Aerosol Properties & Processes: Technical Reference

Flow modeling of the pumped counterflow virtual impactor (PCVI)

This technical reference is intended for users of the pumped counterflow virtual impactor and similar devices (Boulter et al., 2006).

The pumped counterflow virtual impactor (PCVI) is designed to separate large aerosols from an air stream and place them in a separate gas flow. Small aerosols and the original air are exhausted to a pump. So far, the main application has been separating ice crystals from aerosols for laboratory or field measurements. If the flow of added gas is warm and dry then the ice crystals evaporate and the residues are available for analysis (e.g. Cziczo et al., 2003; DeMott et al., 2003; Cziczo et al., 2006; Gallavardin et al., 2008)

Flow modeling of the PCVI was conducted by Mr. Afchine during an extended visit to the Chemical Sciences Division. Software and computer resources were provided by Forschungzentrum Juelich. The PCVI has several characteristics that make the modeling difficult and computer intensive. Flow velocities can be large enough to require compressible flow. The flow geometry is complicated and has large aspect ratios that are a challenge for mesh generation.

Flow modeling was performed in three dimensions using ANSYS CFX version 11. The mesh included approximately 2.5 million nodes. The flow field was calculated by solving the Navier-Stockes equation for a steady state, compressible, and turbulent flow. The Shear Stress Transport (SST) k-w based turbulence model was used, which is more accurate in solving stagnation flow and flow with separation. The flow field was calculated, then particles were injected into the assuming that they have no influence on the flow field because of their low concentration. Particles sizes from 1 to 5 microns were examined. The Schiller-Naumann drag force model was used. Additional forces on the particles are the turbulent dispersion forces in regions where the turbulent viscosity ratio is above the value 5. All particles were lost upon a wall collision with no bounce.

Refer to Boulter et al., 2006 for definitions of the termed input, output, pump and add flows as well as details of the flow geometry and experimental tests of the PCVI.

Flow modeling of the PCVI
Results & Interpretation
Possible Improvements
Other Modeling
Image listing
low flow results
high flow results
Armin Afchine
ICG-I: Stratosphare
Forschungszentrum Juelich
Juelich 52428, Germany

Daniel M. Murphy
Aerosol Properties & Processes
NOAA Chemical Sciences Laboratory
Boulder, CO 80305 USA

Results and Interpretation

Flow modeling was performed for flows of 2.7 and 5.13 standard liters per minute (slpm) at a pressure of 400 mbar and temperature of 298 K. The add flows were 1.2 slpm for both cases and the exhaust flows were 3.6 and 6.84 slpm. These flows represent a fairly low-speed and medium-speed cases with Mach numbers at the tip of the inlet of about 0.25 and 0.55, respectively. A variety of cross sections of various flow parameters are appended to this description. Most results were common to both flow cases. View higher resolution images.

    2.7 slpm vector flow plot 5.13 slpm vector flow plot
  1. The flow in the counterflow region (where the flow direction is opposite the direction of both the input and output flows) is well behaved in both cases. The stagnation planes are reasonably flat. These features can be seen in the vector flow plots.
  2. low flow particle trajectory low flow particle trajectory

    high flow particle trajectory high flow particle trajectory

    high flow streamlines input

  3. The cut points were sharp in this modeling, although there was no attempt to determine just how quickly the transmission changed through the cut point. One may compare the 3 and 4 μm particle trajectories for the low flow case and the 2 and 3 μm trajectories for the high flow case. In both cases modeled transmission goes from 0 to 100% within a 1μm increment in diameter. As expected, the cut point decreased with increasing flow. The gas phase separation seems to be very good: when streamlines are labeled on the input flow none of them appear in the output stream, meaning the output is composed only of the added gas to a high approximation.
  4. low flow particle trajectory low flow particle trajectory low flow particle trajectory

    low flow highly turbulent exhaust high flow highly turbulent exhaust

  5. Originally there was hope that the device could be operated as a dichotomous sampler: the large particles would go into one flow and the small particles could be sampled from the pump flow. The flow modeling makes clear that this is not possible because particle losses are extremely high in the pump flow. This is especially true just below the cut point. For example, comparing the 1, 2, and 3 μm particle trajectories in the low flow case shows that an increasing fraction of the particles are lost with size until just below the cut point essentially all of the particles are lost within the PCVI. As can be seen from these particle trajectories, flow in the exhaust regions of the PCVI has a strong curvature resulting in losses much as in a cyclone. In addition, flow in the exhaust regions is highly turbulent.
  6. high flow particle trajectory
  7. There are losses of large particles by impaction on a step in the inlet diameter. These losses are already significant at the cut point. This is probably one reason that the experimental transmission above the cut point never reached unity (Boulter et al., 2006). For example, look at the trajectories hitting the step for the 3 μm particles in the high flow case.
  8. high flow particle trajectory high flow particle trajectory high flow particle trajectory
  9. The modeling supports the speculation in Boulter et al. (their Figure 3) that unexpectedly large cut points for high-pressure operation may be due to large particles not being fully accelerated in the input nozzle. This may be seen by comparing the velocity scales for 1, 3 and 5 μm particles. The 5 μm particles are only being accelerated to about 70% of the velocity of the 1 μm particles.
  10. high flow particle trajectory high flow streamlines output
  11. There is recirculation in the output flow induced by diameter changes in the output flow, even though those changes are beveled at about 45 degrees. This can be seen in both particle trajectories and gas streamlines. A few particles almost get back to the stagnation region.
  12. low flow particle trajectory high flow particle trajectory
  13. Particles just larger than the cut point are strongly focused into a small portion of the output flow, then are steered by the recirculating flows in the output. This steering seems to be stronger for the higher flow case. The recirculation zones mentioned above are not stable with time. The flow model results may be thought of as a snapshot in time. The result will be that on a cross-section through the output flow the spatial distribution of transmitted particles will be highly non-uniform and the spots of high concentration will move with time. A mixing region downstream of the outlet is essential for consistent results. In our laboratory work, downstream mixing was difficult to achieve for pressures below about 100 mbar.
  14. high flow streamlines add flow high flow streamlines output

  15. An interesting phenomenon is that as the "add" flow splits into going to the output and pump flows there is a high degree of spatial coherence in the flow. Along this particular viewing plane, all of the output streamlines come from the upper half of the add flow tubes.
  16. Listings of the full set of images for low flow results and high flow results are available.

Back to top

Possible improvements

The modeling here suggests some improvements on the PCVI design:

On the other hand, the losses of particles in the pump line are so large that it seems that minor design changes are unlikely to improve particle transmission enough to allow quantitative sampling of particles smaller than the cut point in the pump flow.

Other modeling

This modeling is not complete. Some important effects that need to be investigated are near-sonic flows, the sharpness of the cut point, and the effects of misalignment from machining tolerances. Some of these have been modeled by Mikhail Pekour, Gourihar Kulkami, and Daniel Cziczo at Pacific Northwest Laboratory.

Another topic for future modeling is investigating the details of how the add flow is inserted in order to try to minimize the cut point. The existing design has outstanding rejection of particles smaller than the cut point, but it would be desirable to reduce the cut point. Could the length of the counterflow region be reduced (yielding a smaller cut point) without allowing transmission of a few smaller particles?

Back to top


Boulter, J. E., D. J. Cziczo, A. M. Middlebrook, D. S. Thomson, and D. M. Murphy, Design and performance of a pumped counterflow virtual impactor, Aerosol Sci. Technol., 40, 969-976, 2006.

Cziczo, D. J., P. J. DeMott, C. Brock, P. K. Hudson, B. Jesse, S. M. Kreidenweis, A. J. Prenni, J. Schreiner, D. S. Thomson, and D. M. Murphy, A method for single particle mass spectrometry of ice nuclei, Aerosol Sci. Technol., 37, 460-470, 2003.

Cziczo, D. J., D. S. Thomson, T. L. Thompson, P. J. DeMott, and D. M. Murphy, Particle Analysis by Laser Mass Spectrometry (PALMS) studies of ice nuclei and other low number density particles, Int. J. Mass Spectrom., 258, 21-29, 2006.

Gallavardin, S. J., K. D. Froyd, U. Lohmann, O. Moehler, D. M. Murphy, and D. J. Cziczo, Single particle laser mass spectrometry applied to differential ice nucleation experiments at the AIDA chamber, Aerosol Sci. Technol., 42, 773-791, 2008.

DeMott, P. J., D. J. Cziczo, A. J. Prenni, D. M. Murphy, S. M. Kreidenweis, D. S. Thomson, and R. Borys, Measurements of the concentration and composition of nuclei for cirrus formation, Proc. Nat. Acad. Sci., 100, 14655-14660, 2003.

Back to top