University of Twente Student Theses

Login

Exploratory CFD research on a Coriolis mass flowmeter

Burghout, R.A.B. (2016) Exploratory CFD research on a Coriolis mass flowmeter.

[img] PDF
Restricted to Restricted

8MB
Abstract:The outcome of the three main objectives of the exploratory CFD research on a Coriolis mass flowmeter is described briefly here. The first goal was to generate a FSI model representing a U-shaped flowmeter and try to validate the accuracy of this model by comparing the numerical results with the available literature. A FSI model of the U-shaped flowmeter was made in Ansys. It turned out that two different U-shaped geometries were used in the available literature. For one geometry the first natural frequency was provided, for the other geometry the phase difference as function of the mass flow was provided. The small difference in geometry was only noticed during the last week of the internship, because then I searched for the cause why there was still a difference in the phase difference as function of the mass flow that resulted from Ansys, compared to the phase difference as function of the mass flow stated in the literature. A new FSI model of the U-shaped flowmeter was then quickly made in Ansys and two "final simulations", with refined settings, were executed during the final days of my internship period. It was expected that due to this improvement the existing deviation in the phase difference as function of the mass flow would become much less, but this turned out to be not the case. The results from the two "final simulations" in Ansys resulted in a phase difference as function of the mass flow that is 21% higher than stated in the literature. The conclusion is thus that the calculated phase difference is very inaccurate. Many possible causes of this deviation are mentioned in this report. Further investigation has to be done to find the root cause. The second goal was to find out what the best way is to setup a FSI model and how the acquired data can be processed efficiently and accurately. This has been investigated. The best way to setup a FSI model in Ansys can be read in the report, because it is not possible to summarize all steps here. To retrieve the phase difference out of the sensor data that follows from Ansys, it was determined that the best way to do this is with the use the Fast Fourier Transform in Matlab. The table with displacement results at the sensor positions from Ansys has to be copied into an Excel file. This Excel file is then loaded into Matlab. To use the Fast Fourier Transform correctly in Matlab, the time range with displacement data should contain exactly an integer amount of oscillations. The timestep in Ansys is often a nice rounded number like 0.0001s, 0.0002s, 0.00025s or 0.0005s. So when this timestep is known, the applied frequency at the actuator position should then be chosen in such a way that the selected time range, which will be used for analysis in Matlab, consist out of an integer amount of timesteps and also an integer amount of oscillation periods. The startup effect of a simulation should also be outside this timerange. If all this is not the case, then the Fast Fourier Transform in Matlab will give inaccurate results. The applied frequency in Ansys will thus probably always differ from the actual natural frequency, because a nice rounded frequency (often a frequency that can precisely be divided by a number like 2, 3, or 5) has to be chosen. The current requirement to use a nice rounded frequency can probably be eliminated when the Matlab script is improved, because actual flow meters are also able measure the phase difference when the flow meter oscillates in a natural frequency that is not precisely a nice rounded frequency. So it has to be investigated how this also can be realized with Matlab. The third goal was to generate a FSI model representing a flowmeter of DEMCON and investigate the applicability of FSI analysis on this large and complex geometry. A FSI model of the "Bronkhorst geometry" was made in Ansys. A transient simulation of this model was started when I left Demcon at the end of my internship period. The estimation was that the complete transient solution of the Ansys Bronkhorst model would be available after a couple of weeks. However, the simulation crashed before the simulation was completed due to removal of my personal account at Demcon. The solution is only available for the first 0.05 sec, which is the part that contains a startup effect. Therefore no phase difference as function of the mass flow could be determined based on this short time range. However, based upon the time that it took to solve the first 0.05 sec, an estimation could be done about how long a simulation of 0.3 sec would take. 0.3 sec is probably the total solution time that is needed in order to have a time range with a length of 0.1 sec up to 0.2 sec, that doesn’t contain the startup effect. This time range is then used in Matlab to retrieve the phase difference. The run of the Bronkhorst model is based on a relatively coarse mesh, a laminar model, a timestep of dt = 0.0005s and a RMS value of 0.01. Solving the first 0.05 sec took 57 hours on 6 CPU. A solution time of 0.3 sec for the Bronkhorst geometry would thus mean a solution time of 2 weeks on 6 CPU. The solution will probably not be converged with these settings. So a simulation with a much stricter RMS convergence value would take even longer, for example a month on 6 CPU. Running a simulation with a turbulence model, thus therefore also a coarser mesh, would then take probably multiple months on 6 CPU. As mentioned in discussing the first goal of this internship, the calculated phase difference as function of the mass flow for the relatively simple U-shape model turned out to be very inaccurate. As long as Ansys isn’t able to provide accurate results for this simple U-shape model, then it has no use to do lengthy simulations in Ansys with the large model of the "Bronkhorst geometry".
Item Type:Internship Report (Master)
Clients:
DEMCON BuNova B.V., the Netherlands
Faculty:ET: Engineering Technology
Subject:52 mechanical engineering
Programme:Mechanical Engineering MSc (60439)
Link to this item:http://purl.utwente.nl/essays/71989
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page