You are on page 1of 15

The 12th International Topical Meeting on Nuclear Reactor Thermal Hydraulics (NURETH-12) Sheraton Station Square, Pittsburgh, Pennsylvania

, U.S.A. September 30-October 4, 2007.

Log Number: 061

APPLICATION OF NMR TO FLUID FLOW MEASUREMENTS
Herv´ Lemonnier1 e DER/SSTH/LIEX, CEA/Grenoble, 38054 Grenoble Cedex 9, France herve.lemonnier@cea.fr Jacques Leblond Laboratoire de physique thermique, ESPCI 10 rue Vauquelin, 75231 Paris Cedex 05, France jac.leblond@wanadoo.fr, ABSTRACT In this paper, preliminary area-averaged void fraction and liquid mean velocity measurements are reported. The results show that nuclear magnetic resonance is a very versatile tool to characterize two-phase flow. Some of the reported results have already been obtained by others (see for example Leblond et al. , 1998) and are therefore not original. The Spinflow experiment, originally designed and built in Paris has been moved to Grenoble and the reported results show that the equipment has been correctly implemented in the new environment. A new program has started in Grenoble aiming at the use of magnetic resonance imaging technology and the refinement of its spatial resolution from the present 5 cm to less than 5 mm. It is the authors’ intention to promote the use of NMR techniques as a reference for two-phase flow measurements. KEYWORDS Two-phase flow instrumentation, void fraction, liquid velocity, nuclear magnetic resonance, NMR, MRI. 1 INTRODUCTION

NMR is a well known technique in the fields of solid state physics and chemistry. It is however seldom used for single-phase flow and and its potential for two-phase flow characterization is almost not acknowledged. Recent studies have shown, however, that velocity can be measured locally and accurately in some particular flows. Gatehouse et al. (1994) utilized successfully NMR to measure aortic blood flowrate. More recently, Bittoun et al. (2000) have shown that 3D mapping of myocardium velocity was feasible in vivo. Moreover, Raynaud et al. (2002) have shown that velocity profiles can be determined in a Couette rheometer with a gap of 2 cm and these authors derive the rheological properties of various non-Newtonian suspensions by the direct measurement of the actual local shear rate. There is no need to provide a longer review of existing studies to advocate the use of NMR techniques in fluid mechanics. This paper reports new experiments performed on a new version of the Spinflow experiment. Spinflow was specially designed and built at ESPCI in Paris by Leblond and Javelot (Javelot, 1994) for two-phase flow investigations. The experiment has been moved to Grenoble in the first author’s laboratory. A new flow loop was designed and built and a brand
1 Corresponding

author

(1/15)

Likewise. Single-phase data are shown and the velocity measurement and void fraction calibration procedures are detailed. this observation derives from the existence of the nuclear spin of certain nucleus such as the proton or fluor. When a fluid containing protons is immersed in a magnetic field of strength B0 . Next a short description of the experiment is provided. For times much less that T2 . if by an appropriate mean the direction of the magnetization is tilted from its equilibrium position (aligned with B0 ) then magnetization rotates around the direction of B0 with an angular characteristic pulsation. depending on the magnetic field strength and the (2/15) . The mechanical one-dimensional counterpart of these equations are the equations of motion of the needle of a compass subjected to the terrestrial magnetic field. For water T2 is close to 2 s. it is proportional to the number density of the considered nucleus. Magnetization time evolution under the influence of a magnetic field mainly oriented in the direction of B0 (B0X B0 . the classical (as opposed to that of quantum physics) description of magnetization is far sufficient. Rodriguez (2004) provides the classical presentation of NMR physics and the first principles of measurement and imaging. In this paper we reproduce basically some of the results obtained earlier by Javelot (1994) to show that the Spinflow experiment and its new console has been successfully set up in Grenoble. the so-called adiabatic magnetization. ω0 . In the first section. For the correct understanding of MNR fluid measurements. When the magnetic field is no longer applied. B0Y B0 and B0Z ≈ B0 ) is described by Bloch equations. magnetization does not remain constant but fades out following a first order relaxation with the characteristic time T2 . 2 BASICS OF NMR Any substance can be magnetized provided it is submitted to a magnetic field during a sufficiently long time. it is magnetized in proportion to the strength of B0 . If M is the magnetization and M∞ . then an oscillating and damped return to equilibrium takes place with a period depending on the field strength and the mechanical inertia of the needle. Lemonnier and J. The compass mechanics provides a quite simple 1D image of the spin system evolution. Those equations can both be derived from the angular momentum balance considering an applied torque proportional to the cross product of the magnetization and the magnetic field.H. the relaxation processes can be neglected and magnetization is regarded as constant. we review briefly the principles of NMR measurements and how flow related information can be collected. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 new Apollo Tecmag NMR console has been installed in replacement of the original SMIS equipment. Furthermore. Finally two-phase flow data in bubbly flow regime are shown. Some conclusions are drawn and research perspectives are briefly outlined. The mechanical counterpart of the relaxation processes is a mere fluid (linear) friction. For liquids. (1) Magnetization is aligned with the magnetic field. M(t) = M∞ (1 − exp(−t/T2 )). is its value reached after a sufficiently long time then. the so-called longitudinal relaxation time. If the needle of a compass is deviated from its equilibrium position.

Maximum signal is obtained when the precession axis of the magnetization is perpendicular to the coil axis.H. If a coil is placed around the sample. If the receiver coil is adequately designed. the imaginary and real parts of the signal are proportional to the two components of the magnetization perpendicular to the main field direction as seen in a frame rotating at the Larmor frequency. The maximum effect is obtained when the frequency of this auxiliary field is equal to the Larmor frequency. the overall magnetization of the fluid sample varies with time at the Larmor frequency and with a strength proportional to the mass of the sample. the signal does not depend on the actual position of the sample within the coil. Lemonnier and J. The corresponding frequency is the so-called Larmor frequency of the free precession motion and it given by. it is easy to wait for complete magnetization of the sample before the measurement (3/15) . This is the basis of the NMR signal detection. By adjusting appropriately the reference phase. Usually the coil axis is placed perpendicular to the B0 field direction and the coil signal is detected in phase and quadrature at the Larmor frequency. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 nature of the nucleus under investigation. The free precession of magnetization in the terrestrial field can be observed easily by using very ordinary equipment as shown by Ravanello & Sconza (1987) or Callaghan & Le Gros (1982) who inspired the work of the former group. The process is easy to understand by using again the compass as a mechanical model.2 Void fraction measurement In a static water sample. Applying a field perpendicular to the direction of the terrestrial magnetic field rotates the needle by some angle. the corresponding frequency is close to 5 MHz with a field of strength around 0.12 T. 2. The deviation angle depends on the strength and frequency of this field and on the duration of the field application. the NMR signal amplitude is proportional to the mass of the sample (number of protons of the sample) and the initial magnetization. This is known as the nuclear magnetic resonance phenomenon mainly because the tilt amplitude varies with frequency in a manner similar to that of a 1D oscillator under a variable frequency driving force. ω0 = γB0 (2) where γ is the so-called gyromagnetic radius of the nucleus. For protons. it sees a variable flux and therefore a variable voltage is induced in the coil winding ends (Lenz law). Its strength is proportional to the magnetization strength and depends on the angle between the coil axis and the magnetization. applying a magnetic field perpendicular to the main field B0 rotates the magnetization around this field direction. The ”tilting efficacy” decreases very rapidly when the frequency deviates from the Larmor frequency. In order to be complete it must be said that rotation is retrograde.1 Magnetization detection When magnetization is under free precession conditions. In a static sample. 2. Usually the coil used for detection is also used for initially tilting the direction of magnetization. Likewise.

In an air-water flowing system. The ratio of this two quantities is the liquid fraction of the sample RL3 . a small perturbation field aligned with B0 with a strength varying linearly with space. These methods require tagging some fluid property with the position and two basic procedures: tagging once or possibly twice with position and reading the tags. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 and therefore the signal is proportional to the water content of the sample only. δ. one has simply to compare the signal obtained in two-phase flow to that obtained in single-phase flow with the same residence time. To simplify let us consider a 1D-motion with a constant and uniform velocity and let us consider a particular fluid element with an initial magnetization m0 . A gradient pulse tags the fluid with the position through the phase of its magnetization. In a snapshot. Apply a first gradient pulse with a strength g and duration δ.3 Velocity measurement The velocity measurement in NMR is similar to the double exposure shots or PIV techniques. Before tagging the fluid. the exposure time plays the role of δ. during a short period period of time. Considering a position z where the field is now B0 +gz and comparing to the mid plane of the system (z = 0). the sample diameter is a cylinder with a diameter of 49 mm and a height of approximately 100 mm. it is clear that magnetization rotates faster in z by the amount γgz. it must have been magnetized and the magnetization tilted at right angle in the plane perpendicular to the main field axis. Such fields are provided by so-called gradient coils because the field varies with space as gz. This is very conveniently and non intrusively induced by external magnetic fields. where g is the gradient strength and z is some distance along the direction of the gradient. Next. but the phase of the signal is shifted by the angle γgzδ. During the application of the gradient field. Magnetization at this location is therefore. In Spinflow. The fluid element magnetization becomes. ω0 . m(z) = m0 exp(−iγgzδ) (3) where m0 is the magnetization per unit volume in absence of gradient pulse. precession motion is back again at the initial Larmor pulsation. During time δ magnetization in z is phase shifted by an angle equal to γgzδ. 2. Lemonnier and J. see (2). magnetization depends also slightly on the residence time in the electromagnet according to (1). m1 = m0 exp(−iγgδz1 ) (4) where z1 is the position of the fluid element at the time of the first gradient pulse. When the gradient is suppressed. Moreover. Magnetization is used as a tagging property. δ must be sufficiently small to avoid getting a blurred picture.H. Free precession occurs and by adjusting the phase detection appropriately the measured signal is constant and real. To measure the water fraction of the a two-phase flow. The minus sign in the argument is not essential but accounts for the retrograde nature of the rotation. (4/15) . proceed by adding to B0 . the overall magnetic field depends slightly on z and so does the frequency of the precession motion. This will be denoted as the measurement volume. the magnetization is very weak and has absolutely no mechanical feedback on the flow (no induced mechanical stresses).

the averaged velocity.51 G/cm and γ = 2. on gets. one has. where D is the pipe diameter. m2 = m1 exp(iγgδz2 ) (5) where z2 is the position of the same fluid particle at the time of the second gradient pulse. Convection has moved the fringes pattern over a distance corresponding to approximately 35 fringes. g=3. With a velocity of 50 cm/s and ∆ = 5 ms this is 2. After the first gradient pulse. During the time ∆. magnetization becomes phase shifted by an amount proportional to the distance covered by the fluid element during time ∆.H. among others. mY = m0 sin γgδz (10) with the spatial wave length λ = 2π/γgδ. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 Next wait for a time ∆ and apply a second gradient pulse with a strength −g. oscillates with g with a period inversely proportional to v. As a result. (5/15) . mX = m0 cos γgδz. By introducing P(v) the fraction of the volume V where the velocity ranges between v and v + dv. Javelot (1994). M(g). the fringes are convected by the fluid over the distance v∆ where v is the fluid velocity.5 ms. dV = VP(v)dv and therefore (7) becomes. In other words the magnetization at the end of the sequence. showed how (9) is derived from the exact solution of the Bloch-Torrey-Stejskal equations that describe the evolution of magnetization during the so-called pulsed field gradient spin echo sequence. By repeating this procedure for all the values of g between 0 and gmax one gets a phase shift linearly varying with g and v. When the second gradient pulse is applied. on the time interval ∆ is introduced. A physical picture is useful to understand the velocity measurement. the fringes are ”unfolded” and what remains is only the phase shift corresponding to the fluid displacement approximately 70π.5 mm. M(g) = M(0) P(v) exp(iγgδ∆v)dv (9) (8) which shows M(g)/M(0) is the Fourier transform of P(v) in the variable γgδ∆. m = m0 exp(iγgδ∆v) (6) Consider now a pipe flow with the pipe axis aligned with the magnetic field B0 and the gradient direction. Magnetization after this second pulse becomes. magnetization components (4) perpendicular to the gradient direction show a fringe pattern. magnetization after the double pulse sequence reads. In fact the contributions to (7) are limited to the measurement volume scaled by V = πD3 /4. With typical values such a δ = 1. M(g) = m0 exp(iγgδ∆v)dV (7) where the integral is considered over all the magnetized fluid.67519 108 for protons. This characteristic behavior is shown for example in figure 3(a). Since the overall signal is obtained by summing up the contribution of all the fluid elements in the measuring volume. If v. Lemonnier and J. λ ≈ 71 µm.

" . This change of sign is achieved in a manner similar to the initial tilting of magnetization by applying a second RF pulse that rotates magnetization at 1800 in a direction perpendicular to the first RF pulse. Echo occurs at 2τ . • Steps 7-10: First gradient pulse of strength g and duration δ. This results from the magnetization orientation dispersion induced by the field non uniformity. Though uniform within less than 1 ppm in the measuring volume.51 G/cm. The sequencing of events is controled by the MNR console. Run: 0278. The detailed description of the hardware including coils and signal analog processing equipment is detailed by Javelot (1994). to tilt magnetization at right angle. )  " . • Step 12 :Second RF pulse to change the sign of phase.6(201) 80000 Signal imag part Signal real part First in the series (R) First in the series (I) J @ & ' J . The sequence shown in figure 1(a) consists in the following steps. It is also positive since the sign of the phase has been changed by the previous step and (6) still holds. " 2 −40000 −60000 0 50 100 150 200 250 300 350 Sampling time 100 µs. Numbers refers to labels in the figure. !  % . δ = 1.tnt 400 450 (a) PFGSE sequence (b) Time evolution of magnetization.  #  $ J 60000 ) ? G 40000 Signal amplitude 20000 0 −20000 / % . Figure 1: Diagramatic description of the PFGSE sequence and measured magnetization 3 PULSED FIELD GRADIENT SPIN ECHO SEQUENCE A MNR sequence consists in creating various fields at some selected times and reading the data at some other selected instants. • Step 19: Data acquisition. ∆ = 6. duration tπ = 2tπ/2 • Steps 14-17: Second gradient pulse.5ms. is not exactly uniform and magnetization decreases much faster than expected from the relaxation time (see figure 1(b)). ) . the overall signal reverts to its original value at time 2τ . The main deviation from the idealized sequence described in the previous section is due to the field non uniformity. This is known as the Hahn’s echo technique. (6/15) . the magntic field B0 . duration tπ/2 . However. ) ? G $   !  ' Echo : 200. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 # 4 . !   . • Steps 5-6: Data acquisition to get the initial magnetization. • Step 5: 5 MHz RF pulse.7 ms.H. this dispersion is reversible and by changing the sign at some time τ of the magnetization. gmax = 3. Lemonnier and J.

1992. Air and water are simply mixed 2 m upstream the glass pipe by blowing air through a porous injector designed originally for oxygenation of fish tanks. This is achieved by selecting ∆ = 6. 12) and yields the velocity distribution. Next the PFGSE is reproduced for various values of g ∈ [−gmax . However. (7/15) . Ch. JL = 30 cm/s) is selected and the gradient pulse duration and separation ∆ and δ are adjusted to provide all the relevant data within the range −gmax and gmax . Lemonnier and J. The two-phase mixture then expands downstream the test section into an horizontal liquid tank where air is separated from water by gravity.7 ms and δ = 1. The rest of the circuit is made of PVC commercial piping (44 mm internal diameter).9 l/m. 4 SPINFLOW SET-UP DESCRIPTION The Spinflow flow loop consists of a closed water loop and an open circuit for air. gmax ]. Endress+Hauser Picomag type (30 and 300 l/min) while air is metered through 3 Brooks thermal mass flow meters ( 5. Next the magnetization is inverse-Fourier transformed by using the FFT algorithm (Press et al. the various power amplifiers feeding the coils are rigourously linear and this knowledge is sufficient and permits a calibration of the gradient strength of a particular sequence. Air and water are metered through a set of two electromagnetic flow meters. This methods provides a rather uniform bubble distribution with a diameter ranging from 3 to 5 mm during the experiments to be reported here. 200 ln/min). the current actually flowing in the various coils cannot be measured accurately. This device provides also a constant pressure exit condition to the flow and therefore the single-phase liquid flows in the loop are free from any pulsation and other flow instability. The test section consists of a glass pipe with a diameter of 49 mm and a length of 1. 50. Magnetization at echo is extracted from the time series shown in figure 1(b). Though gradient strength and RF field amplitude should be known from design parameters.H. A mid range condition (QL = 33. The liquid flow originates from a constant head tank filled by an air-lift pump. eddy currents still develop in the various copper walls of the electromagnet and the determination of the actual values of the various fields is almost impossible. though gradient coils are shielded. 5 SINGLE-PHASE LIQUID VELOCITY DISTRIBUTION MEASUREMENTS Single-phase flow is first studied. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 From all the acquired data M(g) of (9) is simply selected as the value of magnetization at time 2τ as shown by the vertical bar in figure 1(b).5 ms. . Moreover. This arrangement has been selected to provide a rather constant inlet pressure to the system. The obtained data is shown in figure 3(a). Pressure taps are set up at each ends of the glass pipe to get the static absolute pressure in the measuring section that is located in the center of the set of coils shown in the electromagnet depicted in figure 2.5 m.

ux = γg∆δv. They are defined by. Equation 9 shows that u and x are conjugated variables for the Fourier transform since. and u is the non dimensional velocity scaled by vmax .H. (12) Data in figure 3(b) is not yet the space distribution of velocity since the normalization condition. Figure 3(a) shows magnetization as a function of the non dimensional gradient strength. while the space distribution velocity is shown as a function of the conjugated non dimensional variable u. g = xgmax . vmax = 1 γgmax ∆δ (11) where v is the averaged velocity of the flow on time ∆. x. ∞ f (u)du = 1 −∞ 8 3 3 1 8 0 7 (13) (8/15) . u = v/vmax . Lemonnier and J. Leblond Application of NMR to fluid flow measurements 5 0 0 3 0 0 NURETH-12 Log: 061 4 9 4 2 9 0 1 0 3 0 1 0 2 9 0 1 0 0 0 A ltitu d e 1 9 0 0 Figure 2: Schematic of the electromagnet showing the test section and the coils set.

01 0 −0. If the velocity profiles remains only sightly variable with the liquid mean velocity. so does the velocity distribution.6 0. This latter quantity is proportional to the space distribution of velocity and is normalized by condition (14).04 0. d.8 10 20 30 40 50 60 4000 3500 3000 FFT Echo amplitude 2500 2000 1500 1000 500 0 -500 -100 Echo imag part Echo real part Echo amplitude Inv FFT imag part Inv FFT real part Inv FFT Mod -0.96 cm/s JL=19.5 0. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 0 50000 40000 30000 20000 Echo amplitude 10000 0 -10000 -20000 -30000 -40000 -0.7 0.4 Velocity (m/s) 0.2 0 0.03 0. or QL ∆δ = cst. This similarity condition reads.4 -0. JL = cst.06 P(v) (s/m) 0.89 cm/s JL=10.07 cm/s 15 10 5 0 (a) Non dimensional space distribution of velocity (b) Dimensional space distribution of velocity Figure 4: Space distribution of velocity in single-phase flow for variable averaged liquid velocity.11 cm/s JL=40. is not achieved. The value of gmax deduced from the calibration (17) has been used.09 0. vmax (14) where γgmax ∆δJL represents the magnetization phase shift after the PFGSE sequence produced (9/15) . δ = 1.8 25 JL=50. u=v/vmax 40 50 −5 −0.1 0 0.07 0.6 -0.8 1 -50 0 u=z/zmax 50 100 (a) Magnetization at echo (b) Space distribution of velocity Figure 3: Magnetization at echo and its inverse Fourier transform.3 0.1 0. Lemonnier and J.5 ms and ∆ varied according to (14). 0.11 cm/s JL=40.89 cm/s JL=10.2 ND Gradient 0.06 cm/s JL=29.02 0.05 p(u) 0.06 cm/s JL=29. that which is expected from fully developed turbulent flow. The actual value of the integral is evaluated directly from the data of figure 3(b) and is used as a normalization constant to provide the non dimensional space distribution of velocity shown in figure 4(a).2 0. ⇒ γgmax ∆δJL = cst.01 −10 0 10 20 30 n.96 cm/s JL=19. velocity.4 0.6 0.07 cm/s 20 JL=50.H.08 0.

the volume averaged non dimensional velocity is computed from.51 G/cm. From the distributions of figure 4(a). As a result. (10/15) . Figure 5(b) shows that the magnitude of the initial magnetization is a decreasing function of the liquid mean velocity. ∞ < u> 3 = | | up(u)du −∞ (15) Owing to the scale relation (11). According the scale relations (11) the following relation holds for the dimensional. In figure 4(a).3 m). < v> 3 . < u> 3 | | = GZ JL γ∆δ (17) Figure 5(a) shows that the linear behavior implied by (17) is achieved and also provides the calibration of the maximum gradient strength in the sequence. (Javelot.7 l/mn to zero while the liquid superficial velocity varies from 50 cm/s to zero and condition (14) is assumed with δ = 1. p(u). A best fit of the data provides the value gmax = 0. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 by an uninform flow at velocity JL . the following behavior is expected. which is in excellent agreement with the known values for water (Carr & Purcell. < u> 3 = γGZ ∆δ< v> 3 ≈ γGZ ∆δJL | | | | (16) where with the additional assumption of developed flow the last equality results.5 ms in addition. This procedure also assumes the velocity profile to be concave. is related to < u> 3 | | | | by. 1954). It has been however assumed that the initial magnetization inside the measuring volume is uniform. Lemonnier and J. At this point. velocity distributions. the velocity scale vmax is known and dimensional velocity distributions can be computed. P(v) = 1 vmax p v vmax (18) The dimensional data corresponding to those of figure 4(a) is now shown in figure 4(b).H. With this known value.0351 T/m=3. The overall behavior (1) is expected with t = L/JL being the averaged residence time in the electromagnet (L ≈ 1. M0 = M∞ (1 − exp(L/JL T2 )) (19) The data fit provides a value of T2 = 2. Hence the following relation is expected. it must be noticed that the absolute magnitude of the signal has not been used for velocity determination. the dimensional averaged velocity. Space distribution similarity results. P(v) and non dimensional. the liquid flowrate QL is varied from 56. 1994) has shown that the velocity profile can be reconstructed from P(v) by assuming flow axisymmetry.36 s.

002 0 0 Best fit Data 18000 16000 14000 12000 Magnetization 10000 8000 6000 4000 2000 0 Best fit Data 0. By using the procedure developed for (11/15) . By ensemble.004 0.6 0. M(g) = M(g) = M(g) = m0 XL exp(iγgδ∆v)dV m0 RL3 P(v) exp(iγgδ∆v)dv m0 RL3 P(v) exp(iγgδ∆v)dv (20) (21) (22) where XL is the liquid phase presence function and P(v) now represents the instantaneous space distribution of the velocity in the liquid phase and RL3 represents the liquid fraction in the measuring volume.7 0.8 (a) Gradient calibration (b) Initial magnetization vs liquid mean velocity Figure 5: Calibration of the gradient strength and of the initial magnetization as a function of JL . These quantities are related by. Lemonnier and J.01 0. When time.4 0.averaging (21).016 0.6 0 0.008 0.1 0. Since every individual P(v) satisfies the normalization condition so does PL (v).H.3 0. one gets the time or ensemble.014 <u>3/γ∆δ (SI units) 0. 6 LIQUID VELOCITY AND VOID FRACTION MEASUREMENTS IN TWO-PASE FLOWS It is worth noting at this point that the relation between velocity v and magnetization is linear.006 0. By introducing the averaged distribution of velocity weight by the liquid content of the measuring volume on gets. the fluctuating behavior of the velocities is sometimes one order of magnitude larger than in single-phase flow.averaged velocity distribution RL3 P(v) weighted by the liquid content of the measurement volume.5 Mean velocity (m/s) 0.018 0.2 0.or time averaging magnetization.5 0.3 0.4 Mean velocity (m/s) 0. each P(v) satisfies individually the normalization condition (13) and therefore M(0) = m0 VRL3 .2 0.012 0.or ensemble. PL (v) = M(g) = RL3 P(v) RL3 PL (v) exp(iγgδ∆v)dv (23) (24) m0 RL3 PL (v) exp(iγgδ∆v)dv = RL3 M0 (0) where M0 (0) is the initial magnetization in single-phase conditions.1 0.02 0. This of the upmost importance since in two-phase flow. the liquid superficial velocity. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 0.

• αv .5 ms and gmax = 3. Run 0297 is close to the bubble-slug transition as suggested by the appearance of negative velocities in the distribution shown in figure 7(a).6 -0.2 ND Gradient 0.13 cm/s. single-phase flow one gets PL (v) directly from M(g). Magnetization no longer oscillates as it does in single-phase flow (see figure 3(a)) and as a consequence the velocity distribution broadens significantly. JL = 30 cm/s.4 0.4 cm/s.51 G/cm.4 -0. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 0 600000 500000 400000 10 20 30 40 50 60 -200 16000 14000 12000 FFT Echo amplitude 10000 8000 6000 4000 2000 0 -2000 -100 -150 -100 Echo imag part Echo real part Echo amplitude Bexp(-Ax2) Velocity (cm/s) -50 0 50 100 150 200 Inv FFT imag part Inv FFT real part Inv FFT Mod b exp(-(u-c)2/2a2) 300000 Echo amplitude 200000 100000 0 -100000 -200000 -300000 -0. This latter value is calculated from ∞ vL = −∞ vPL (v)dv (25) It is striking to see that negative values of the velocity have been recorded. The corresponding velocity distribution are shown in figure 6(b).H. They are related to negative velocities that can be observed directly downstream the bubble swarms existing the test section and that become further downstream small Taylor bubbles. ∆ = 4 ms. Velocity distributions broadens gently with the increase of the void fraction and so does the averaged velocity. Figure 6(a) shows the magnetization as a function of gradient strength for two-phase bubbly flow.8 -0. JG = 8.8 1 -50 0 u=z/zmax 50 100 (a) Magnetization at echo (b) Liquid velocity distribution Figure 6: Magnetization at echo and liquid velocity distribution in bubbly flow. The mean liquid velocity. JL = αL < v >L3 . The only remaining condition being that M(g) must be statistically converged to get a statistically converged value of PL (v). from the averaged liquid velocity and the superficial velocity.2 0 0. Therefore. Equation (24) also says that velocity information is carried only by the phase of M while its magnitude is proportional to the averaged liquid fraction. JL = 30 cm/s and JG = 8. These conditions are close to the transition to slug flow: bubble swarms and small Taylor bubbles are seen exiting the test section. δ =1. as expected from the void fraction value. Figure 7 shows the evolution of the velocity distributions with the increase of the gas superficial velocity. with MNR and the pressure measurements there are 3 different means to estimate the area-averaged liquid fraction αL = RL3 . run 0297. vL = 38. Lemonnier and J. and RG3 ≈ 22%.13 cm/s and RG3 ≈ 22%.6 0. (12/15) .

2 0 0. In addition. (13/15) . JL=50 cm/s αM0. JG=8. JL=30 cm/s RG3.25 αM .15 0.2 0.2 αM0vL (cm/s) αM .H. JG=3. JL = 50 cm/s Figure 7: Liquid velocity distributions as a function of the gas superficial velocity for two different values of the liquid superficial velocity.00 cm/s 4.13 cm/s 0291.62 cm/s 0293.6 Velocity (m/s).5 0 −1 0 −0. JG=10. JL=50 cm/s RG3.2 (a) Liquid velocity distribution.9 cm/s 0300. JL=30 cm/s 0.25 50 0.8 1 2.4 0. JG=0. from the hydrostatic pressure measurement. JG=1. 0. • RG3 .22 cm/s 0301. JL=10 cm/s y=x 0 0.5 4 3. JL=30 cm/s αM0. JG=5.5 1 0.00 cm/s 4 P(v) (s/m) 3 2 1 P(v) (s/m) −0. Off diagonal points reveal measurement inconsistencies resulting most likely from the lack of statistical convergence of PL and provides therefore an internal consistency check of the data. from the magnitude of the initial magnetization. RG3 40 0 0. The comparison of these values is shown in figure 8(a).05 10 0 0 0 10 20 30 JL (cm/s) 40 JL=30 cm/s JL=50 cm/s JL=10 cm/s y=x 50 60 (a) Comparison of void fraction estimates (b) Comparison of JL with JL measured by NMR Figure 8: Consistency check of the NMR data.59 cm/s 0276.15 30 0. JL=50 cm/s 1 1.5 2 1. Lemonnier and J. ∆p = αL ρl gH. This procedures again provides another consistency check of the data.8 Velocity (m/s).5 3 0302. JL = 30 cm/s (b) Liquid velocity distribution.1 αVL 0. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 7 6 5 0297.63 cm/s 0290. JG=4. JG=2. JL=10 cm/s 0 RG3. JG=0.05 0. • αM0 . M(0) = αL M0 (0).5 0 0.16 cm/s 0292.3 60 0.4 0.1 20 0.2 0.2 0.6 0. by combining the liquid fraction measured from M(0) and vL one can measure directly JL .

Lemonnier.. Mag. (14/15) . Itti. Herment. J. 1982. It is non intrusive. Two-phase flow velocity and void fraction measurements revisited by using nuclear magnetic resonance. 1998. 95–118. Multidimensional MR mapping of multiple components of velocity and acceleration by Fourier phase encoding with a small number of encoding steps.. B. E.. D. The solution of Bloch Equations in the presence of varying B1 field . D. 2007. Two-phase flow characterization by nuclear magnetic resonance. Am. O.D. Javelot. 630–638. 35. Nuclear Engineering and Design. Universit´ Paris 6. In: Proc of the 7th Int. I. 44. P. Callaghan. & Bataille. 2000. & Longmore. A. S. 2007). Conf. NMR diagnotic is believed to provide a reference method to validate the former technique. Magnetic Resonance in Medecine.P. Multiphase Flows.. H. Hoult.. J. 709–713. Mousseaux. & Lebon. & Tasu. It is well known that turbulence measurements in bubbly flow is dramatically dependent on the signal processing algorithms (see for example.. Carr. Resonance. M. Lebrun.. Effects of diffusion on free precession in nuclear magnetic resonance experiments.. R´ alisation d’un dispositif utilisant la RMN pour caract´ riser les ecoulements e e ´ multiphasiques.. Magnetic Resonance in Medicine. Jolivet. 184. linear and therefore sensitive to the sign of the velocity. S. Collins. J. 1991. J. 1991). N. 31. Y. D. Phys. 504– 512. e Lance. J. Lance & Bataille. L.. E. REFERENCES Bittoun. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 7 PERSPECTIVES AND CONCLUSIONS Data interpretation and diagnostic improvements by using selection techniques (Hoult. Lemonnier and J. However. Leblond.. T. A systematic comparison with thermal anemometry and optical probes techniques will be next undertaken. Turbulence in the liquid phase of a uniform bubbly air-water flow. M. 1976. J. & Leblond. Fluid Mechanics.H. 1954. S.An approach to selective pulse analysis.. NMR is basically sensitive to fluid displacement and mean density (24). thesis. E.. 50(6). Rev. it is believed that this short introduction to NMR measurements has raised some interest and demonstrates the tremendous potential of NMR measurements in fluid mechanics. Real time blood flow imaging by spiral scan phase velocity mapping. J. & Le Gros. Gatehouse.. paper S7-Thu-C-58. M. J. Phys. 222. & Purcell. 1994. D. Nuclear spins in the Earth’s magnetic field. 1994. 1976) have not been addressed here for the sake of space limitation and are developed elsewhere (Lemonnier & Leblond. D.. 69–86. Ph. H. Firmin. 229–237. P. Durand. 723– 730.. Javelot. E.. 94(3).

J. Giornale di Fisica.S. (15/15) .H. 2004.. Teukolski. 46(3).. J. A. O. 1987. 28(3). 185– 195. F. Baudez. W. Rev.. Guilbaud.. & Flannery. P. & Sconza.. the free precession and the spin echoes of the terrestrial magnetic field. P. W. P. Fisica. Moucheront. 2002. A. 272– 286. Ravanello. Bertrand. Two experiments on nuclear magnetic resonance.. J. Vetterling.. the art of scientific computing. 50(3). Second edn. A. B. Direct determination by nuclear magnetic resonance of the thixotropic and yielding behavior of suspensions. 1992.C. Raynaud. S. Cambridge Unviersity Press. Lemonnier and J. T. Principles of magnetic resonance imaging. A. H.P.. Mex. Leblond Application of NMR to fluid flow measurements NURETH-12 Log: 061 Press. Rodriguez. 709–732. Journal of Rheology. Numerical recipes in fortran. & Coussot..