tions 5.23, 5.24 and 5.25, are determined. The following, general procedure to get a random numberxrand, distributed according to an arbitrary probability density function pdf(x), is applied:
• Determine the associated cumulative distribution function cdf(x) analytically if possible, else by numerical integration. Due to the normalisation of pdf(x), i.e. R∞
0 pdf(x)dx= 1, it follows by definition thatcdf(x)∈[0,1].
• Choose a linear distributed random number prand ∈[0,1].
• Determine xrand which fulfilscdf(xrand) =prand.
In combination with the relative mean values given in Equations 5.23, 5.24 and 5.25, the secondary diameter, the wall-normal and wall-tangential restitution coefficient of child j can be calculated as follows:
Dsec,j = Dsec,j
D10,sec · D10,sec
D10,prim ·D10,prim, cN,sec,j = vN,sec,j
vN,10,sec · vN,10,sec vabs,10,prim
· vabs,10,prim
vN,prim,i , cT,sec,j = fdev· vT,sec,j
vT,10,sec
· vT,10,sec vabs,10,prim
· vabs,10,prim
vT,prim,i
(6.1)
withfdev =
( +1 if the parcel is forward scattered,
−1 if the parcel is backward scattered. It is important that the last terms in the given equations, D10,prim, vabs,10,prim
vN,prim,i and
vabs,10,prim
vT,prim,i , must not be substituted by Dprim,i, vvabs,prim,i
N,prim,i and vvabs,prim,i
N,prim,i respectively: The model correlations have been obtained using mean impacting values and not individual ones. If those were applied, the primary distributions would be superimposed on the actual secondary distributions leading to different results.
To determine the mean primary properties, information from the previous timestep is evaluated, cf. the discussion in Section 1.4.3. Only if all spray drops experience ap-proximately the same impact conditions like in the reference case, it is possible to do the averaging over all impacting drops. Else, e.g. for the impact on an oblique plate in Chapter 6, the rotationally symmetric spray has to be divided in sectors with approx-imately homogeneous impact conditions, and the averaging has to be done for every sector separately. Depending on the sector in which a parcel impinges, different means are then used in the calculations. For very asymmetric and spatially diversified sur-faces, where it is no longer possible to do an averaging at all, the individual primary properties have to be used although this is known to comprise errors.
6.1.2 Numbers and number rates of child parcels
The number of secondary parcels generated per impact is a user-defined quantity and should be chosen not too large with respect to computational effort. In case of a mean impact angle 60◦ ≤ α10 ≤ 90◦ (α = 90◦ signifies normal impact), it is suggested to define two child parcels per impinging parcel i, i.e. Nsec,i = 2. For smaller α10, three child parcels are proposed, i.e.Nsec,i= 3: two in forward direction and one in backward direction. An adaption to other values ofNsec,i can be easily done.
The jump from Nsec,i = 2 to Nsec,i = 3 at a mean impact angle of α10 = 60◦ is rather arbitrary at the moment. It may have an influence on a graphical presentation when simulations are post-processed and a sufficient number of Lagrangian parcels is required.
It has to be stressed, however, that any model suffers from a similar problem, which is inherent to Lagrangian particle tracking where the multitude of physical spray droplets has to be represented by a small number of discrete parcels, see Section 1.3.2.
The definition of the number nsec,j of physical drops associated to every child parcel is an important aspect in the implementation of a model, because the values strongly in-fluence the diameter and velocity distributions: The number rates1 present a weighting factor of a drop value - analogously to the factors wn,i·ηval,i used in Equation 4.9 for the measured values.
Two different ways can be chosen to define the number rates, see Figure 6.1. Consider exemplarily a drop size distribution:
• Dividing the diameter range in a finite number of discrete size values/classes, the same number of parcels can be chosen for every size value/class. The number rates of the parcels then have to represent the mass/number fractions per size value/class to obtain the correct distribution.
• If the diameter values for the child parcels are chosen according to the associated distribution function, the distribution is already represented. The number rates then have to be equal for all parcels.
A mixture of both ways is in principle also possible but rather error-prone. In this work the second possibility is chosen and for each secondary direction a constant value for the child number rates is determined such that the averaged secondary to primary mass ratio from the measurements, see Equation 5.26, is accounted for:
Nsec,i= 2 :
nsec,1 = qm,sec forward
qm,prim
· mprim,parcel
msec,single drop
(fdev,1 = +1),
nsec,2 = qm,sec backward
qm,prim · mprim,parcel
msec,single drop
(fdev,1 =−1), Nsec,i= 3 :
nsec,1/2 = 0.5·qm,sec forward
qm,prim · mprim,parcel
msec,single drop
(fdev,1 = +1),
nsec,3 = qm,sec backward
qm,prim · mprim,parcel
msec,single drop
(fdev,1 =−1). (6.2)
1nprim,i andnsec,j respectively are scaled with the timestep value of the continuous phase in CFX and thus denoted also as number rate.
Figure 6.1: Illustration of the two methods to define the number rates. Consider a simplified, discrete diameter distribution with three values. On the one hand, the distribution could be presented by three parcels - one per size value - with weights (number rates) of 0.25, 0.5 and 0.25. On the other hand, it is also possible to choose four parcels - one for size D1 and D3 and two for sizeD2 - with equal weights of0.25.
The ratio between the mean primary parcel mass and the mean mass of a single sec-ondary drop can be approximated as
mprim,parcel
msec,single drop
≈nprim·
D30,prim
D30,sec 3
, (6.3)
where the evaluation of the measurement gives D30,prim
D30,sec ≈0.789. (6.4)
The averaged primary number rate can be calculated for a specific simulation as nprim= total injected mass flow
total number of Lagrangian primary parcels·π/6·ρ·D330,prim . (6.5) It is also possible to evaluate nprim directly from the previous timestep like the value D30,prim if unknown2.
Remarks on Kuhnke’s and Roisman’s/Horvat’s model In these models, de-scribed in Section 1.4.2, diameters of secondary parcels are chosen in the same way as in the new model, i.e. random numbers distributed according to the respectively given distribution function are used. As discussed above, number rates then have to be set to a constant value (which has to be determined in such a manner that the overall secondary to primary mass ratio is respected). Yet this is not done. Instead a rather intuitive but imprecise procedure is carried out in both models:
• In Kuhnke’s model, see Equation 1.83, the child number rates are determined as nsec,j = qm,sec
qm,prim ·mprim,i·F . (6.6)
2The experiment yields D30,prim ≈52.9µm.
mprim,i =π/6·ρ·D3prim,i·nprim,i is the total mass of the impacting parcel andF is a function which divides the secondary mass between the child parcels generated for the considered impact
F = pdfWeibull(Dsec,j) P
k=1,2,3D3sec,k·pdfWeibull(Dsec,k). (6.7) Even if the primary spray was set up in such a way that all primary parcels carried the same massmprim, the functionF would not be constant for all parcel impacts, because it depends on the randomly chosen values of the secondary diameters. Also the secondary to primary mass ratio qm,sec/qm,prim is calculated using a random number, see Equation 1.79. The secondary number rates are therefore not set to a constant value.
• In Roisman’s/Horvat’s model, see Equation 1.100 the child number rates are also defined as in Equation 6.6. The function F is different but depends on the secondary diameters, which are random numbers distributed according to the size distribution, too. Analogously to Kuhnke’s model, the values of nsec,j are thus not constant.
Summarised, the definition of the child number rates is arguable in both models: If the parcels were considered with equal weights, the distributions of the drop diameters and velocities would be correctly represented, because the values are chosen randomly according to the distributions. Yet, the distributions must not be evaluated for the parcels but for the actual drops, i.e. every parcel has to be weighted with its number rate. Since these are not constant in this case, the distributions are incorrectly deformed.