The wake flow past an axisymmetric body of revolution at a diameter-based Reynolds number $Re=u_{\infty }D/\nu =5000$ is investigated via a direct numerical simulation. The study is focused on identification of coherent vortical motions and the dominant frequencies in this flow. Three dominant coherent motions are identified in the wake: the vortex shedding motion with the frequency of $St=fD/u_{\infty }=0.27$, the bubble pumping motion with $St=0.02$, and the very-low-frequency (VLF) motion originated in the very near wake of the body with the frequency $St=0.002$–$0.005$. The vortex shedding pattern is demonstrated to follow a reflectional symmetry breaking mode, whereas the vortex loops are shed alternatingly from each side of the vortex shedding plane, but are subsequently twisted and tangled, giving the resulting wake structure a helical spiraling pattern. The bubble pumping motion is confined to the recirculation region and is a result of a Görtler instability. The VLF motion is related to a stochastic destabilisation of a steady symmetric mode in the near wake and manifests itself as a slow, precessional motion of the wake barycentre. The VLF mode with $St=0.005$ is also detectable in the intermediate wake and may be associated with a low-frequency radial flapping of the shear layer.

The Poiseuille and thermal transpiration flows of a dense gas between two parallel plates are investigated on the basis of the Enskog kinetic equation under the diffuse reflection boundary condition. In contrast to the case of an ideal gas, the density and the gradients of pressure and the normal stress component in the flow direction are not uniform in the direction normal to the plates for a dense gas. The non-uniform normal stress gradient contributes also to the acceleration or deceleration of the thermal transpiration flow for small Knudsen numbers. The profiles of mass and heat flows as well as the net mass flows are obtained for various Knudsen numbers and ratios of the molecular diameter to the distance of plates. In the analysis of the Poiseuille flow, most characteristics of a force-driven flow with a small force are recovered. However, for the case of a dense gas, differences between the force-driven and the present pressure-driven flows are observed even within the linearized regime for small force and pressure gradient, especially at the microscopic level. The behaviour of the velocity distribution functions, in particular, the way in which they approach the ones for the Boltzmann equation as the molecular diameter becomes smaller, is clarified.

Drop impact experiments allow the modelling of a wide variety of natural processes, from raindrop impacts to planetary impact craters. In particular, interpreting the consequences of planetary impacts requires an accurate description of the flow associated with the cratering process. In our experiments, we release a liquid drop above a deep liquid pool to investigate simultaneously the dynamics of the cavity and the velocity field produced around the air–liquid interface. Using particle image velocimetry, we analyse quantitatively the velocity field using a shifted Legendre polynomial decomposition. We show that the velocity field is more complex than considered in previous models, in relation to the non-hemispherical shape of the crater. In particular, the velocity field is dominated by degrees 0 and 1, with contributions from degree 2, and is independent of the Froude and the Weber numbers when these numbers are large enough. We then derive a semi-analytical model based on the Legendre polynomial expansion of an unsteady Bernoulli equation coupled with a kinematic boundary condition at the crater boundary. This model explains the experimental observations and can predict the time evolution of both the velocity field and the shape of the crater, including the initiation of the central jet.

The Hall–Vinen–Bekharevich–Khalatnikov (HVBK) model is widely used to numerically study quantum turbulence in superfluid helium. Based on the two-fluid model of Tisza and Landau, the HVBK model describes the normal (viscous) and superfluid (inviscid) components of the flow using two Navier–Stokes type of equations, coupled through a mutual friction force term. We derive transport equations for the third-order moments for each component of velocity involving the fourth-order moments, which are classical probes for internal intermittency at any scale, and revealing the probability of rare and strong fluctuations. Budget equations are assessed through direct numerical simulations of the HVBK flow. We simulate a forced homogeneous isotropic turbulent flow with Reynolds number of the normal fluid (based on Taylor's microscale) close to 100. Values from 0.1 to 10 are considered for the ratio between the normal and superfluid densities. For these flows, an inertial range is not discernible and the restricted scaling range approach is used to take into account the finite Reynolds number (FRN) effect. We analyse the importance of each term in budget equations and emphasize their role in energy exchange between normal and superfluid components. Some interesting features are observed: (i) transport and pressure-related terms are dominant, similarly to single-fluid turbulence; and (ii) the mathematical signature of the FRN effect is weak despite the low value of the Reynolds number. The flatness of the velocity derivatives is finally studied through the transport equations and their limit for very small scales, and it is shown to gradually increase for lower and lower temperatures, for both normal fluid and superfluid. This similarity highlights the strong locking of the two fluids. The flatness factors are also found in reasonable agreement with classical turbulence.

To investigate the initial development of the Rayleigh–Bénard–Marangoni (RBM) instability in a relatively deep domain, direct numerical simulations for a large range of Marangoni and Rayleigh numbers were performed. In the simulations, the surface was assumed to be flat and surface cooling was modelled by a constant heat flux. The small-scale dynamics of the flow and temperature fields near the surface was fully resolved by using a non-uniform vertical grid distribution. A detailed investigation of the differences in physical mechanisms that drive the Rayleigh- and Marangoni-dominated instabilities is presented. To this end, various properties such as the maturation rate of convection cells, the fluctuating kinetic energy and the surface characteristic length scale were studied. It was confirmed that buoyancy forces and surface-temperature-gradient-driven Marangoni forces enhance one another in promoting the development of the RBM instability. When using a relevant measure of the effective thermal boundary layer thickness as length scale, both the critical Marangoni and Rayleigh numbers, obtained for the purely Marangoni- and purely Rayleigh-driven instabilities, were found to be in good agreement with the literature.

Hydraulic fracturing for oil and gas production from shale formations, as well as natural geological phenomena, involve the propagation of thin viscous films within elastic media. For viscous fluids, stress diverges as the thickness of the film tends to zero, arresting the propagation of the film, and thus implying the contact line paradox. For free-surface films, this paradox is resolved by considering a precursor film, leading to Tanner's law. This approach was extended recently for viscous films between a thin elastic plate and a rigid solid, allowing calculation of the film propagation rate. In this work, we examine the effect of a pre-wetting layer on the rate of propagation of a viscous flow within an infinitely deep and long domain. We analyse the linear and nonlinear dynamic problems, and perform a self-similarity analysis. We find that peeling front propagation scales as time to the power of $1/9$ and $1/3$ for thin and thick pre-wetting layer limits, respectively. Our results contribute to the understanding of the contact line paradox in elastic media and the crucial role of the pre-wetting layer in resolving it.

We investigate the fundamental time scales that characterise the statistics of fragmentation under homogeneous isotropic turbulence for air–water bubbly flows at moderate to large bubble Weber numbers, $We$. We elucidate three time scales: $\tau _r$, the characteristic age of bubbles when their subsequent statistics become stationary; $\tau _\ell$, the expected lifetime of a bubble before further fragmentation; and $\tau _c$, the expected time for the air within a bubble to reach the Hinze scale, radius $a_H$, through the fragmentation cascade. The time scale $\tau _\ell$ is important to the population balance equation (PBE), $\tau _r$ is critical to evaluating the applicability of the PBE no-hysteresis assumption, and $\tau _c$ provides the characteristic time for fragmentation cascades to equilibrate. By identifying a non-dimensionalised average speed $\bar {s}$ at which air moves through the cascade, we derive $\tau _c=C_\tau \varepsilon ^{-1/3} a^{2/3} (1-(a_{max}/a_H)^{-2/3})$, where $C_\tau =1/\bar {s}$ and $a_{max}$ is the largest bubble radius in the cascade. While $\bar {s}$ is a function of PBE fragmentation statistics, which depend on the measurement interval $T$, $\bar {s}$ itself is independent of $T$ for $\tau _r \ll T \ll \tau _c$. We verify the $T$-independence of $\bar {s}$ and its direct relationship to $\tau _c$ using Monte Carlo simulations. We perform direct numerical simulations (DNS) at moderate to large bubble Weber numbers, $We$, to measure fragmentation statistics over a range of $T$. We establish that non-stationary effects decay exponentially with $T$, independent of $We$, and provide $\tau _r=C_{r} \varepsilon ^{-1/3} a^{2/3}$ with $C_{r}\approx 0.11$. This gives $\tau _r\ll \tau _\ell$, validating the PBE no-hysteresis assumption. From DNS, we measure $\bar {s}$ and find that for large Weber numbers ($We>30$), $C_{\tau }\approx 9$. In addition to providing $\tau _c$, this obtains a new constraint on fragmentation models for PBE.

With an immersed-boundary lattice-Boltzmann method, we consider the transit of a three-dimensional initially spherical capsule with a viscoelastic membrane through a cross-slot microchannel. The capsule is released with a small initial off-centre distance in the feeding channel, to mimic experiments where capsules or cells are not perfectly aligned with the centreline. Our main objective is to establish the phase diagram of the capsule's deformation modes as a function of the flow inertia and capsule membrane viscosity. We mainly find three deformation modes in the channel cross-slot. For a capsule with low membrane viscosity, a quasi-steady mode occurs at low Reynolds numbers ($Re$), in which the capsule can reach and maintain a steady ellipsoidal shape near the stagnation point, for a considerable time period. With $Re$ increasing to 20, an overshoot–retract mode is observed. The capsule deformation oscillates on an inertial–elastic time scale, suggesting that the dynamics is mainly driven by the balance of the inertial and membrane elastic forces. The membrane viscosity slows down the capsule deformation and suppresses the overshoot–retract mode. A capsule with high membrane viscosity undergoes a continuous-elongation mode, in which its deformation keeps increasing during most of its journey in the channel cross-slot. We summarise the results in phase diagrams, and propose a scaling model which can predict the deformation modes of a viscoelastic capsule in the inertial flow regime. We also discuss implications of the present findings for practical experiments for mechanical characterisation of capsules or cells.

Reconstructing high-resolution flow fields from sparse measurements is a major challenge in fluid dynamics. Existing methods often vectorize the flow by stacking different spatial directions on top of each other, hence confounding the information encoded in different dimensions. Here, we introduce a tensor-based sensor placement and flow reconstruction method which retains and exploits the inherent multidimensionality of the flow. We derive estimates for the flow reconstruction error, storage requirements and computational cost of our method. We show, with examples, that our tensor-based method is significantly more accurate than similar vectorized methods. Furthermore, the variance of the error is smaller when using our tensor-based method. While the computational cost of our method is comparable to similar vectorized methods, it reduces the storage cost by several orders of magnitude. The reduced storage cost becomes even more pronounced as the dimension of the flow increases. We demonstrate the efficacy of our method on three examples: a chaotic Kolmogorov flow, in situ and satellite measurements of the global sea surface temperature and three-dimensional unsteady simulated flow around a marine research vessel.

Vertically neutral collapse of a pulsating bubble occurs when the boundaries above or below the bubble balance the buoyancy effect over a pulsation. In this study, the vertically neutral collapse of a bubble near a vertical rigid wall below the free surface is investigated. The boundary integral method (BIM) is employed to model the bubble dynamics with an open-domain free surface. Moreover, this method is validated against several buoyant bubble experiments. Bubble dynamics in such conditions are associated with three dimensionless parameters: the bubble-free surface distance $\gamma _{{f}}$, bubble–wall distance $\gamma _{{w}}$ and buoyancy parameter $\delta$. We derive the Kelvin impulse of a spherical bubble and the algebraic relationship for vertically neutral collapse, which proves to be accurate for predicting vertically neutral collapse when the bubble is relatively far from the boundaries. Four patterns of the vertically neutral collapse of the bubble for different $\gamma _{{w}}$ and $\gamma _{{f}}$ are identified: (i) formally downward jet; (ii) annular collapse; (iii) horizontal jet; and (iv) weak jet. Despite the downward jet shape, the ‘formally downward jet’ is in the vertically neutral collapse state in terms of the profile of toroidal bubbles and the orientation of local high-pressure zones around the bubble at jet impact. A bulge with a high curvature above the bubble in the ‘annular collapse’ pattern is formed during bubble collapse under two local high-pressure zones at the left and right extremities of the bubble. The ‘horizontal jet’ pattern has the greatest potential to attack the wall, and the power laws of the moment of the jet impact, jet velocity and bubble displacement with respect to the theoretical Kelvin impulse are discussed. In particular, we quantitatively illustrate the role of the free surface on bubble migration towards the wall through the variational power-law exponents of the bubble displacement with respect to $\gamma _{{w}}$.

High resolution aeroacoustic source analysis is a prerequisite to address the noise concerns and release the full benefits of wake-ingesting propellers. In this work, the aeroacoustic sources of a two-bladed propeller ingesting the wake of an aerofoil are investigated using large eddy simulation in conjunction with two different source identifying approaches. The first approach is the numerical beamforming that utilizes both the classical and wavelet-based beamforming techniques, which determine the phase variations of sources at the low to mid frequencies and reveal that the high-frequency sources are phase-independent. To further improve the spatial resolution of source identification, a new near-field aeroacoustic source analysis approach based on the acoustic analogy is developed in this work. In particular, the on-surface source terms emanating the far-field noise are derived based on the Ffowcs Williams and Hawkings equation for low Mach number flows and constant rotating propellers. Through the incorporation of the simulation results into the proposed source analysis approach, various types of aeroacoustic sources are identified and studied by visualizing their distributions on the propeller surfaces, correlating to flow features and examining the noise spectra and directivity. While the leading edge sources are highly correlated with the wake interaction process, the sources at the mid-chord and the trailing edge of the blade can maintain their strength across most revolving angles. Overall, the proposed analysis approaches extend the capability of computational fluid dynamics and enable the detailed study of noise generation mechanisms of wake-ingesting propeller noise.

This work investigates the stability and transition to turbulence in a diverging channel subjected to a time-varying trapezoidal-shaped inflow boundary condition. Numerical simulations are performed for different deceleration rates and Reynolds numbers while maintaining a constant acceleration rate. The flow transition begins with two-dimensional primary instability with the formation of inflectional velocity profiles, followed by local separation and the emergence of an array of shear layer vortices. We divide simulation cases systematically into three categories based on the onset of secondary instability and the generation of streamwise vorticity. At low and medium Reynolds numbers (type I), the spanwise vortex rolls formed by inflectional instability remain two-dimensional and diffuse at the channel centre without exhibiting further instabilities. At high Reynolds numbers and deceleration rates (type II), the rolled shear layer exhibits secondary instability during the zero mean inflow phase, followed by local incipient turbulent structure formation. The streamwise vorticity that develops over the shear layer structures causes oscillations with a spanwise wavelength similar to those associated with the elliptic instability in a counter-rotating vortex pair. Using the Lamb–Oseen approximation of vortices in conjunction with the dynamic mode decomposition algorithm of the three-dimensional flow field, we captured successfully the characteristics of the secondary instability. The third category (type III) is characterized by periodic unsteady separation, secondary instability, and merging of shear layer vortices, which occurs when Reynolds numbers are high and deceleration rates are low.

This paper proposes a new approach for evaluating numerically the forces and moments applied to a circular cylinder that is immersed in a fluid and which translates and rotates near a plane wall. Under the proposed approach, the flow is decomposed into inner and outer flows. The inner flow represents the flow in the thin interstice between the cylinder and the wall, and is obtained as an analytic expression using lubrication theory. The outer flow represents the flow far from the interstice, which does not depend on the magnitude of the gap between the cylinder and the wall, when the gap is small. The outer flow is obtained using numerical simulation as a function of both the Reynolds number and the slip coefficient. The force and moment coefficients are then obtained, as functions of the Reynolds number, slip coefficient and gap-to-diameter ratio, by combining the inner and outer solutions. Importantly, since the outer flow does not depend on the gap-to-diameter ratio, the parameter space to be explored by numerical simulations is greatly reduced compared to using finite gap ratio simulations. Moreover, the numerical difficulties associated with resolving the interstitial flow are avoided. The proposed approach can be extended to a wide range of rolling bodies, including spherical particles and wheels, and should significantly reduce the computational expense required to model the hydrodynamic forces and predict the subsequent motion of such bodies.

By using direct numerical simulations (DNS) of bubbly flows with passive scalars, we show a transition in the scalar spectra from a $k^{-5/3}$ to a $k^{-3}$ scaling with the wavenumber $k$, in contrast with those of single-phase isotropic turbulence. For cases with a mean scalar gradient in the horizontal direction, the scalar spectrum decays faster than $k^{-3}$ at high wavenumbers. While the $k^{-3}$ scaling is well established in the bubbly flow velocity spectrum, the scalar spectrum behaviour is not fully understood. We find that the transition length scale of the scalar spectra is comparable to or below the bubble diameter and decreases with the molecular diffusivity of the scalar in the liquid. We use DNS to compute the scalar spectrum budget and show that the scalar fluctuations are produced by the mean scalar gradient at length scales above the bubble diameter, contrary to the velocity fluctuations. At length scales below the bubble diameter, the net scalar transfer scales as $k^{-1}$ inducing the $k^{-3}$ scaling of the scalar spectra. This finding is consistent with the hypothesis proposed by Dung et al. (J. Fluid Mech., vol. 958, 2023, p. A5) about the physical mechanism behind the $k^{-3}$ scaling. We also show dependencies of the bubble suspension's convective scalar diffusivity on the gas volume fraction and molecular diffusivity that differ based on the direction of the mean scalar gradient. For a mean scalar gradient in the vertical direction, we find and qualitatively explain a significant effect of the molecular diffusivity in the gas on the convective scalar diffusivity.

Zonal jets and inertia–gravity waves are ubiquitous on planets such as Earth, Jupiter and Saturn. Motivated by the modification of energy flux of balanced flow by inertia–gravity waves, this paper studies the impact of near-inertial waves (NIWs) on zonal jets on a $\beta$-plane. Using a two-dimensional quasi-geostrophic and NIW coupled system on a $\beta$-plane (Xie & Vanneste, J. Fluid Mech., vol. 774, 2015, pp. 143–169), we find NIWs catalytically impact several features of zonal jets. The NIWs inhibit jet formation due to the waves’ catalytic induction of downscale mean energy flux. As the strength of NIWs increases, a critical point exists beyond which zonal jets are annihilated. The jet spacing is captured by the Rhines scale $L\sim \sqrt {U/\beta }$ with $U$ estimated from the upscale energy flux induced by the mean flow alone, which again shows that the NIWs’ impact is catalytic. Also, the temporal asymmetry of NIWs leads to the spatial asymmetry of jet dynamics. The jet profiles are asymmetric with a stronger shear on the left flank. And similar to the left turning of vortex dipole under the impact of NIWs, the NIW-modified jets migrate poleward. The NIWs also show a catalytic role in jet migration: the net momentum flux directly induced by NIWs is of secondary importance in the zonal mean momentum dynamics and impedes jet migration, while the advective effect of NIW-modified mean flow dominates the jet migration velocity.

We investigate the Lagrangian properties of fingering convection in a parameter regime that is far beyond marginality, but that does not lead to staircase formation. We find that most fluid particles perform very short vertical excursions before switching direction. Only a small fraction of them travels vertically in the same direction for lengths substantially larger than the typical spatial scales associated with fingering instabilities. These long travels are associated with vertical velocities, buoyancy anomalies and advective fluxes constrained within a relatively narrow range of values. Despite their limited occurrence, they contribute the overwhelming majority of the up-gradient buoyancy flux. A description of fingering convection emerges that partitions the fluid into an incoherent background, where fluxes may be down-gradient, and a set of relatively long-lived, coherent, buoyancy-carrying structures that travel vertically, interacting with each other.

Surface buoyancy gradients over a quasigeostrophic fluid permit the existence of surface-trapped Rossby waves. The interplay of these Rossby waves with surface quasigeostrophic turbulence results in latitudinally inhomogeneous mixing that, under certain conditions, culminates in a surface buoyancy staircase: a meridional buoyancy profile consisting of mixed zones punctuated by sharp buoyancy gradients, with eastward jets centred at the sharp gradients and weaker westward flows in between. In this article, we investigate the emergence of this buoyancy staircase limit in surface quasigeostrophic turbulence and we examine the dependence of the resulting dynamics on the vertical stratification. Over decreasing stratification ($\mathrm {d}N(z)/\mathrm {d}z\leq 0$, where $N(z)$ is the buoyancy frequency), we obtain flows with a longer interaction range (than in uniform stratification) and highly dispersive Rossby waves. In the staircase limit, we find straight jets that are perturbed by eastward propagating along jet waves, similar to two-dimensional barotropic $\beta$-plane turbulence. In contrast, over increasing stratification ($\mathrm {d}N(z)/\mathrm {d}z\geq 0$), we obtain flows with shorter interaction range and weakly dispersive Rossby waves. In the staircase limit, we find sinuous jets with large latitudinal meanders whose shape evolves in time due to the westward propagation of weakly dispersive along jet waves. These along jet waves have larger amplitudes over increasing stratification than over decreasing stratification, and, as a result, the ratio of domain-averaged zonal to meridional speeds is two to three times smaller over increasing stratification than over decreasing stratification. Finally, we find that, for a given Rhines wavenumber, jets over increasing stratification are closer together than jets over decreasing stratification.

We report flow measurements in rotating Rayleigh–Bénard convection in the rotationally constrained geostrophic regime. We apply stereoscopic particle image velocimetry to measure the three components of velocity in a horizontal cross-section of a water-filled cylindrical convection vessel. At a constant, small Ekman number $Ek=5\times 10^{-8}$, we vary the Rayleigh number $Ra$ between $10^{11}$ and $4\times 10^{12}$ to cover various subregimes observed in geostrophic convection. We also include one non-rotating experiment. The scaling of the velocity fluctuations (expressed as the Reynolds number $Re$) is compared to theoretical relations expressing balances of viscous–Archimedean–Coriolis (VAC) and Coriolis–inertial–Archimedean (CIA) forces. Based on our results we cannot decide which balance is most applicable here; both scaling relations match equally well. A comparison of the current data with several other literature datasets indicates a convergence towards diffusion-free scaling of velocity as $Ek$ decreases. However, at lower $Ra$, the use of confined domains leads to prominent convection in the wall mode near the sidewall. Kinetic energy spectra point at an overall flow organisation into a quadrupolar vortex filling the cross-section. This quadrupolar vortex is a quasi-two-dimensional feature; it manifests only in energy spectra based on the horizontal velocity components. At larger $Ra$, the spectra reveal the development of a scaling range with exponent close to $-5/3$, the classical exponent for inertial range scaling in three-dimensional turbulence. The steeper $Re(Ra)$ scaling at low $Ek$ and development of a scaling range in the energy spectra are distinct indicators that a fully developed, diffusion-free turbulent bulk flow state is approached, sketching clear perspectives for further investigation.

The tidal flow through a channel connecting two basins with different tidal regimes can be optimally controlled by means of a turbine fence or array to maximise the extracted mechanical power. The paper gives the optimal control strategy as a function of the blockage ratio $\sigma$, i.e. the ratio of the turbine cross-section to the cross-section of the local passage of a turbine. The results presented are a physically consistent generalisation of the results of Garrett & Cummins (Proc. R. Soc. Lond. A, vol. 461, 2060, pp, 2563–2572), valid only for $\sigma =1$ and turbine efficiency of one, now for arbitrary blockage ratio $0 < \sigma \leqslant 1$. Published research over the past decade on the same topic has taken the momentum equation and the turbine drag force as a starting point. The new approach presented here, in contrast, takes the energy equation as the starting point and uses the relative volume flow as the control variable. As the work shows, this new approach has three advantages. First, starting with the energy equation allows us to derive an optimal flow control problem resulting in an Euler–Lagrange equation using the physically consistent and experimentally validated actuator disk model for the free surface flow of Pelz et al. (J. Fluid Mech., vol. 889, 2020) in a direct and formal way. The optimal control problem is solved (a) numerically and (b) analytically. In the latter case, the turbine characteristics are approximated by a rational function in the relevant design and operating range. The analytical solution (b) validated against the numerical solution (a) is surprisingly concise and easy to apply in practice, as shown by use cases. Second, instead of the induction factor, we use the volume flow that is the same for all turbines in a cascade, i.e. a row of turbines in the direction of flow, which significantly reduces the complexity of the optimal control task of turbine arrays. Third, we obtain a well-founded energy estimate, whereas previous methods overestimate the energy yield due to inconsistent turbine disc models (for the consistency and valid parameter ranges of different models, also in comparison with experiments, see Pelz et al., J. Fluid Mech., vol. 889, 2020). The results can be used for the conceptual design of turbine arrays, but also for a sound physically realistic and consistent resource assessment of tidal power for a system consisting of two basins, a channel and a turbine fence with $0<\sigma \leqslant 1$ and operated in a complete tidal cycle.

An artificial neural-network-based subgrid-scale (SGS) model, which is capable of predicting turbulent flows at untrained Reynolds numbers and on untrained grid resolution is developed. Providing the grid-scale strain-rate tensor alone as an input leads the model to predict a SGS stress tensor that aligns with the strain-rate tensor, and the model performs similarly to the dynamic Smagorinsky model. On the other hand, providing the resolved stress tensor as an input in addition to the strain-rate tensor is found to significantly improve the prediction of the SGS stress and dissipation, and thereby the accuracy and stability of the solution. In an attempt to apply the neural-network-based model trained for turbulent flows with a limited range of the Reynolds number and grid resolution to turbulent flows at untrained conditions on untrained grid resolution, special attention is given to the normalisation of the input and output tensors. It is found that the successful generalization of the model to turbulence for various untrained conditions and resolution is possible if distributions of the normalised inputs and outputs of the neural network remain unchanged as the Reynolds number and grid resolution vary. In a posteriori tests of the forced and the decaying homogeneous isotropic turbulence and turbulent channel flows, the developed neural-network model is found to predict turbulence statistics more accurately, maintain the numerical stability without ad hoc stabilisation such as clipping of the excessive backscatter, and to be computationally more efficient than the algebraic dynamic SGS models.