7 Replies Latest reply on Feb 19, 2015 5:34 PM by Bill McEachern

    mixing efficiency - Coefficient of Variation (CoV)

    Jake Spooner

      Does anyone have any experience of using CFD results to assess the mixing efficiency in a system using a coefficient of variation (CoV) calculation?

       

      I am trying to simulate some physical lab tests which have measured dye concentrations using LIF (laser induced fluorescence) to calculate the CoV. Although hydraulic losses in the model match the physical results well, the CoV values are significantly different.

       

      I have used the point parameters function to extract detailed concentrations across the test section to carry out the CoV calculation from the CFD model. The LIF measurement uses the dye concentration of each pixel (200,000) to calculate the CoV. The CoV from the CFD model using the point parameters function to extract the data does not really vary once the number of points increases over 500-1000, although I have been using 7000 points, still significantly less than the pixel count.

       

      My thoughts are that possibly:

      - the model isn't representing the turbulence or secondary flows adequately

      - the CFD results aren't suitable for a comparative CoV calculation

       

      Any thoughts?

        • Re: mixing efficiency - Coefficient of Variation (CoV)
          Vikram Manthri

          Did you happen to solve your problem ? Were you able to match the COV results with your PLIF measurements ?

          • Re: mixing efficiency - Coefficient of Variation (CoV)
            Bill McEachern

            This isn't exactly my area but I have done comparisons for flow variations in the discrete channels of a multi channel electrostatic precipitator and things matched up quite nicely for the COV on velocity in that case. Experimentally they have one point per channels and in flow I used average velocity in each channel. In my case I had multiple simulation points for each experimental point. In the case stated its the other way round. Have you tried increasing the mesh density in your model to match or be closer to the experimental density or maybe more practically reducing the experimental density. Just a suggestion as the experimental density is very high and it may be in part a contributing factor to the mismatch. Anyway just a thought.

            • Re: mixing efficiency - Coefficient of Variation (CoV)
              Bill McEachern

              and one more thing I discovered in that electrostatic precipitator model: The client had suggested a scope of model - just the top section. However I did not get a close match. I included the whole machine and it was very a very respectable match. The lesson was make sure your scope in the simulation is adequate to capture the real flow field. The downstream resistances can not be simplified out and expect a good match. It wasn't mentioned what your model was like compared to the experimental set up but it was intimated that they are identical. If they are not you might want to make them that way and try again.

                • Re: mixing efficiency - Coefficient of Variation (CoV)
                  Vikram Manthri

                  Bill, when you calculated COV did you do the volumetric flow rate sampling ( dividing the area such that flow rate through division*avg velocity of division is constant) of the cross section. If so, can you tell me how did you do it in solidworks?

                    • Re: mixing efficiency - Coefficient of Variation (CoV)
                      Bill McEachern

                      I used goal data - I had a flow manager (inlet pipe, followed by a tapered section, a set of perforated plates, then a very short rectangular section,) to try and even it out the flow. It was  quickly followed by the main body which was composed of say call it 32 rectangular long  channels.  They all then merged in a compartment and off to s single discharge. I put a goal surface in each channel and got he avg or bulk avg velocity and used that to compute the variation. I compared it to experimental data - very close all things considered. You could do the same approach or you could get the bulk data and do it on that.