substrate characteristics can affect the overlying flow. This section presents these analytical solutions with specific emphasis on their particularisation at the interface, which then serves as a boundary condition for the overlying free flow. This is complemented by a brief discussion on the role of substrate depth in these solutions.

The Darcy-Brinkman equation in an isotropic permeable substrate with permeability KK and depth hh reads

∇p=−νK−1u+νeff∇2u,y∈[−h,0].(6)\nabla p=-\nu K^{-1}\mathbf{u}+\nu_{\mathrm{eff}}\nabla^2\mathbf{u},\qquad y\in[-h,0]. \tag*{(6)}

where pp is the fluid-volume-average (‘intrinsic average’) pressure and u\mathbf{u} is the total-volume-average (‘superficial average’) velocity. νeff=ν/rν\nu_{\mathrm{eff}}=\nu/r_\nu is an effective viscosity perceived by the macroscale shear of the substrate flow, where the viscosity ratio rνr_\nu is usually regarded as a constant.

Solution for the mean flow

Let us first consider the mean flow field Φ(y)\Phi(y) averaged in xx and zz. Assume a constant pressure gradient ∂xP\partial_xP, a mean velocity shear at the interface ∂yU∣0\left.\partial_yU\right|_0 (approaching from below), and a no-slip condition on the floor U∣−h=0\left.U\right|_{-h}=0. Then the mean-flow solution to equation (D 1) can be directly obtained as the sum of a pressure-driven and a shear-driven part,

U(y~)=UPr(y~)+USh(y~),U(\widetilde{y})=U_{\mathrm{Pr}}(\widetilde{y})+U_{\mathrm{Sh}}(\widetilde{y}),
UPr(y~)=(−∂xP) ν−1K(1−cosh⁡(y~)cosh⁡(h~)),(7)U_{\mathrm{Pr}}(\widetilde{y})=(-\partial_xP)\,\nu^{-1}K\left(1-\frac{\cosh(\widetilde{y})}{\cosh(\widetilde{h})}\right), \tag*{(7)}
USh(y~)=(∂yU∣0)Krν(sinh⁡(y~+h~)cosh⁡(h~)),(8)U_{\mathrm{Sh}}(\widetilde{y})=\left(\left.\partial_yU\right|_0\right)\sqrt{\frac{K}{r_\nu}}\left(\frac{\sinh(\widetilde{y}+\widetilde{h})}{\cosh(\widetilde{h})}\right), \tag*{(8)}

where the dimensionless coordinate y~\widetilde{y} and depth h~\widetilde{h} are

y~=y/K/rν,h~=h/K/rν.(9)\widetilde{y}=y/\sqrt{K/r_\nu},\qquad\widetilde{h}=h/\sqrt{K/r_\nu}. \tag*{(9)}

For the DNS cases in this study, UPrU_{\mathrm{Pr}} is significantly smaller than UShU_{\mathrm{Sh}}, so U≈UShU\approx U_{\mathrm{Sh}}. Also, if the depth hh is large enough, h~≫1\widetilde{h}\gg1, we can approximate (D 2b) by an exponential decay

USh(y~)∝ey~=ey/K/rν(10)U_{\mathrm{Sh}}(\widetilde{y})\propto e^{\widetilde{y}}=e^{y/\sqrt{K/r_\nu}} \tag*{(10)}

in the region not too close to the floor. The decaying exponent in equation (D 4) can thus be used to estimate the viscosity ratio rνr_\nu. In this paper, rνr_\nu is estimated by fitting the mean velocity profile from DNSs between y=0y = 0 and the location where U≈0.1U∣0U \approx0.1U|_0, while ensuring (D 2) roughly holds. The resulting values of rνr_\nu in this study are in the range 0.05–0.98, and the corresponding Darcy-Brinkman solutions are shown for comparison with the DNS mean velocity profiles in Figure 5, Figure 6, Figure 7, Figure 8, Figure 14 and Figure 18.

Solutions for the fluctuations at the interface

With the mean-flow part subtracted, a zero-mean field ϕ(x,z,y)\phi(x,z,y) can be represented by a Fourier series in xx and zz such that ϕ(x,z,y)=∑α≠0ϕ^(αx,αz,y)ei(αxx+αzz)\phi(x,z,y) = \sum_{\boldsymbol{\alpha}\ne0}\hat{\phi}(\alpha_x,\alpha_z,y)e^{i(\alpha_xx+\alpha_zz)} where αx\alpha_x and αz\alpha_z are the xx- and zz- components of the planar wavevector α\boldsymbol{\alpha}. Each Fourier mode ϕ^(αx,αz,y)\hat{\phi}(\alpha_x,\alpha_z,y) follows the transform of Eq. (6),

iαxp^=rν−1ν(∂yy2−α2−rνK−1)u^,(11)i\alpha_x\hat{p}=r_\nu^{-1}\nu\left(\partial_{yy}^{2}-\alpha^2-r_\nu K^{-1}\right)\hat{u}, \tag*{(11)}
∂yp^=rν−1ν(∂yy2−α2−rνK−1)v^,(12)\partial_y\hat{p}=r_\nu^{-1}\nu\left(\partial_{yy}^{2}-\alpha^2-r_\nu K^{-1}\right)\hat{v}, \tag*{(12)}
iαzp^=rν−1ν(∂yy2−α2−rνK−1)w^,(13)i\alpha_z\hat{p}=r_\nu^{-1}\nu\left(\partial_{yy}^{2}-\alpha^2-r_\nu K^{-1}\right)\hat{w}, \tag*{(13)}

where α=αx2+αz2\alpha= \sqrt{\alpha_x^2+\alpha_z^2}.

We define a wavevector-aligned coordinate system, of which the column basis vectors e//\boldsymbol{e}_{//}, ey\boldsymbol{e}_y, and e⊥\boldsymbol{e}_{\perp} are obtained by the rotation transform

[e// ey e⊥]=Q⋅[ex ey ez],Q≡[αx/α0−αz/α010αz/α0αx/α],(14)[\boldsymbol{e}_{//}\ \boldsymbol{e}_y\ \boldsymbol{e}_{\perp}] = Q\cdot[\boldsymbol{e}_x\ \boldsymbol{e}_y\ \boldsymbol{e}_z],\qquad Q\equiv \begin{bmatrix} \alpha_x/\alpha& 0 & -\alpha_z/\alpha\\ 0 & 1 & 0\\ \alpha_z/\alpha& 0 & \alpha_x/\alpha \end{bmatrix}, \tag*{(14)}

where ex\boldsymbol{e}_x, ey\boldsymbol{e}_y and ez\boldsymbol{e}_z are the unit vectors in xx, yy and zz. Eq. (D 5) can then be written as

iαp^=rν−1ν(∂yy2−α2−rνK−1)u^//,(15)i\alpha\hat{p}=r_\nu^{-1}\nu\left(\partial_{yy}^{2}-\alpha^2-r_\nu K^{-1}\right)\hat{u}_{//}, \tag*{(15)}
∂yp^=rν−1ν(∂yy2−α2−rνK−1)v^,(16)\partial_y\hat{p}=r_\nu^{-1}\nu\left(\partial_{yy}^{2}-\alpha^2-r_\nu K^{-1}\right)\hat{v}, \tag*{(16)}
0=rν−1ν(∂yy2−α2−rνK−1)u^⊥,(17)0=r_\nu^{-1}\nu\left(\partial_{yy}^{2}-\alpha^2-r_\nu K^{-1}\right)\hat{u}_{\perp}, \tag*{(17)}

where [u^// v^ u^⊥]T=QT⋅[u^ v^ w^]T[\hat{u}_{//}\ \hat{v}\ \hat{u}_{\perp}]^{\mathsf{T}}=Q^{\mathsf{T}}\cdot[\hat{u}\ \hat{v}\ \hat{w}]^{\mathsf{T}}. The equation for u^⊥\hat{u}_{\perp}, (D 7c), decouples from the rest of the problem and can be solved by itself under appropriate boundary conditions. Separately, equations {(xref}D 7a,b) combined with the continuity equation,

iαu^//+∂yv^=0,(18)i\alpha\hat{u}_{//}+\partial_y\hat{v}=0, \tag*{(18)}

form the equation system for u^//\hat{u}_{//}, v^\hat{v}, and p^\hat{p}.

At the free-flow interface, y=0y = 0, let us assume that the values of velocity shear rates ∂yu^//∣0\partial_y\hat{u}_{//}|_0 and ∂yu^⊥∣0\partial_y\hat{u}_{\perp}|_0 and pressure p^∣0\hat{p}|_0 are given. At the floor boundary, y=−hy = -h, no-slip, no-penetration conditions, u^//∣−h=v^∣−h=u^⊥∣−h=0\hat{u}_{//}|_{-h}=\hat{v}|_{-h}=\hat{u}_{\perp}|_{-h}=0, are imposed. The solutions to equations {(xref}D 7)}–{(xref}D 8) yield then a constitutive relationship between velocity and stress at the interface:

\begin{bmatrix} \hat{u}_{//}|_0\\ i\hat{v}|_0\\ \hat{u}_{\perp}|_0 \end{bmatrix} = \boldsymbol{H}_{Br}\cdot \begin{bmatrix} \partial_y\hat{u}_{//}|_0\\ -i\nu^{-1}\hat{p}|_0\\ \partial_y\hat{u}_{\perp}|_0 \end{bmatrix}, \qquad \boldsymbol{H}_{Br}\equiv \begin{bmatrix} \mathcal{L}_{slip} & \alpha\mathcal{K}_{slip} & 0\\ \alpha\mathcal{N}_{trsp} & \alpha\mathcal{K}_{trsp} & 0\\ 0 & 0 & \mathcal{L}_{slip}^{\perp} \end{bmatrix}, \tag*{(19)}\end{bmatrix}

in which HBr\boldsymbol{H}_{Br} is a real-valued matrix with dimension of length, and is an inherent characteristic of a substrate. It contains five independent coefficients, which are functions of α\alpha and hh,

Lslip=K/rν fLs(α~,h~),(20)\mathcal{L}_{slip} = \sqrt{K/r_{\nu}}\, f_{Ls}(\tilde{\alpha},\tilde{h}), \tag*{(20)}
Kslip=K fKs(α~,h~),(21)\mathcal{K}_{slip} = K\, f_{Ks}(\tilde{\alpha},\tilde{h}), \tag*{(21)}
Ntrsp=(K/rν) fNt(α~,h~),(22)\mathcal{N}_{trsp} = (K/r_{\nu})\, f_{Nt}(\tilde{\alpha},\tilde{h}), \tag*{(22)}
Ktrsp=K fKt(α~,h~),(23)\mathcal{K}_{trsp} = K\, f_{Kt}(\tilde{\alpha},\tilde{h}), \tag*{(23)}
Lslip⊥=K/rν fLs⊥(α~,h~),(24)\mathcal{L}^{\perp}_{slip} = \sqrt{K/r_{\nu}}\, f^{\perp}_{Ls}(\tilde{\alpha},\tilde{h}), \tag*{(24)}

where

fLs=(tanh⁡(α~Kh~)−(α~/α~K)tanh⁡(α~h~))2α~K2 ξ,(25)f_{Ls} = \frac{\left(\tanh(\tilde{\alpha}_{K}\tilde{h})-(\tilde{\alpha}/\tilde{\alpha}_{K})\tanh(\tilde{\alpha}\tilde{h})\right)^{2}}{\tilde{\alpha}_{K}^{2}\,\xi}, \tag*{(25)}
fKs=(1+(α~/α~K)2)(1−sech⁡(α~h~)sech⁡(α~Kh~))−2(α~/α~K)tanh⁡(α~h~)tanh⁡(α~Kh~)ξ,(26)f_{Ks} = \frac{\left(1+(\tilde{\alpha}/\tilde{\alpha}_{K})^{2}\right)\left(1-\operatorname{sech}(\tilde{\alpha}\tilde{h})\operatorname{sech}(\tilde{\alpha}_{K}\tilde{h})\right)-2(\tilde{\alpha}/\tilde{\alpha}_{K})\tanh(\tilde{\alpha}\tilde{h})\tanh(\tilde{\alpha}_{K}\tilde{h})}{\xi}, \tag*{(26)}
fNt=1−sech⁡(α~h~)sech⁡(α~Kh~)−(α~/α~K)tanh⁡(α~h~)tanh⁡(α~Kh~)α~K2 ξ,(27)f_{Nt} = \frac{1-\operatorname{sech}(\tilde{\alpha}\tilde{h})\operatorname{sech}(\tilde{\alpha}_{K}\tilde{h})-(\tilde{\alpha}/\tilde{\alpha}_{K})\tanh(\tilde{\alpha}\tilde{h})\tanh(\tilde{\alpha}_{K}\tilde{h})}{\tilde{\alpha}_{K}^{2}\,\xi}, \tag*{(27)}
fKt=tanh⁡(α~h~)−(α~/α~K)tanh⁡(α~Kh~)α~K2 ξ,(28)f_{Kt} = \frac{\tanh(\tilde{\alpha}\tilde{h})-(\tilde{\alpha}/\tilde{\alpha}_{K})\tanh(\tilde{\alpha}_{K}\tilde{h})}{\tilde{\alpha}_{K}^{2}\,\xi}, \tag*{(28)}
fLs⊥=tanh⁡2(α~Kh~)α~K2,(29)f^{\perp}_{Ls} = \frac{\tanh^{2}(\tilde{\alpha}_{K}\tilde{h})}{\tilde{\alpha}_{K}^{2}}, \tag*{(29)}

and where α~\tilde{\alpha} is the dimensionless wavenumber defined by

α~≡αK/rν,(30)\tilde{\alpha} \equiv\alpha\sqrt{K/r_{\nu}}, \tag*{(30)}
α~K=1+α~2,\tilde{\alpha}_{K} = \sqrt{1+\tilde{\alpha}^{2}},
ξ=1−(α~/α~K)tanh⁡(α~h~)tanh⁡(α~Kh~)−(α~/α~K)2sech⁡(α~h~)sech⁡(α~Kh~).\xi= 1-(\tilde{\alpha}/\tilde{\alpha}_{K})\tanh(\tilde{\alpha}\tilde{h})\tanh(\tilde{\alpha}_{K}\tilde{h})-(\tilde{\alpha}/\tilde{\alpha}_{K})^{2}\operatorname{sech}(\tilde{\alpha}\tilde{h})\operatorname{sech}(\tilde{\alpha}_{K}\tilde{h}).

The five attenuating functions f∗(α~,h~)f_{*}(\tilde{\alpha},\tilde{h}) in (D 11) have values in the range of 0 - 1 and account for the effect of substrate depth in (D 10).

On the effect of depth

Figure 30 shows the values of the five attenuating functions f∗f_{*} against h~\tilde{h} and the dimensionless wavelength λ~=2π/α~\tilde{\lambda}=2\pi/\tilde{\alpha}. For fLsf_{Ls}, fKsf_{Ks}, fNtf_{Nt} and fLs⊥f^{\perp}_{Ls}, a threshold h~≈2−3\tilde{h}\approx2-3 can be established, beyond which f∗f_{*} becomes essentially depth-independent, and thus the substrate can be seen as ‘sufficiently deep’ with regards to Lslip\mathcal{L}_{slip}, Kslip\mathcal{K}_{slip}, Ntrsp\mathcal{N}_{trsp} and Lslip⊥\mathcal{L}^{\perp}_{slip}. Such a threshold, however, cannot be established for fKtf_{Kt}. The contour lines of fKtf_{Kt} for large waves follow λ~∼h~\tilde{\lambda}\sim\tilde{h}, with λ~≈5h~\tilde{\lambda}\approx5\tilde{h} for fKt=0.9f_{Kt}=0.9. This scaling behaviour indicates that pressure fluctuations with wavelength λ≳5h~\lambda\gtrsim5\tilde{h} will always perceive the presence of the floor impeding them from inducing any significant transpiration across the interface, regardless of how deep the substrate may be relative to K\sqrt{K}. Therefore, ‘sufficiently deep’ for Ktrsp\mathcal{K}_{trsp} is not a property exclusively inherent to the substrate, but instead depends on the typical length scales of the overlying pressure fluctuations, λp\lambda_{p}.

For near-wall turbulence, the scale λp\lambda_{p} in wall units is typically λp+≈200\lambda_{p}^{+}\approx200 for active eddies (see the figures of spectra in § 4.2 and § 5.1), so for a permeable substrate with K+≲10\sqrt{K^{+}}\lesssim10, we have λ~p=λp/K/rν≳10\tilde{\lambda}_{p}=\lambda_{p}/\sqrt{K/r_{\nu}}\gtrsim10. In such a scenario, figure 30 shows that, for

Darcy-Brinkman-based attenuating coefficients

Figure 30. Darcy-Brinkman-based attenuating coefficients f∗f_{\ast} as functions of dimensionless wavelength λ~≡2π/α~\tilde{\lambda} \equiv2\pi/\tilde{\alpha} and depth h~\tilde{h}.

fKtf_{Kt}, a sufficiently deep substrate must have at least h~≳5\tilde{h} \gtrsim5. Consequently, as an initially deep substrate becomes gradually shallower, the overlying turbulence will first perceive the suppression of pressure-induced transpiration, i.e. Ktrsp\mathcal{K}_{trsp} decreasing at large scales, whereas the effect of the substrate depth on the other coefficients will only be perceived later, when h~≈2\tilde{h} \approx2–33.

Appendix E

Roughness function versus upscaled coefficients from homogenisation

Here we report the scaling of the roughness function obtained in our DNSs with the admittance coefficients for our substrate geometries obtained following the second-order homogenisation model of Bottaro & Naqvi [4] and Naqvi & Bottaro [46]. The coefficients are obtained a priori from a series of simple auxiliary problems on the substrate geometry based on a Stokes-flow assumption. The first coefficient is the conventional slip length, LsL_s, which essentially gives the uu-∂u/∂y\partial u/\partial y admittance. The second is the ‘interfacial permeability’, KintfK_{\mathrm{intf}}, which gives the uu-pp admittance, and from the problem symmetries also the vv-∂u/∂y\partial u/\partial y admittance. The final coefficient is the intrinsic permeability, KyK_y, which gives the vv-pp admittance, and is thus analogue to the KeqtK^{t}_{eq} of equation (4). Results for the DNS-measured roughness function ΔU+\Delta U^+ as a function of the pre-computed LsL_s, KintfK_{\mathrm{intf}} and KyK_y are shown in Figure 31. Overall, the collapse is generally better with KeqtK^{t}_{eq}, as shown in Figure 17(d). We note that the methods used to obtain the coefficients in Bottaro & Naqvi [4] and Naqvi & Bottaro [46] differ for rough and porous substrates. For rough substrates, it is assumed that there is a bottom impermeable wall blocking the flow, and this results in Ky=0K_y = 0 (Bottaro & Naqvi [4]). For permeable substrates, the auxiliary problems are solved in a mesoscale domain that penetrates into the substrate a few grains, typically of order ∼5L\sim5L. A Darcy-like transpiration boundary condition is set at the bottom of the mesoscale domain. As a result, this Darcy transpiration is transferred to the solution, yielding Ky=KK_y = K. For our present substrates, we would assume that at least for substrate depths h=1Dh = 1D-2D2D, if not for h=5Dh = 5D-7D7D, the rough-wall framework is more appropriate, and thus Ky=0K_y = 0, as portrayed in 31(c). If we overlooked the presence of a bottom wall in our substrates and assumed Ky=KK_y = K for all of them, the result would be as shown in Figure 13(f).

Velocity deficit as a function of upscaled substrate coefficients

Figure 31. Velocity deficit ΔU+\Delta U^+ for the present DNS substrates as a function of their upscaled coefficients derived from homogenisation [4, 46]. Symbols are as in Figure 9. (a) slip length, LsL_s; (b) interface permeability, KintfK_{\mathrm{intf}}; (c) intrinsic permeability, KyK_y.

Appendix F

In this final appendix we compile roughness function results for other substrate topologies from both experimental and numerical literature. Most of these have larger permeabilities K+K^+ than the scope of the present work. As a result, we would expect some of our underlying assumptions to fail, such as assuming Stokes flow for the microscale flow when deriving expression 5.1 for KeqtK_{\mathrm{eq}}^{t}, or that the direct effect of texture granularity on the overlying turbulence is small. The collapse of roughness function data with Keqt+K_{\mathrm{eq}}^{t+}, and with K+K^+ for reference, is portrayed in 32, showing reasonably good agreement even at these larger K+K^+ values. We note that the substrates of Habibi Khorasani et al. (2024) appear to follow a lower ΔU+\Delta U^+ trend, especially for small Keqt+K_{\mathrm{eq}}^{t+}. These mesh substrates are different from the rest in that their exposed interface is perfectly flat. This implies an absence of interfacial roughness, probably suppressing partially the increase in drag compared to substrates with roughness. Following this, we have conducted an additional set of simulations for our mesh topologies, where we made the interface flat as in Habibi Khorasani et al. (2024) or, conversely, extended the protruding ligaments from their original height ℓ/2\ell/2 to ℓ\ell. The parameters of these simulations are summarised in Table 3. The flat-interface cases follow the same trend of the substrates of Habibi Khorasani et al. (2024), while the taller-protrusion cases show a trend similar to all the other substrates with rough interfaces. We note that this figure includes results for various irregular grain topologies [72, 15, 31], and they follow the same trend of regular ones.

CaseReτRe_{\tau}L+L^{+}g/Lg/Lh/Lh/Lϵ\epsilonK+\sqrt{K^{+}}Keqt+\sqrt{K_{\mathrm{eq}}^{t+}}ΔU+\Delta U^{+}
MFPd-36-78182.735.97/950.875.345.338.24
MFPd-48-56182.747.85/950.583.303.293.64
MFPd-48-72182.747.813/1850.815.945.948.20
MFPd-48-78182.747.87/950.877.127.119.48
MTPd-48-72182.747.813/1850.815.945.099.77
MTPd-48-78182.747.87/950.877.126.059.50

Table 3. Parameters for mesh substrates with flat interfaces (MFPd) and tall interfacial protrusions (MTPd). LL is the grain pitch, gg the gap size, hh the substrate depth, ϵ\epsilon the porosity, KK the permeability, KeqtK_{\mathrm{eq}}^{t} the transpiration equivalent permeability, and ΔU+\Delta U^{+} the resulting roughness function. The ‘+’ superscripts indicate viscous scaling.

Figure 32. Velocity deficit ΔU+\Delta U^{+} as a function of (a)(a) the permeability K+K^{+} and (b)(b) the equivalent permeability Keqt+K_{\mathrm{eq}}^{t+} for the present DNSs and other substrates in the literature. ● and lines faded in the background, present staggered cubes; ◁ meshes of cases MPd; ▷, meshes of cases MFPd; ▽, meshes of cases MTPd; ◻, randomly packed spheres from [72]; ◇, staggered cubes from [35]; +, collocated spheres from [33]; ◯, reticulated foams from [15]; ×, randomly packed spheres from [31]; △, mesh lattices from [26].

References

  1. [1]Abderrahman-Elena, Nabil, Fairhall, Chris T & García-Mayoral, Ricardo 2019 Modulation of near-wall turbulence in the transitionally rough regime. J. Fluid Mech. 865, 1042–1071.DOI
  2. [2]Abderrahman-Elena, Nabil & García-Mayoral, Ricardo 2017 Analysis of anisotropically permeable surfaces for turbulent drag reduction. Phys. Rev. Fluids 2 (11), 114609.DOI
  3. [4]Bottaro, Alessandro & Naqvi, Sahirsh B. 2020 Effective boundary conditions at a rough wall: a high-order homogenization approach. Meccanica 55 (9), 1781–1800.DOI
  4. [5]Bou-Zeid, Elie, Anderson, William, Katul, Gabriel G. & Mahrt, Larry 2020 The persistent challenge of surface heterogeneity in boundary-layer meteorology: A review. Bound.-Layer Meteorol. 177 (2-3), 227–245.DOI
  5. [6]BREUGEM, WP, BOERSMA, BJ & UITTTENBOGAARD, RE 2006 The influence of wall permeability on turbulent channel flow. J. Fluid Mech. 562, 35–72.DOI
  6. [7]BREUGEM, WIM-PAUL & BOERSMA, BENDIKS-JAN 2005 Direct numerical simulations of turbulent flow over a permeable wall using a direct and a continuum approach. Phys. Fluids 17 (2), 025103.DOI
  7. [8]BRINKMAN, H. C. 1949 A calculation of the viscous force exerted by a flowing fluid on a dense swarm of particles. Flow Turbul. Combust. 1 (1).
  8. [10]CHEN, ZISHEN & GARCÍA-MAYORAL, RICARDO 2023 Examination of outer-layer similarity in wall turbulence over obstructing surfaces. J. Fluid Mech. 973, A31.DOI
  9. [11]CHUNG, DANIEL, HUTCHINS, NICHOLAS, SCHULTZ, MICHAEL P & FLACK, KAREN A 2021 Predicting the drag of rough surfaces. Annu. Rev. Fluid Mech. 53, 439–471.DOI
  10. [12]COOPER, JAMES R., OCKLEFORD, ANNIE, RICE, STEPHEN P. & POWELL, D. MARK 2017 Does the permeability of gravel river beds affect near-bed hydrodynamics? Earth Surf. Process. Landf. 43 (5), 943–955.DOI
  11. [13]DARCY, HENRY 1856 Les fontaines publiques de la ville de Dijon: exposition et application des principes à suivre et des formules à employer dans les questions de distribution d’eau, vol. 1. Victor dalmont.
  12. [14]EFSTATHIOU, CHRISTOPH & LUHAR, MITUL 2018 Mean turbulence statistics in boundary layers over high-porosity foams. J. Fluid Mech. 841, 351–379.DOI
  13. [15]ESTEBAN, LB, RODRÍGUEZ-LÓPEZ, E, FERREIRA, MA & GANAPATHISUBRAMANI, B 2022 Mean flow of turbulent boundary layers over porous substrates. Phys. Rev. Fluids 7 (9), 094603.DOI
  14. [16]FAIRHALL, C. T., ABDERRAHAMAN-ELENA, N. & GARCÍA-MAYORAL, R. 2019 The effect of slip and surface texture on turbulence over superhydrophobic surfaces. J. Fluid Mech. 861, 88–118.DOI
  15. [17]FAIRHALL, C. T. & GARCÍA-MAYORAL, R. 2018 Spectral analysis of the slip-length model for turbulence over textured superhydrophobic surfaces. Flow Turbul. Combust. 100 (4), 961–978.DOI
  16. [18]FANG, HONGWEI, HAN, XU, HE, GUOJIAN & DEY, SUBHASHISH 2018 Influence of permeable beds on hydraulically macro-rough flow. J. Fluid Mech. 847, 552–590.DOI
  17. [19]FLORES, OSCAR & JIMENEZ, JAVIER 2006 Effect of wall-boundary disturbances on turbulent channel flows. J. Fluid Mech. 566, 357–376.
  18. [20]GARCÍA-MAYORAL, RICARDO, CHUNG, DANIEL, DURBIN, PAUL, HUTCHINS, NICHOLAS, KNOPP, TOBIAS, MCKEON, BEVERLEY J., PIOMELLI, UGO & SANDBERG, RICHARD D. 2024 Challenges and perspective on the modelling of high-re, incompressible, non-equilibrium, rough-wall boundary layers. J. Turbul.DOI
  19. [21]GARCÍA-MAYORAL, RICARDO & JIMÉNEZ, JAVIER 2011 Hydrodynamic stability and breakdown of the viscous regime over riblets. J. Fluid Mech. 678, 317–347.
  20. [22]GARCÍA-MAYORAL, RICARDO & JIMÉNEZ, JAVIER 2012 Scaling of turbulent structures in riblet channels up to Reτ ≈550. Phys. Fluids 24 (10).DOI
  21. [23]GÓMEZ-DE-SEGURA, GARAZI & GARCÍA-MAYORAL, RICARDO 2019 Turbulent drag reduction by anisotropic permeable substrates–analysis and direct numerical simulations. J. Fluid Mech. 875, 124–172.
  22. [24]GÓMEZ-DE-SEGURA, G., SHARMA, A. & GARCÍA-MAYORAL, R. 2018a Turbulent drag reduction using anisotropic permeable substrates. Flow Turbul. Combust. 100 (4), 995–1014.DOI
  23. [25]GÓMEZ-DE-SEGURA, G., SHARMA, A. & GARCÍA-MAYORAL, R. 2018b Virtual origins in turbulent flows over complex surfaces. In Proc 2018 CTR Summer Program, pp. 277–286.
  24. [26]HABIBI KHORASANI, SEYED MORTEZA, LUHAR, MITUL & BAGHERI, SHERVIN 2024 Turbulent flows over porous lattices: alteration of near-wall turbulence and pore-flow amplitude modulation. J. Fluid Mech. 984, A63.DOI
  25. [27]IBRAHIM, JOSEPH I, GÓMEZ-DE-SEGURA, GARAZI, CHUNG, DANIEL & GARCÍA-MAYORAL, RICARDO 2021 The smooth-wall-like behaviour of turbulence over drag-altering surfaces: a unifying virtual-origin framework. J. Fluid Mech. 915, A56.DOI
  26. [28]JIMÉNEZ, JAVIER 2004 Turbulent flows over rough walls. Annu. Rev. Fluid Mech. 36, 173–196.
  27. [29]JIMENEZ, JAVIER, UHLMANN, MARKUS, PINELLI, ALFREDO & KAWAHARA, GENTA 2001 Turbulent shear flow over active and passive porous surfaces. J. Fluid Mech. 442, 89–117.
  28. [31]Karra, Shashank K., Apte, Sourabh V., He, Xiaoliang & Scheibe, Timothy D. 2023 Pore-resolved investigation of turbulent open channel flow over a randomly packed permeable sediment bed. J. Fluid Mech. 971.DOI
  29. [32]Kawano, Koki, Motoki, Shingo, Shimizu, Masaki & Kawahara, Genta 2021 Ultimate heat transfer in ‘wall-bounded’ convective turbulence. J. Fluid Mech. 914, A13.arxiv.org/abs/2004.08831
  30. [33]Kim, Taehoon, Blois, Gianluca, Best, James L & Christensen, Kenneth T 2020 Experimental evidence of amplitude modulation in permeable-wall turbulence. J. Fluid Mech. 887.DOI
  31. [34]Kong, F & Schetz, J 1982 Turbulent boundary layer over porous surfaces with different surface geometries. In 2th Aerospace Sciences Meeting, p. 30.DOI
  32. [35]Kuwata, Y & Suga, K 2016a Lattice boltzmann direct numerical simulation of interface turbulence over porous and rough walls. Int. J. Heat Fluid Flow 61, 145–157.DOI
  33. [39]Ligrani, Phillip M & Moffat, Robert J 1986 Structure of transitionally rough and fully rough turbulent boundary layers. J. Fluid Mech. 162, 69–98.DOI
  34. [41]Luchini, P 1996 Reducing the turbulent skin friction. In Computational methods in applied sciences’ 96 (Paris, 9–13 September 1996), pp. 465–470.
  35. [42]Lévy, Thérèse 1983 Fluid flow through an array of fixed particles. Int. J. Eng. Sci. 21 (1), 11–23.DOI
  36. [43]Manes, Costantino, Poggi, Davide & Ridolfi, Luca 2011 Turbulent boundary layers over permeable walls: scaling and near-wall structure. J. Fluid Mech. 687, 141–170.DOI
  37. [44]Manes, Costantino, Pokrajac, Dubravka, McEwan, Ian & Nikora, Vladimir 2009 Turbulence structure of open channel flows over permeable and impermeable beds: A comparative study. Phys. Fluids 21 (12), 125109.DOI
  38. [45]Moody, Lewis F. 1944 Friction factors for pipe flow. J. Fluids Eng. 66 (8), 671–678.DOI
  39. [46]Naqvi, Sahrish B & Bottaro, Alessandro 2021 Interfacial conditions between a free-fluid region and a porous medium. Int. J. Multiph. Flow 141, 103585.DOI
  40. [47]Orlandi, P. & Leonardi, S. 2006 DNS of turbulent channel flows with two- and three-dimensional roughness. J. Turbul. 7, N73.DOI
  41. [48]Orlandi, P. & Leonardi, S. 2008 Direct numerical simulation of three-dimensional turbulent rough channels: parameterization and flow physics. J. Fluid Mech. 606, 399–415.DOI
  42. [49]Orlandi, Paolo, Leonardi, S & Antonia, RA 2006 Turbulent channel flow with either transverse or longitudinal roughness elements on one wall. J. Fluid Mech. 561, 279–305.DOI
  43. [52]Rosti, Marco E, Cortelezzi, Luca & Quadrio, Maurizio 2015 Direct numerical simulation of turbulent channel flow over porous walls. J. Fluid Mech. 784, 396–442.DOI
  44. [53]Rousseau, Gauthier & Ancey, Christophe 2020 Scanning piv of turbulent flows over and through rough porous beds using refractive index matching. Exp. Fluids 61 (8), 172.DOI
  45. [54]Sanchez-Palencia, E. 1982 On the asymptotics of the fluid flow past an array of fixed obstacles. Int. J. Eng. Sci. 20 (12), 1291–1301.DOI
  46. [55]Sharma, Akshat 2020 Turbulent flows over canopies. PhD thesis, University of Cambridge.DOI
  47. [56]Sharma, Akshat & García-Mayoral, Ricardo 2020a Scaling and dynamics of turbulence over sparse canopies. J. Fluid Mech. 888, A1.DOI
  48. [57]Sharma, Akshat & García-Mayoral, Ricardo 2020b Turbulent flows over dense filament canopies. J. Fluid Mech. 888, A2.DOI
  49. [58]Sharma, Akshat, Gómez-de Segura, Garazi & García-Mayoral, Ricardo 2017 Linear stability analysis of turbulent flows over dense filament canopies. In Tenth International Symposium on Turbulence and Shear Flow Phenomena. Begel House Inc.DOI
  50. [59]Shen, Guangchen, Yuan, Junlin & Phanikumar, Mantha S 2020 Direct numerical simulations of turbulence and hyporheic mixing near sediment–water interfaces. J. Fluid Mech. 892, A20.DOI
  51. [60]Stroh, A, Schäfer, K, Frohnapeel, B & Forooghi, P 2020 Rearrangement of secondary flow over spanwise heterogeneous roughness. J. Fluid Mech. 885, R5.DOI
  52. [62]Suga, K, Matsumura, Y, Ashitaka, Y, Tominaga, S & Kaneda, M 2010 Effects of wall permeability on turbulence. Int. J. Heat Fluid Flow 31 (6), 974–984.DOI
  53. [63]Suga, K, Mori, M & Kaneda, M 2011 Vortex structure of turbulence over permeable walls. Int. J. Heat Fluid Flow 32 (3), 586–595.DOI
  54. [64]Taylor, GI 1971 A model for the boundary condition of a porous material. part 1. J. Fluid Mech. 49 (2), 319–326.DOI
  55. [65]Voermans, JJ, Ghalsiberti, M & Ivey, GN 2017 The variation of flow and turbulence across the sediment–water interface. J. Fluid Mech. 824, 413–437.DOI
  56. [66]Wang, Wenkang, Chu, Xu, Lozano-Durán, Adrián, Helmig, Rainer & Weigand, Bernhard 2021 Information transfer between turbulent boundary layers and porous media. J. Fluid Mech. 920, A21.DOI
  57. [67]Wangsawijaya, D.D., Jaiswal, P. & Ganapathisubramani, B. 2023 Towards decoupling the effects of permeability and roughness on turbulent boundary layers. J. Fluid Mech. 967, R2.DOI
  58. [68]Wangsawijaya, D. D., Baidya, R., Chung, D., Marusic, I. & Hutchins, N. 2020 The effect of spanwise wavelength of surface heterogeneity on turbulent secondary flows. J. Fluid Mech. 894, A7.DOI
  59. [69]Xie, W., Fairhall, C. T. & García-Mayoral, R. 2024 Resolving turbulence and drag over textured surfaces using texture-less simulations: the case of slip/no-slip textures. J. Fluid Mech. 1000, A36.DOI
  60. [70]Zagni, Anthony FE & Smith, Kenneth VH 1976 Channel flow over permeable beds of graded spheres. J. Hydraul. Div. 102 (2), 207–222.DOI
  61. [72]Zippe, Hans J & Graf, Walter H 1983 Turbulent boundary-layer flow over permeable and non-permeable rough surfaces. J. Hydraul. Res. 21 (1), 51–65.DOI

Paper details

vlnc:W_bfhxjy8q
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents