identification of flow regimes using raw eit

WCIPT8 - 8th WORLD CONGRESS ON INDUSTRIAL PROCESS TOMOGRAPHY Iguassu Falls, PR, Brazil, September 26 to 29, 2016 ISIPT - The International Society for Industrial Process Tomography

IDENTIFICATION OF FLOW REGIMES USING RAW EIT MEASUREMENTS 1,3

1

3,

2

A. Dupré , G. Ricciardi , S. Bourennane , S. Mylvaganam CEA Cadarache, STCP/LHC Laboratory, 13115 Saint-Paul-Les-Durance, FRANCE 2 University College of Southeast Norway, Department of EE, IT & Cybernetics, Faculty of Technology, Campus Porsgrunn, NORWAY 3 AMU, CNRS, Centrale Marseille, Institut Fresnel UMR 7249, 13013 Marseille, FRANCE * [email protected] 1

ABSTRACT In multiphase flow studies, the distribution of the different phases fall into categories called ‘flow regimes’ depending on the prevailing rheological parameters. Analysis and measurements of the different phases are simplified by ascertaining the distribution of phases. The knowledge of flow regimes helps to select the right model and suitable control actions to alleviate conditions involving dangerous flow regimes such as slugs or to select pre-computed sensitivity matrix in back-projection algorithms for enhanced online image processing. EIT (Electrical Impedance Tomography) has been used in flow regime studies in the recent past. Usage of raw measurements from direct time series analysis of the raw EIT data prevents image processing, giving rise to easier and faster recognition of flow regimes using the non-invasive EIT sensor arrays. Simple algorithms can be implemented online providing a priori knowledge of the flow regimes. In the approach proposed by the authors, the time series of each inter-electrode normalized capacitance measurements taken over a suitable duration are characterized by the average and the standard deviation (SD). In an EIT system, criteria based on the eigenvalues of the matrices evolving from twin plane raw data are used to identify stratified, annular and intermittent flows and to quantify the instability level for transitional flows. For intermittent flows, a further analysis of the Fast Fourier Transform (FFT) of the time series is applied to derive the frequency of plugs and slugs. Keywords Eigenvalues, Electrical Impedance Tomography (EIT), Flow regimes, Raw data, Time series Nomenclature EIT ECT ERT FFT SD GRM SVM Symbols Ci,j C Cmean CSD rHF/LF

Electrical Impedance Tomography Electrical Capacitance Tomography Electrical Resistance Tomography Fast Fourier Transform Standard Deviation Gamma -Ray Meter Support Vector Machines th th Capacitance value between i and j electrodes in the ECT sensor Capacitance matrix Matrix of the mean of time series of capacitance measurements Matrix of the SD of time series of capacitance measurements Matrix of ratio of high frequencies and low frequencies

1

Context of this study

The objective of the present study is to propose simple tests on raw electrical capacitance tomographic (ECT) measurements to determine a set of parameters and identify the flow regimes in multiphase flows. Such information is of great value in various research fields associated with the study of multiphase flows, and can enhance imaging algorithm for applications requiring deeper understanding of different phenomena associated with the multiphase flow. It is of particular interest to use non-invasive and non-intrusive instrumentation techniques for the flow regime identification since they do not perturb the flow. Furthermore, electrical tomography techniques features high frame acquisition rates suitable for the study of fast evolving flows. 1

WCIPT8 - 8th WORLD CONGRESS ON INDUSTRIAL PROCESS TOMOGRAPHY Iguassu Falls, PR, Brazil, September 26 to 29, 2016 ISPT - The International Society for Industrial Process Tomography

For this purpose, using the experimental setup of the multiphase flow rig in University College of Southern Norway (USN), we have considered tomographic capacitance measurements of a horizontal air/water flow. The method can be generalized to electrical impedance tomography (EIT) sensors and flows of other media. Lee et al. (2008) have trained neural networks to recognize the flow regime from raw data. Other researchers have also tried to extract information out the spectral behaviours of the signals, such as Hervieu (1998) that quantified the instability degree of a transitional flow. We have based our approach on the analysis of the eigenvalues of the capacitance matrix, as was suggested in the paper by Fang and Cumberbatch (2005). In this study, the authors propose a criterion for distinguishing symmetric configurations (annular and core flows) from asymmetric configurations (stratified flow). Besides, they prove that the analysis of eigenvalues provides an identification method that is invariant under rotation. In Section 2, we present the multiphase flow rig in USN used in generating various flow conditions in pipe sections which can be tilted, enabling horizontal and inclined flows. In addition to conventional pressure and mass flow rates measurements for air and water, measurements from ECT module, volume-averaged void estimates from a gamma-ray meter (GRM) and high-speed camera recordings in the transparent section of the pipe are also available. In section 3, we describe the set of experiments that helped to propose the criteria used in identifying of the flow regimes. In section 4, the criteria derived are applied to ECT raw data obtained from extensive experiments involving different flow regime scenarios. In section 5, suggestions for future work are disclosed.

2 Multiphase flow rig The USN multiphase flow rig enables the injection of mineral oil, water and air in a horizontal or nearhorizontal pipe made of Plexiglas (see Figure 1). The input mass flow rates are accurately controlled and monitored by Coriolis flow meters. One can select water mass flow rate up to 150 kg/min and air mass flow rate up to 5 kg/min. The pipe inner diameter is 56 mm and its length is 15 m long so the flow is fully developed in the measurement section at its end (see Figure 2). In the horizontal configuration, stratified, wavy, annular, plug and slug flows can be generated with the water and air mass flow rates approximately following the flow regime map for air/water flow in 50 mm horizontal pipes proposed by Mandhane (1974). Only the dispersed bubble flow cannot be generated because of the limitations of water mass flow rates. Besides, it is possible to position the test section with tilt angles up to ± 10° to the horizontal. The transparent Plexiglas test section is equipped with a dual-plane 12-electrodes ECT sensor, a GRM, a high-speed camera and differential pressure measurements. A schematic of the dual plane ECT sensor system is given in Figure 3, in which a stratified flow scenario is presented.

2


Figure 1. P&ID of the multiphase flow loop with installed tomography and gamma sensor systems as explained in Pradeep et al. (2014)

Figure 2. Test section with sensor placements as part of the tilted pipe with multiphase flow. Transparent section for high-speed camera based studies, multimodal tomographic systems at the far right of the pipe section. Differential pressure transmitters.

Figure 3. Twin plane ECT tomographic system – Array of 12 electrodes on the periphery of the pipe section with stratified flow in this schematic, Cij is the capacitance between electrodes i and j, with i ,j = 1,2, ..12

3


3 Experiments performed using multiphase flow rig A total of 72 experiments for various flow regimes have been performed for this study. The location of the different flow conditions are shown in Figure 4, approximately 25% of which are intermittent flows, 25% annular flows, and 50% stratified or wavy flows. Mostly transition flows have been studied in order to gain an insight on what characterizes each flow regime and to find out what happens to the set of criteria in the transition region.

Figure 4. Set of 72 experiments and the identified flow regimes superimposed on map by Mandhane (1974)

During each experiment, the following measurands are monitored: raw data from twin plane ECT module, GRM signal, high-speed video recording, the differential pressure and the flow using Coriolis flow meters, thus providing a set of time series for analysis. Capacitance measurements were acquired at 100 frames per second over 60 seconds, thus leading to data set of 6000 frames. High-speed videos were captured at 130 frames per second over 10 seconds, an image acquisition rate suitable for distinguishing plug from slug and stratified from wavy flow. The GRM was operated at 20 Hz so intermittent flows can be identified and their appearance frequency of slug/plug units obtained. The ECT module operates at an AC electrical field frequency of 1 MHz and acquires 100 frames per second. 12 large electrodes (11.7 mm wide and 85 mm long) are placed on the outer diameter of a wall made of dielectric material. A capacitance measurements between electrode i and j consists in measuring the electrical charge on j resulting from a unit electrical potential applied on i. Since the capacitance measurements are symmetric (Ci,j=Cj,i), there are 66 independent measurements in a frame from a 12-electrode sensor. The self-capacitance is defined by Fang and Cumberbatch (2005) as i,i = − ∑ ≠ i,j. A frame can be conveniently expressed as a 12x12 capacitance matrix C: 1,1

=( ⋮ 12,1

⋯ ⋱ ⋯

1,12

⋮ )

(1)

12,12

The capacitance measurements can be normalized, i.e. linearly interpolated between the corresponding measurements for the pipe full of air, and the pipe full of water. 4


4 Criteria used for identifying flow regimes From the set of 72 experiments, two criteria have been found to distinguish intermittent (plug or slug), asymmetric (stratified or wavy) and annular flows. In an earlier study, the methods available for data fusion have been described (Pradeep et al, 2011). In Figure 5, we have different methods for fusing the data and getting process relevant information. Our strategy in this paper is to use statistical methods coupled to the eigenvalues of the capacitance matrix. Earlier works by Alme et al (2006) and Alme (2007) have looked into soft sensing methods based on the inferential methods mentioned in Figure 5.

Figure 5. Data fusion in the context of capacitance based tomographic measurements. Note the classification: estimation methods and inferential methods, adapted from Chaminda el al 2012.

For each set of experiments, the 60 seconds recording from ECT sensor provides N=6000 capacitance matrices. Let us consider the following matrix of time series of the capacitance values obtained using the ECT system: 1,1[n]

=(

⋮ 12,1[n]

⋯ ⋱ ⋯

1,12[n]

⋮

) , = 1,2, … ,

(2)

12,12[n]

We define the mean and standard deviation (SD) capacitance matrices as follows:

mean

=(

mean( 1,1[ ]) ⋮ mean( 12,1[ ])

⋯ ⋱ ⋯

mean( 1,12[ ]) ⋮ ) mean( 12,12[ ])

SD = (

SD( 1,1[ ]) ⋮ SD( 12,1[ ])

⋯ ⋱ ⋯

SD( 1,12[ ]) ⋮ ) SD( 12,12[ ])

(3)

where the mean and the standard deviation operators of a time series are defined as: 1

mean( [ ]) = ∑ =1 [ ] 1

SD( [ ]) = √ ∑ =1( [ ] − mean( [ ]))2 .

(4) (5)

Finally, we consider the eigenvalues of these mean and SD matrices, which are shown in Figure 6 and Figure 7 respectively. This methodology of fusing the ECT data helps to collapse the data (from 12x12x6000 matrix to two 12-elements arrays) and facilitate rotationally invariant tests to be performed. For example, the eigenvalues corresponding to a stratified flow will remain unchanged if the sensor is rotated.

5


Our first criterion identifies intermittent flows (plug and slug). It is based on the CSD matrix: the leading eigenvalue consistently exceeds 1.5, while for annular flows it was around 0.7 and even lower for stratified and wavy (see Figure 7). The second criterion identifies an annular flow. It is based on the Cmean matrix: the absolute value ratio of the second largest with the smallest eigenvalue exceeds 0.75 (see Figure 6). Though the transition is gradual, based on video recordings, the transition to annular was set when the water film covers the top of the pipe and the threshold estimated. Despite the arbitrariness of the threshold, and knowing changing it will shift the transition line, we are encouraged to recover the same curvature (see Figure 3) as in Mandhane (1974).

Figure 6. The 12 eigenvalues of the mean matrix for typical plug and slug flows (in green), stratified and wavy flows (in blue), annular flows (in red) and full pipe (in black). Error bars indicate variability in the 72 measurements.

Figure 7. The 12 eigenvalues of the SD matrix for typical plug and slug flows (in green), stratified and wavy flows (in blue), annular flows (in red) and full pipe (in black). Error bars indicate variability in the 72 measurements.

For each experiment, the flow regime has been determined using the flow diagram shown in Figure 8. In Figure 1, the transition zones have been superimposed on existing flow regime maps available in the literature (Mandhane, 1974), showing the versatility of the EIT based classification and fast 6


identification of flow regimes. The attentive reader will note that the transition regions when increasing the pipe diameter (from 50 to 56mm) shift to regions with higher air and water flow rates (up and right). This fact is substantiated by visual observations and data from the GRM and the high-speed camera. The transition line from wavy to annular flows has been defined as the complete wetting of the pipe, which was observed visually or on the videos by a glare originating from the top of the pipe. The intermittent flows have been validated using the GRM which shows abrupt fluctuations of its signal. Figure 8 shows the diverse criteria used in the classification/identification of the flow regimes in the pipe with multiphase flow. As the capacitance values are normalised and as the criteria are based on the eigenvalues and some selected statistical properties of these, we do not expect drastic changes in the set of criteria for the flow regime identification strategy discussed here. It is now evident that plug and slug (or stratified and wavy) flows differ in their spectral behaviour leading to a third indicator based on the spectral fingerprint of these flow regimes. This hypothesis needs testing and verification using more experimental results. EIT sensor Raw data

Capacitance matrix of time series 3rd indicator

SD Me an

CSD Eigenvalues

SD[12]

??

Cmean

> 1.5 ?

yes

Plug/Slug

Eigenvalues

Eigenvalues

no

[11]

mean

�� mea n[1]

> 0.75 ?

yes

no

Tests to be defined

Annular

Stratified/Wavy

Figure 8. Flow diagram for flow regime identification. All the criteria used in this study are presented using the colour code used in Figures 6 and 7. Some future perspectives are also shown.

5 Future work Questions remain, notably how general the proposed criteria perform with different values of pipe diameter, sensor design and geometry, and nature of phases. Ongoing extra measurements will help further validate (or refine) our algorithms. So far, new data in different regions of the flow regime map for the horizontal configuration has increased our confidence in the two criteria proposed. Additionally, measurements for tilted configuration (±3°) featuring high amplitude waves resembling intermittent flows also fitted correctly in the identification algorithm based on eigenvalues. Future work will focus on testing the entire flow regime map, using different combination of media (mineral oil and water) in combination with electrical resistance tomography (ERT) sensor. To validate the universality of this methodology in ascertain the flow regimes in various dimensions of pipe lines, some more experiments should be run using different pipe dimensions. Besides, a research on a third indicator based on the frequency components (i.e. spectral behaviour) of the signals can probably help 7


in distinguishing a slug from a plug, and a stratified from a wavy flow and in identifying this flow by a combination of the techniques describe in this paper. Being a classification problem, involving many flow regimes and parameters, future study related to multiphase flow and flow regime identification, can also involve the usage of Support Vector Machines (SVM), which due its versatility and swift performance has found in many fields, (Haykin, 2009). Finally, a numerical method for simulating the sensor response is under development. It will help to estimate void fraction over the cross-section of the pipe section, once the algorithm described here has identified the flow regime.

6

Acknowledgements

Antoine Dupre is grateful to Morten Christian Melaaen of USN, Isabelle Tkatschenko, Gaelle Pestel of CEA, and others for making the research sabbatical to USN possible. STATOIL, Norway provided part of the multiphase flow facilities and the equipment used in this study. Mr. Eivind Fjelldalen of USN has been responsible for the hardware involved in the data logging from the sensors used in monitoring the multiphase loop. Docent Dr. Finn Aakre Haugen of USN is responsible for the LabVIEW program used in steering and monitoring the multiphase flow. Many thanks to Fredrik Hansen of USN for his invaluable help with the operation of the multiphase flow rig and with the data logging from all the measurement systems coupled to the rig. Master student from Faculty of Applied Engineering: Chemistry, University of Antwerpen, Belgium, Stijn Vyncke in ERASMUS+ student exchange, currently at USN, was partly involved in some of the measurements performed with the multiphase flow facilities at USN. Prof. Malcolm Byars of Process Tomography Ltd, Manchester provided support and advice in conjunction with the usage of ECT System coupled to the multiphase flow rig at USN.

REFERENCES ALME K. J., Material distribution and interface detection using EIT, PhD thesis, Telemark University College and NTNU, Norway, 2007 ALME K.-J., MYLVAGANAM S., Electrical capacitance tomography: sensor models, design, simulations, and experimental verification, Sensors Journal, IEEE 6(5), 1256–1266., 2006 FANG W., CUMBERBATCH E., Matrix properties of data from electrical capacitance tomography, Journal of Engineering Mathematics (2005), 51 127-146 HAYKIN S.O., Neural Networks and Learning Machines, 3rd Edition, Pearson, 2009 HERVIEU E., SELEGHIM P., An objective indicator for two-phase flow pattern transition, Nuclear Engineering and Design (1998), 184 421-435 LEE J.Y., ISHII M., KIM N.S., Instantaneous and objective flow regime identification method for the vertical upward and downward co-current two-phase flow, International Journal of Heat and Mass Transfer (2008),3442-3459 MANDHANE J.M., Gregory G.A., and Aziz K., A flow pattern map for gas-liquid flow in horizontal pipes, Int. Journal of Multiphase Flow (1974), 1 537-553 PRADEEP C., RU Y., MYLVAGAAM S., Neural network based interface level measurement in pipes using peripherally distributed set of electrodes sensed symmetrically and asymmetrically, IEEE Transactions on Instrumentation and Measurement, Vol:61 , Issue: 9, pp: 2362-2373) 2012 PRADEEP C., Tomographic approach to automatic and non-invasive flow regime identification, TUC PhD thesis, 2015

8