Experimental deposition of NaCl particles from turbulent flows at gas turbine temperatures

,


INTRODUCTION
e deposition of micron-sized particulates from turbulent gas flow has been a subject of investigation for over years.Interest from within the gas turbine community has become very significant over the last years due to numerous threats to engine life.Ingested sand, dust, volcanic ash, salt, and ice crystals can drive various damage mechanisms which can substantially reduce component life [ ].Studies of deposition within secondary air systems studies have shown that significant blockage of film cooling hole can occur for high pressure turbine (HPT) blade conditions [ ].In the main gas path particle effects include fouling and erosion of compressor surfaces, and accretion onto nozzle guide vanes and HPT blades [ , ].
Computational studies of such geometries have tended to apply standard and widely-available numerical models for the interaction of particles with gas turbulence, including the discrete random walk [ ]. ese have been shown to be inappropriate for such modelling [ ]; to address this the continuous random walk model [ ] has been applied to gas turbine flows [ ]. is model has been assessed at ambient conditions [ ], and now requires validation at enginerepresentative temperatures.
is paper presents experimental data for assessment of the validity of the continuous random walk model at temperatures and Reynolds numbers representative of conditions in gas turbine secondary air systems.Horizontal pipe flow experiments are reported, using a sodium chloride (NaCl) aerosol.Tests are undertaken with both isothermal and wall-gas temperature gradient conditions in order to assess thermophoretic effects.ermophoresis is a particle force due to temperature gradients within the gas phase, proportional to the temperature gradient, in the negative direction.When T gas > T wall the rate of deposition observed is increased in comparison to isothermal conditions ('increasing' thermophoresis), for T gas < T wall the reverse is observed ('decreasing' thermophoresis).To the authors' knowledge, only one study [ ] has addressed decreasing thermophoresis experimentally.

. BACKGROUND
Deposition is presented in this paper in terms of non-dimensional deposition velocity V + d , and non-dimensional particle relaxation time τ + p , In the absence of significant body forces except those due to particle-gas relative motion, particle motion can be divided into three regimes,  [ , ], can be seen to have higher deposition rates in the diffusional and diffusion-impaction regimes, and a reduced dependency of V + d on τ + p .Based on expected particle sizes ( .µm) and secondary air systems conditions (T g , ρ g , u * ), particle τ + p values are expected to fall into the diffusion-impaction and inertia-moderated regimes.In the diffusion-impaction regime, particles have enough inertia to be effected by turbulent eddies, which can give the particles sufficient wall-normal velocity to deposit onto boundary surfaces.A strong dependence on τ + p is seen.In the inertia-moderated regime particle response to turbulence fluctuations decreases as particle inertia becomes to large to be effected by all but the largest eddies.
e majority of particle deposition experiments applicable to gas turbine engines have assessed main gas path geometries [ , ], using engine(-scale) components at enginerepresentative temperatures.
ese studies have shown a significant particle temperature dependency on deposition rate; above an adherence or so ening temperature threshold the rate of deposition has been shown to increase very significantly.ese findings corroborate a number of studies of deposition in secondary air systems; [ , ] investigated sand ingestion and blocking of film cooling holes.All reported increased rate of hole blockage with increasing metal temperature above a threshold and varying composition effects.
e more complex geometries undertaken by the above studies make it hard to decouple flow-related and geometryrelated effects on deposition.By removing much of the geometric complexity, the experimental results presented here a empt to address the underlying physics of the problem.

. THERMOPHORETIC PARAMETER
A new thermophoretic parameter is derived below.Initially we take a force balance on an aerosol particle suspended in a gas, giving net acceleration, Note that forces due to Brownian motion and li are neglected due to large relative particle size and high particle to gas density ratio respectively.Here the drag component of acceleration is given by where particle Reynolds number Re p is defined by e drag coefficient is defined by as the particle to gas density ratio is so high (ρ p /ρ g ∼ 1 000).e thermophoretic acceleration, a T h , is experienced by particles proportional to the negative direction of temperature gradient, Here ∇T g is the gas temperature gradient and Φ the thermophoretic coefficient (Φ = Φ (Λ, Kn)).
e thermal conductivity ratio Λ is defined Λ = k p /k g , where k p and k g are the particle and gas (translation component only, k g = 15/4µR) thermal conductivities.Knudsen number is defined Kn = 2l g /d p , where l g = µ π/ 2P g ρ g is molecular mean free path of the gas, and d p particle diameter.In the case of hot gas, cool wall, an increase in particle deposition rate is seen ('increasing' thermophoresis).In the reverse case e expression of Beresnev and Chernyak was derived from an analysis based on the S-model, a higher-order linearisation of the Boltzmann equation collision integral than the (more widely used) BGK equation.e thermophoretic coefficient is dependent functions f k,1 , for k = 1 − 4, Functions f k,1 are tabulated in the original paper for varying R BC (= √ π/2Kn).Use of the expression necessitates interpolation of Φ from the given values of R BC .
We define a normalised thermophoretic parameter P + th to a empt to collapse the deposition curves at varying thermophoretic conditions.Starting from the equation for thermophoretic acceleration, Eq. , this can be wri en which is non-dimensionalised by (u * ) 2 /d p , is can then be expressed in terms of τ + p and the Epstein number, E p = d p ∇T g /T g , is is plo ed against normalised deposition fraction, f d , the ratio of deposition fraction for non-isothermal to isothermal matched tests.All test conditions (Re, d p , T g ) are kept constant between the matched isothermal and nonisothermal tests, with the exception of the wall temperature, which is changed, Deposition fraction is calculated from where m d is the measured deposited mass during a test, and m in the the injected mass.For T gas > T wall we expect f d > 1; the reverse for T gas < T wall .If f d is correlated against P + T h , for a given aerosol, the change in deposition fraction with thermophoresis from isothermal conditions can be calculated.

. METHODS . Experimental design
An experimental rig was built in a horizontal pipe flow configuration, Fig. .Gas was compressed air, pre-dried and supplied by the laboratory barg line.Heating of the gas was achieved using inline pipe heaters (Omega AHPheaters, rated to • C, max flow rate l/min), gas path (GP ) and a section wrapped in heater tape (Omega Ultrahigh temperature heater tape, STH -, W, rated to • C).Gas path (GP ) passed through the atomiser, carrying the generated particles into the mixing chamber.Gas path (GP ) enabled the atomiser to be bypassed during the heating processes.
Heating of the test piece body was undertaken using a Wild Barfield oven.Flow regulation and measurement was done using a pair of mass flow controllers (FMA-A, -SLPM (MFC ), and FMA-A, -SLPM (MFC )) and an orifice plate, calibrated using one of the mass flow controllers.Gas velocities ranging -m/s and Reynolds numbers 6 500 − 10 000 were achieved, corresponding to friction velocities .-. m/s.Gas and metal/wall temperatures up to • C and • C respectively were achieved.• , giving a deposition area of mm 2 .A D h distance existed between mixing chamber and test insert to allow the flow to fully develop.Full rotation about the longitudinal axis was achievable in order to assess gravitational effects.e test insert was clamped into the test piece using an array of six clamps, and sealed using a metal gasket (James Walker, Supagraf Laminated S ).
Two test pieces, EXPα and EXPβ, were manufactured from stainless steel grades and respectively.Internal geometries and instrumentation were identical.Surface roughness profiles were measured using a Taylor Hobson surface profiliometer, 'Talysurf', probe tip diameter .µm, and a Gaussian high-pass filter with cut-off wavelength .mm applied to remove waviness.Roughness is characterised by the arithmetic mean roughness R a .Calculated surface roughnesses were EXPα: R a = 0.22 µm, EXPβ: R a = 1.12 µm.e difference likely occurred due to different post-experiment cleaning methods.Expected engine component finishes of .-. µm (machined) and ∼ 1.4 µm (shot peened) indicate that a comparison between deposition on test pieces EXPα and EXPβ is relevant to engine conditions.
Instrumentation was in the form of three axial thermocouples measuring test piece body temperature (T B1−3 ), and a fourth to measure gas temperature mm downstream of end of the test insert (T TS ).A pressure tapping in the test piece body wall at the upstream edge of the test insert gave test section static pressure P TS , and could also be used to calculate test section friction velocity (Sensortronics,mbar)).
A diverging-straight-converging mixing chamber was designed for the introduction of the particulate laden flow to the heated gas.e mixing chamber was instrumented with a static pressure tapping (P MX ), and a thermocouple upstream of the particulate injection (T MX ).

. Particle generation
NaCl particles were produced from solution of NaCl (Sigma Aldrich BioXtra, S ) at ambient temperature and mixed with the heated flow.Particle diameters .-. µm were generated, giving non-dimensional particle relaxation times τ + p of .-. at ambient conditions, and .-. at the high temperature conditions.NaCl was chosen due to its close relation to sea salt, with the advantage that its single compound nature allowed simpler measurement of the deposited mass.Its high melting temperature, • C, made possible to use at engine-like conditions, unlike the majority of aerosol particulates generated from solution.
A TSI model VOAG (vibrating orifice aerosol generator) was used to provide the NaCl particles.e atomiser pumps saline solution through an orifice, which is oscillated using a signal generator and piezoelectric crystal.Tuning of the vibrational frequency to the volumetric flow rate of the solution and droplet diameter can produce mono-disperse droplets.A single droplet is produced per cycle, which are dispersed into a drying column where the solvent evaporates, leaving solid NaCl aerosol particles.e final particle diameter is dependent on solution volumetric concentration of NaCl.Particles are assumed close to spherical as were seen to be circular when viewed under a microscope.
An optical particle sizer (OPS), TSI model , was used to provide both particle size distributions and bulk concentration measurements.e OPS uses a laser sca ering/shadow principle to measure optical (rather than aerodynamic equivalent) size.e OPS samples .l/min air, reporting the size distribution and total count.e total number of particles in the bulk flow is then calculated by multiplying by the actual volume flow rate of the flow from which the sample is drawn.
e OPS unit was used to assess the nature of the particle size distribution; distributions with a geometric standard deviations of 1.25 were produced.

. Deposition measurement
A new technique was developed for the measurement of deposited NaCl.
e electrical conductivity, measured in S/cm, or more commonly µS/cm, of a NaCl-water solution is linearly dependent on NaCl concentration; upon dissolving into solution, the Na + and Cl − ions disassociate.e internal surfaces of test insert and test section were cleaned repeatedly using a swab, which was rinsed in deionised water.
e conductivity of this NaCl-deionised water solution was measured, from which the mass of deposited salt was calculated, based on solution mass and a calibration of solution concentration against conductivity.Solution conductivity is dependent on temperature, which is calibrated to a datum value ( • C) using temperature compensation coefficient α.
Calibration of the electrical conductivity meter cell constant using standard solutions (potassium chloride, KCl) of known conductivity (Hanna Instruments HI-, HI-).Manual calibrations of solution concentration against conductivity, temperature compensation coefficient, and swabbing efficiency were carried out.Solution conductivity was indeed seen to be strongly linearly correlated to NaCl concentra- Swabbing efficiency η s , the ratio of measured fraction to a known deposited mass, was found to have a mean of ., with standard deviation .
(nine trials).A Kern KB--N balance was used to measure solution mass.

. Experimental uncertainty
A list of uncertainties in the individual measurements taken in each test is given in Table .e 'source' column indicates the origin of the value: manufacturer-specified (m), manually calibrated (c), assumed (a).

Item
Measurement accuracy/uncertainty Source where U Xi /X i is the relative uncertainty in variable X i , and (X i /R) ∂R/∂ X i the uncertainty magnification factor.For V + d and τ + p , this leads to total uncertainties of .% and .% respectively.
. RESULTS AND DISCUSSION .Ambient temperature experiments A number of ambient temperature tests were carried out to demonstrate the experiment was operating in a comparable manner to other published data, and to investigate the effects of d p , Re, and surface roughness R a on deposition rate.e experimental conditions are summarised in Table .e experimental data are plo ed by particle diameter and bulk Reynolds number in Fig. .e green region represent the data of Kvasnak et al. [ ], who carried out deposition experiments in a horizontal channel flow configuration at ambient conditions, with a measurement surface made 'sticky' to retain all impacting particles with the application of a thin layer of freon.Regarding the current experiment, some sca er of the data points is seen, though this is almost OPS bins particles by size.Bin width varies based on particle size.We used the mean bin particle diameters for this study: ., ., ., ., ., ., ., .µm.universally the case with particle deposition experiments, and mostly within the spread of published data.
Two strong trends are seen.For τ + p ≤ 8 the experimental data follow the Kvasnak et al. data   .High temperature experiments e high temperature experiments were carried out in the same manner as the ambient temperature campaign, but with gas and metal heated to two temperature conditions representative of the intermediate pressure turbine secondary air system.ese are referred to as 'T ' (lower gas temperature) and 'T ' (higher gas temperature); experimental conditions are outlined in Table .e high temperature isothermal data are presented in . .

Table .
Operating conditions for high temperature experiments.
discriminated by d p and Re it is seen that again this is due to increasing Re.Both the Re = 10 000 data points show this effect, V + d reducing compared to the same particle size at a lower Re.e two points at τ + p = .make a useful comparison: d p = µm, Re = 10 000 (yellow square), and d p = µm, Re = 7 800 (turquoise triangle).e higher Re case reduces V + d by 46% in comparison to the lower Re case.A number of papers on particle deposition have proposed or accepted Reynoldsindependence of V + d , for example Liu and Agarwal [ ].It appears that V + d may not be Re-independent when a dry/nonsticky boundary condition is in place.
Non-isothermal experiments were undertaken with temperature pairings A,B,D,E as outlined in Table .Gas temperatures were kept constant for T ( • C), T ( -. E -. -. Table .Gas and wall temperature pairings A-E for high temperature tests.Indicative P + T h values shown for the tabulated gas temperatures with d p = µm, Re = (T ), Re = (T ).Temperature conditions T D and T E were undertaken at temperature gradients which gave decreasing thermophoresis.T D shows significant reduction in V + d in comparison to the isothermal data.At low τ + p (< 2.5), V + d stays fairly constant with τ + p , then increases for τ + p > ., tending back towards the isothermal data line.e effect of varying d p , Re at constant τ + p indicated a reduction in V + d of % between the d p = µm, Re = 7 800 and d p = µm, Re = 10 000 cases. is is similar to the reduction at isothermal conditions (46%).
e T E experiments used a larger temperature gradient than T D ; these show a similar trend to the T D experiments but of larger magnitude.A larger reduction of 68% between the two varied d p , Re experiments at constant τ + p is seen for this higher temperature gradient.e T experiments were undertaken at higher gas and wall temperatures than T .ey show a similar trend to T ; although the T C (isothermal) and T D data points sit higher than the equivalent cruise points, the reduction in deposition fraction is comparable.
e T E experiment showed extremely low deposition, close to the detection limit of the conductivity meter.
Log of normalised deposition fraction is plo ed against thermophoretic parameter P + T h for the six thermophoretic conditions in Fig. .Particle thermal conductivity is k p = 4.9 W/m/K for T g > 120 • C. A first order correlation by linear regression is calculated, Log of normalised deposition fraction ln f d against thermophoretic parameter P + T h .
e linear fit is good for both increasing and decreasing thermophoresis, with more sca er seen for P + T h < −3 × 10 −6 .e sca er is likely due to these points representing experiments undertaken with a high, decreasing thermophoresis, where the deposited mass of NaCl was low.

. CONCLUSIONS AND SUMMARY
A new experimental rig has been built, and an experimental campaign carried out to assess the deposition characteristics of NaCl particles at gas and metal temperatures representative of a gas turbine engine secondary air system.A pipe flow geometry has been utilised to allow comparisons to be made to other existing deposition data at ambient conditions.As the experimental data is for the assessment and validation of numerical models, this geometry is a suitable first step.A method to measure the deposition based on solution electrical conductivity has been developed and validated.Experimental uncertainty has been addressed.e initial ambient temperature campaign demonstrated that for τ + p < 8, V + d behaves as would be expected in comparison to other ambient measurements.For τ + p > 8 a reduction in V + d is seen. is is thought to be related to some particles rebounding or being removed upon impact, and is noted mainly for the highest nominal particle kinetic energy cases.Due to the experimental set-up we were not able to assess this effect further during the test campaign.A study where NaCl particles were impacted at known angles onto a plate, and the rebound characteristics measured (for example [ ]), could be carried out to investigate this effect.e effects of two engine-representative roughnesses on deposition were assessed; increasing surface roughness at this scale (R a < d p ) was seen to increase deposition by a mean of 1.2×.
At high temperatures, the isothermal tests showed similar trends to those at ambient conditions; at low τ + p the values are well-matched, with a similar tail-off of V + d occurring for τ + p > 4. Both thermophoretic directions (increasing and decreasing deposition) are observed; for T gas < T wall very substantial reductions are seen in deposition velocity.e development of a thermophoretic parameter showed that for there existed a linear relationship between ln f d and P + T h for the majority of the experimental data.ese provide a large data set for assessment of models for particle -turbulence interaction in numerical simulations.

Figure .
Figure .Summary of some key experimental data for particle deposition in turbulent vertical pipe (markers) and horizontal channel (green region) flows at ambient conditions.Adapted from [ ].

Figure .
Figure .Schematic of experimental rig

Figure .
Figure .Experimental test piece with test insert removed

Figure .
Figure .Measured solution conductivity dependence on NaCl concentration.Plo ing axis reversed as experimental analysis requires C m = C m (G).Conductivity at • C.

Figure .
Figure .V + d against τ + p for all ambient temperature tests, coloured by nominal particle kinetic energy E k, p .Lines show theoretical deposition curves for coefficient of restitution r = 0.0, r = 0.5, r = 0.96 from [ ].
Fig. , and compared to the ambient temperature experiments.It can be seen that the data are very comparable to those at ambient temperatures in general.For τ + p > 4 the hot V + d values are in general lower than the ambient experiments at similar τ + p .A similar tail off appears in V + d for larger τ + p akin with the ambient temperature experiments.As the data

Figure .
Figure .Comparison of all isothermal hot data with ambient temperature experiments.

Table .
Operating conditions for ambient temperature experiments.
• C) conditions.e majority of experiments were undertaken at T conditions, varying τ + p by changing d p and Re.Results for the thermophoretic experiments are shown in Fig. .Tests T A and T B were carried out with increasing thermophoresis, T g > T w , at Re = 7 800.Substantial