Despite the severe delays caused by the Covid19 pandemic, the last batch of experimental measurements was delivered on schedule to the partners of the ENERXICO project. The first data set was uploaded to the cloud, so that all collaborators could retrieve it at their own leisure. These data sets correspond to the multiphase flow experiments conducted at the Instituto de Ingeniería, UNAM. On this second stage of the experimental program, we focused on the flows processes involving high-viscosity mixtures of liquids and gases, which are at the heart of some flow assurance problems arising in the oil & gas arena.
The experiments were conducted with different combinations of liquid and gas flow rates. In total, 25 flow mixtures were included in the experimental matrix. Nearly 300.000 measurements were registered for the high-viscosity regime. Previously, the complementary set of experiments in the low-viscosity limit generated around 600.000 data points. The obtained data is being pre-processed for quality control. Upon verification, the accepted time-series undergoes a statistical analysis designed to identify patterns in the data. This statistical analysis is based on spectral decompositions, sample entropy evaluations and cross-correlations, among other procedures.
Besides the forgoing laboratory tests, a numerical model was prepared to set a performance reference for the computations done with other codes, such as SPH or Alya. The benchmark code created in OpenFOAM by Valente Hernández successfully simulated various multiphase flows with low-viscosity mixtures (i.e. with water and air). Figure 1 illustrates the arrangement of the phases along the pipeline. Instead of tackling the complex problem of hydrodynamic slugging from the start, the inlet condition was set to produce a pulsating flow as a first approximation. In spite of this modeling simplification, a proper slugging initiation was achieved through the growth of unstable interfacial waves on an initially stratified flow. The simulations show that these wave fronts continue to grow until the entire cross-section of the pipe is blocked, which is in full agreement with all previous knowledge. The long-term evolution, however, is still a matter of debate and motivates further research.
Since the simulations reproduce the well-known dynamical properties of this flow pattern, the next step is to model the process in the high viscosity regime. The validation with the experimental data will then allow an adequate assessment of the code’s reliability. Such validations are essential since the way in which the inlet conditions are specified have a definitive influence on the development of the flow along the pipe. Other specific technical details, like the use of mixed or separate phases, will also be addressed. For instance, it has been observed that the use of mixed phases favours the formation of wavy flow, while the use of separate phases leads to a fluctuating state of the relevant fields that favours an early appearance of the slug flow.
Figure 1. Numerical simulations. Phase contours produced by a slug flow in a relatively-long horizontal pipe. The low viscosity flow is constituted by an air-water mixture (red corresponds to the gaseous phase and blue to the liquid phase). The formation of the slug units is evident in this sequence of snapshots. The superficial velocities are 0.75m/s and 1 m/s for the water and the air, respectively.
These flow structures are similar to the ones observed in the CT Scans. Figure 2 depicts the sequence of tomograms showing the slug units that result with three different flow conditions. From top to bottom, the liquid flow rate is increased while the gas flow rate remains constant. An obvious challenge for any code is to correctly reproduce these sequences for the known inlet and outlet conditions, as well as fluid properties.
(a) Low liquid and gas flow rates
(b) Intermediate liquid flow rate and low gas flow rate
(c) High liquid flow rate and low gas flow rate
Figure 2. Tomograms showing the structural arrangement of the high viscosity liquid in a two-phase flow.
The relevance of these measurements stems from the simple fact that great majority of the models so far reported in the scientific literature do not encompass the high-viscosity regime. Around 99% of the published articles deal with low viscosity flows, because the primary motivation behind those studies relates to the production of high API crude oils (i.e. oils with low viscosities and densities). Even though this situation has been changing over time, the trend still progresses slowly. An example of the inadequacy of the available models is portrayed in Fig. 3. In the case of air-glycerin (high-viscosity) mixtures, the measured pressure gradients differ significantly from the values predicted with the (robust) Lockhart & Martinelli correlation. For each flow condition, the three pairs of data points show the value at their respective locations on the test section. Clearly, the pressure gradients are under-predicted with a considerable margin of error.
Figure 3. Comparison of experimental pressure drop and Lockhart & Martinelli correlation for an air-glycerin flow.
Far more important than the visualizations provided by the tomograms is the measurement of the actual values of the volume fractions of the phases. The CT Scans allow for a time-resolved measurement of the local void fractions and liquid holdups. Figure 4 illustrates the synchronous variation of the local liquid holdup and the pressure for two distinctive cases, namely: a constant gas flow rate combined with a low or a high liquid flow rate. It goes without saying that these measurements are essential to the characterization of the flow, because other properties may be derived through an application of the principles of mass and momentum conservation.
(a) Low liquid rate (b) High liquid flow rate
Figure 4. Synchronized time-series of the local pressure and liquid holdup.
Collaboration between the IPN, ININ and the BSC
Further experimental results were obtained by Ignacio Carvajal’s group at the Instituto Politécnico Nacional (IPN). This work concerns a combustion process, whereby the flame interacts with a water spray (formed with an annular nozzle) to moderate the temperature on the outer layers of the flame front. The respective data sets were already shared with Daniel Mira and Oriol Lehmkuhl from the BSC. Our Spanish colleagues will fine-tune their Alya Code to reproduce the specific features of this process. As a preliminary step towards this goal, a benchmark simulation was performed with ANSYS-Fluent by Mauricio de la Cruz. The two images shown in Fig. 5 illustrate the temperature and velocity contours in the combustion domain.
Figure 5. Turbulence-intensity contours in the wake induced by a bank of flat tubes at high Reynolds numbers.
The work outlined above was complemented with the analysis of the pressure drop produced along pipes carrying air-water and air-oil mixtures. The data was obtained in the fluid dynamics laboratory at the IPN. Fig. 6 shows the differences between the two cases; the non-linear behavior of the pressure gradient in the case of air-oil mixtures is quite evident and intriguing. An upcoming article is being drafted to report on the corresponding findings.