Electromagnetic flow sensors’

John Kanwisher and Kenneth Lawson

Woods Hole Oceanographic Institution, Woods Hole, Massachusetts 02543

Abstract
Flow sensors based on the principle of electromagnetic induction were investigated as
alternatives to commonly used mechanical devices utilizing rotors and propellers. Proto-
type sensors were constructed showing considerable promise. Measurement accuracy in
excess of 1 cm see-l seems feasible with devices suited to long term battery operation. The
inertial effects and many of the reliability problems inherent in moving part devices would
be overcome by USC of an electromagnetic sensor.

Ocean currents have been routinely mca- field (E = V X B ) . The magnetic field pro-
sured by differences in potential caused by duced by a coil depends on the electrical
the clcctrically conductive seawater moving current and the number of turns. The
through the vertical component of the power supplied to the coil varies with 12.
Earth’s magnetic field (von Arx 1950). Thus, if the sensitivity of a coil is to bc in-
Blood flow in animals is also determined by creased lo-fold, the coil power must go up
electromagnetic induction, the magnetic 100 times. Large coils are efficient, but
field being produced by a coil placed next sensitivity increases with the square root of
to the blood vcsscl. The method is also used linear dimensions, while volume varies with
industrially for monitoring flow in pipes. the cube. Constructional simplicity is there-
But the use of a locally generated field has fore a useful criterion for choosing coil size.
found only occasional application in ocean- Coils of IO-cm diameter and l-cm2 cross
ography (Olson 1972; Tucker 1972; Bow- section are convenient. With a DC power
den and Fairbairn 1956). The continually of 1 W the field from such a coil reacts with
disappointing performance of mechanical a water flow of 1 m set-l to produce a po-
current meters has motivated us to take an tential difference of about 100 PV at the
independent look at the possibilities of elec- electrodes. If 10 mW of power are used,
tromagnetic flow sensors. Here WC describe the electrode output will still be 10 pV.
first the theory involved in the design of Such potentials are readily measured.
such a sensor. Then we show what results It is impossible to get two electrodes to
one can get in an actual device. Finally remain within a few microvolts of potential
we discuss potential uses, such as current difference because of chemical uncertainty
meters, to which this kind of flow sensor at the electrode surface. They can easily
can be applied. have a static offset 1,000 times this large.
If, however, the magnetic field is periodi-
Theory cally reversed, the polarity of the electrode
The flow sensor ( Fig. 1) contains a cop- potential due to water flow will change,
per coil to produce a magnetic Eicld. A set while the static offset remains constant. The
of electrodes measures the voltage gradient electrode voltages can be detected in syn-
across the face of the coil which is induced chrony with the field reversals. The magni-
in the water when it flows through the field. tude of the change is a function of the ve-
Faraday’s principle of induction states that locity of the flow. Electrode errors caused
the voltage field induced is the vector by electrochemical effects and input am-
product of the velocity and the magnetic plificr drift are almost entirely canceled.
---- It is ncccssary only that these offsets do not
1 Contribution No. 3448 from the Woods Hole change during one complete cycle of the
Oceanographic Institution. This work was SUP- field. Further, reversing the field means
ported by National Science Foundation grant
GA-31987. that we can use AC amplifiers; large gains
LIMNOLOGY AND OCEANOGRAPHY 174 MARCH 1975, V. 20(2)
 Electromagnetic flow sensors 175

maximum current. When the field is re-

versed, the coil voltage very rapidly satu-
rates at the limit of available power supply
voltage. The current rises according to Ia
(1 - e+“) where x is the coil time constant.
When the current limit is reached, no fur-
thcr changes in current can occur, and the
magnetic field is stable. This technique al-
lowed cfficicnt use of the prototype coil at
a field rate of 100 Hz.
Electrical noise, which tends to mask the
signal, is contributed by the input ampli-
fier, the chemical uncertainties at the elec-
trode surface, and the resistance of the wa-
ter. While the last is the ultimate limit to
sensitivity, the noise in the input amplifier
e.m. SENSOR is most of the total, Amplifier noise in-
BOTTOM VIEW SENSITIVITY AXIS

I
creases at lower frequencies, but remains
fairly constant above about 100 Hz. This,
as well as fast response, mandates a high
ELECTRODE,COVER
field reversal rate.

The electromagnetic sensor

The flow sensor contains a coil of wire
CROSS SECTION
and a set of electrodes. The coil was wound
on a Lucite bobbin and cast in epoxy resin.
ELEiK;;DE
\\ FIELD COIL CABLE The Ag-AgCl electrodes were recessed into
the face oft the sensor. The electronics used
Fig. 1. Schematic drawing of a round e.m. with the prototype sensor provide a -t- 0.037
current sensor.
A field current for a magnetizing force of
50 A-turns. (DC coil power was 0.22 W.)
are possible without risk of saturation bc- From geometry and calculated field
cause DC effects are excluded. This detec- strength, the electrode voltage was expected
tion technique is critical to the operation to be about 36 PV m-l set-l. C’alibration in
of our electromagnetic (e.m. ) flow sensor. a tow tank showed a linear response to ve-
In a perfectly symmetrical coil, there locity. The sensitivity referenced to the
would be no voltage on the electrodes in- electrode voltages was 30 PV m-l set-l.
duced by the changing field. In real coils, Electronics-The electronics provides a
some imbalance is inevitable. An error volt- reversing field, an electrode amplifier capa-
age, indistinguishable from water flow, al- ble of responding to submicrovolt signals,
ways exists as the product of the rate of and a detector which demodulates the sig-
change of field and asymmetry. To avoid nals to produce a voltage level which is a
this, measurement of the electrode poten- function of water flow velocity only. A
tials is delayed until the field stabilizes. It functional block diagram is shown in Fig.
is possible to drive the coil with an alter- 2. The complete schematic for the elec-
nating voltage, but some changing field tronics is available from us in an unpub-
effect will exist however long the measure- lishcd manual.
ment at the electrodes is delayed. Thcrc- The input resistance of the “follower-
fore, the coil is driven with an electronic with-gain” configuration used in the elcc-
circuit which delivers an accurately limited trode amplifier is more than lo7 ohms, This
 176 Kanwisher and Lawson

SENSOR ELECTRODES

Fig. 2. Block
El--
diagram
OSCILLATOR
4 X FIELD
FREQUENCY

of the flow
LOGIC

sensor electronics.

is so large compared to the nominal 50-ohm Discussion-In considering the possibil-

resistance of seawater plus electrodes that ity of changing to an e.m. sensor in a spe-
salinity effects and aging produce ncgligi- cific application such as a current meter it
ble change in sensitivity. Use of feedback is important to see clearly what advantages
with operational amplifiers assures large, it has over prcscnt mechanical systems,
stable, and prcdictablc gain, The circuit which are finally starting to give data after
used has a voltage gain of 20,000, sufficient 15 years of effort. In fairness, one must
to raise the lcvcl of the voltage at the clec- note that roughly the first 8 years were
trodes to the millivolt range. Were the gain spent finding out how to put an instrument
much lower, drift and noise in the DC am- in the ocean and get it back. During that
plifier following detection would be bleak period many of us spent long weeks
troublesome. The electrode amplifier gain scanning the horizon for the orange floats
is divided between an input cross-coupled that weren’t there. Only in the past few
pair of operational amplifiers and a differ- years have returns been frequent enough to
encc amplifier following these. This ar- focus on the operation of the instrument
rangement provides very large rejection of itself. And only in the past year has it been
any but differential signals from the elec- realized how complex the mooring motions
trodes. on a surface buoyed mooring are. It is now
The dctcctor uses solid state devices in apparent that these motions can lead to
a manner analogous to a DPDT center-off several-fold overestimates of the energy in
switch. The logic signals that drive it are shorter time periods ( MODE 0).
obtained by dividing the coil period into Having come this far and expended this
exactly equal parts. After each field rcver- much time and money on the Savonius
sal, the switch is open so that changing rotor, one must have convincing arguments
field effects arc blocked. During positive for change. The following are some advan-
field half-cycles, the switch connects the tages of the e.m. sensor.
output (amplified electrode voltage) to the 1. It has a large linear dynamic range
dif fcrential DC amplifier, During negative and keeps a constant calibration. Flows
field half-cycles, the polarity of this con- over a range of at least l,OOO-fold can be
ncction is reversed. The result is that any measured.
signals that are not synchronous with the 2. The sensor is simple and rugged. The
field drive will produce zero average out- electronics required are straightforward and
p11t. relatively simple; this should contribute to
 Electromagnetic flow sensors

Fig. 3. Photograph of the complete flow sensor system with which the mcasurcments cited in this
articli were mad& -

reliability. The simplicity of the device is an important consideration, we see no limi-

illustrated (Fig. 3) by its construction on tation in increasing electrical current for
a 5- x 15-cm printed board. Without any greater sensitivity, although the heat gen-
adjustments after completion, this unit pro- eratcd in the coil will eventually cause con-
duccd a baseline error of less than 1 cm vection currents.
secl. 5. An electromagnetic flow sensor has
3. Almost any flow sensitivity can be had no moving parts, so it will not have unde-
by averaging the signal over a longer time. sirable inertial properties. With no bearing
This follows because the noise is random problems, mechanical limitations for both
and centered on zero. WC operated the de- high and low speed flows do not exist.
vice in a tub with the output connected to 6. The electrical output changes sign
a pen recorder such that full scale repre- with the direction of the flow. This means
sentcd 1.5 cm see-*. If the clcctrical time that a sensor of this type on a moored cur-
constant of the amplifier output is 30 set, rent meter will not tend to produce a net
the noise band is about 0.025 cm set-l. output in response to pumping motions.
With no flow, the recorder will draw a trace The device does not “rectify” such apparent
within such limits for many days. The ex- vertical currents. A single axis sensor is
trcmely small zero drift over this time could convenient for measuring flow along a
be attributed to the final DC amplifier, channel in that displacements along one
4. The power requirements are rcason- axis can bc summed electrically.
able for long term battery operation since 7. A second set of clectrodcs can be
one can trade time for sensitivity. The 30- mounted on the sensor at a right angle to
set 0.025 cm set-l noise band dcscribcd the first. The outputs of the two detectors
above required 0.22 W of coil power. For will bc the rectangular coordinates of the
a noise band of 0.25 cm set-l ( 10 times current flow, appropriate sign included.
larger) the coil power need only bc 2.2 mW. These coordinates contain complete infor-
The clcctronics need not consume more mation defining magnitude and direction of
than a few milliwatts. Where power is not flow in the plant of the sensor. Such a de-
178 Kanwisher and Lawson

vice is in use to monitor lateral as well as which make them worthy of study. These
fore and aft ship movements (Tucker et al. systems seem to 11s conceptually similar to
1970). e.m. devices. The choice between e.m. and
8. A determination of flow is completed acoustic flow sensors may rest on a balanc-
for each cycle of field reversal. It is possi- ing of engineering considerations not yet
ble to integrate the detector output in syn- apparent.
chrony with the field so as to eliminate any Errors-The device creates the magnetic
need to average large numbers of field cy- conditions in water which generate a flow
cles. The time of integration still sets the signal in the form of a square-wave voltage
bandwidth, however, so that time acuity at the electrodes. This is amplified and de-
and sensitivity must be balanced against tected to provide a DC level, the sign and
one another. For flows above 5 cm set-1 magnitude of which are a function of flow
one can see velocity detail with a time velocity. There are three kinds of errors
acuity down to at least 0.04 see; this may affecting the accuracy of this process: off-
make the device useful in turbulence stud- set errors or baseline drift, scale errors (i.e.
ies. sensitivity changes ) , and the catastrophic
Hydrodynamic properties-The flow prop- errors which involve dynamic limitation or
erties of the e.m. sensor may limit its use in device malfunction, and which usually re-
some applications. But a number of varia- sult in wildly inaccurate readings, or no
tions are possible to minimize such restric- readings. These errors are discussed below.
tions. For instance, to the extent that the Baseline clrift-Our device uses an AC
water next to the sensor is retarded by skin amplifier to magnify changing voltages to
friction, the indicated flow will be lower some arbitrarily large level; these relatively
than that in the bulk of the water. This can large changes are then demodulated to pro-
be overcome in part by making the sensor duce a DC level equal to the magnitude of
surface concave or by shaping the sensor as the change. Since demodulation must oc-
a toroid rather than a disk. The cutout cur at a finite level, the DC drift of the out-
center would allow free passage of the wa- put amplifier will affect zero stability. In
ter past the sensor. A toroidal shape would this instrument, a lm set-l current produced
probably be less sensitive to tilt and to a about 700 mV of DC output. The stability
biased indication of current flow resulting of the DC amplifier is about 5 PV deg-‘.
from any vertical pumping motion of the This implies that temperature effects from
mooring. this source are less than 1 X 10m5m see-l
Another sensor configuration would in- deg-l. Aging effects with DC amplifiers
volvc using two parallel field coils about a are such that drift is typically 15 PV
diameter apart. This would result in fairly month-l and this contributes further to
uniform distribution of magnetic field drift.
within the open “cylinder” defined by the A second source of zero drift exists in
two toroids. The struts connecting the long term unidirectional changes in the
toroids would support the electrodes. This electrode potentials. If the electrode differ-
would give a flow determination of the wa- ence voltage changes at sonic consistent
ter mass between the coils; very little dis- rate, an output will result that will look like
turbance by the sensor of this flow would a signal. For example, if the electrodes
be anticipated. Olson (1972) has reported change at a 1 PV set-l rate, the equivalent
good results with this double coil arrange- signal will be l/(2 x field frequency) times
ment. this. In the prototype, this would represent
Acoustic flow sensors-Flow sensors an error of 0.015 cm sccl. We find that the
based on various sound propagation tech- Ag-AgCl electrodes are within a millivolt
niques (pulse time-of-flight, continuous- of each other over long periods, so the ac-
wave phase-shift, doppler) have hydro- cumulated error from this source is appar-
dynamic properties and reliability potentials ently very small. If the amplifier is not
 Electromagnetic flow sensors 179

continuously powered, howcvcr, as might DC level close to zero, Flows past the sen-
be the case in devices intended for inter- sor in excess of some limit will cause non-
mittent sampling, bias currents from the linear amplification of the electrode diffcr-
input operational amplifiers can cause sig- ence voltage, This will be indicated as a
nificant errors as the electrodes adjust to a steady DC output far in excess of nominal
new equilibrium. This effect does not ap- scale. In practice this is no problem.
pear to be significant after about 15 set for
stabilization, Potential uses of the e.m. sensor
A third source of baseline error arises
from electrical lcakagc between the field Moored current meter-The e.m. flow
sensor seems to bc an appropriate device
coil and the water. If this resistance is not
for use in a moored current meter. Several
extremely high (or perfectly balanced with
respect to the electrodes, which is very un- configurations are possible, some of which
likely) the field drive voltage will divide could bc adaptations of existing instru-
across this leakage resistance and the elec- mcnts. For example, an e.m. sensor with a
single electrode set could substitute for the
trode plus water path resistance, causing
mechanical rotor in any instrument that
a signal which produces offset. Changes
in this resistance are likely, so that the off- lines up with the current. An output con-
set cannot be permanently nullcd. A leak- sisting of magnitude and direction results.
age of 10” ohms will cause an offset on the The resultant polar form cannot be aver-
order of 1 cm sec- l, We insulated the coil aged directly, but the system would be suit-
carefully and did not identify leakage as an able for stable moorings and slowly chang-
error source in the sensor described here, ing currents.
although some earlier and less careful ef- Another approach would involve a two-
forts resulted in large errors. axes e.m. sensor. The orthogonal outputs of
Scale errors-Constant scale factor dc- such a device can be directly averaged.
pends on stable coil current and geometry Such a device could cope with mooring
and unchanging gain in the sense electron- mo,vemcnts on a time scale similar to the
ics and detector. Coil current is established averaging period.
with reference to a zener diode and has a If bearing information is available as sine
temperature coefficient of less than 0.02% and cosine of compass angle ( as it would be
per deg C. Most of the sources OFbaseline with a magnetomctcr compass), the orthog-
drift have some small dependence on field onal components from the e,m. sensor can
drive level, so that care in regard to, field be multiplied by the bearing information
stability may be important even though pre- and the result averaged and stored. This
cision in measurement of magnitude may might be the most straightforward and re-
not be. liable approach to a true vane. However,
The gain of the electrode amplifier and existing vector averaging electronics can be
detector also affects the scale factor. With utilized if the e.m. rectangular components
reasonable design it is easy to feel confident arc first converted to polar form. This
in a scale accuracy of better than 1% over simulates the format of the rotor/vane com-
time and temperature. bination. A very important advantage with
Catastrophic errors-Apart from the ob- both systems lies in the essentially instan-
vious failure of critical components, the taneous ability to resolve the directional
component of the water velocity vector.
limits imposed by dynamic range of clec-
This occurs because a vane, with its atten-
trical devices are significant. The internal dant inertia and friction, is not required.
circuitry that we used can compensate elec- Magnetometer compass-Conventional
trode differentials up to about 12 mV. For compasses suffer some of the disadvantages
electrode offset potentials above this, the inherent to electromechanical devices.
device is completely inoperative. The out- They have fragile bearings, uncertain dy-
put will be some arbitrary and unchanging namic propertics, and are difficult to read
180 Kanwisher and Lawson

t ? .5 KNOTS b-1 HOUR--I

Fig. 4. Continuous 4%hr recording of the flow through the Eel Pond channel.

electrically. They are listed as the greatest ford 1970) show currents with a finely
source of trouble in a recent upwelling ex- stratified, rapidly changing structure. In
periment on the Oregon coast (Pillsbury et this sense the contribution of moored cur-
al. 1974). These considerations may point rent meters may be to confirm this com-
to the use of magnetometers using high pcr- plexity.
meability toroidal cores for directional in- Other applications-We have tested our
formation. The Earth’s magnetic field pro- sensor by monitoring the flow through the
ducts an imbalance in the magnetic core channel leading into Eel Pond here in
which can be electrically sensed: the re- Woods Hole. We had expected to see a
sult is an output signal which varies as the simple tidal movement. The record in Fig.
sine of the magnetic heading. This informa- 4 shows otherwise, The tidal flow is ob-
tion allows the orthogonal e.m. sensor out- scured by a short period (about 16 min)
puts to be resolved on a time scale of 1 set oscillation which is constantly present and
or less. The entire instrument case can ro- accounts for 10 times as much flow as the
tate randomly without afEecting the valid- tide. It varies 3-fold in amplitude on a
ity of the current data. schedule which we cannot correlate. The
A magnetometer of the type described record shown is full scale 2 25 cm see-l.
can operate with sufficient accuracy ( -t- 3 The cross section of the channel is well
deg or better) at low power levels com- defined (2.5 X 11 m). The integrated flow
patible with long term measurement prob- during the 8 min of a half cycle can be ex-
lems, i.e. 10 mW or less. We have a work- pressed as volume OFwater moving through
ing prototype which we are combining the channel. This volume distributed over
with the e.m. sensor to produce an all elec- the known area of the pond ( 49,000 m2)
tronic current meter with no moving parts. tells the change in height. Since the time
It may turn out that when oceanography constant OF the pond is short compared to
finally gets a working current meter, the 16 min, the same change in height must
real ocean will appear more complicated occur outside the channel. This periodic
than we can treat conceptually. Certainly movement up and down represents a long
the initial results of in situ potentials as an slow wave driving the water back and forth
index of water movement (Drevcr and San- through the channel. The waves shown in
 Electromagnetic flow sensors 181

Asterias in Eel Pond Channel

.3 knots for 3.5sec

= .53 displacement

channel=2.5 x lim=27.5m2
volume= 14.6m3

Fig. 6. Flow past the sensor when water in a

tub is stirred with a single stroke.

I - -13,5secl- in a tub which was stirred with a single

stroke of a paddle on the side opposite the
Fig. 5. Flow into Eel Pond as boat goes out.
sensor (Fig. 6).
Such. a stroke generates a large single
eddy and also imparts an overall rotary mo-
the record have amplitudes of 1.5 to 5 cm. tion to water in the tub. The eddy main-
Vineyard Sound outside the harbor is tains itself and rides on this overall circula-
about 6,000 m wide and 15 m deep. A tion. As the eddy passes the sensor its
seiche determined by these dimensions rotary motion adds on to that of this over-
( Neumann and Pierson 1966) has a period all circulation. The sensor therefore sets a
of 18 min. The constantly oscillating water greater instantaneous velocity. In Fig. 6
movement in the Eel Pond channel is thus many passages of the eddy past the sensor
probably due to such a seiche. The flow are apparent. With time both the rotation
sensor (or current meter) makes a useful of the eddy and also that of the overall wa-
wave-meter for such long period scichcs ter slow down.
and also for tides. The record as shown is We wcrc surprised to find such a record
the differential wave height and must be when we attempted to test the sensor in the
integrated if height is to be recorded di- laboratory. Our analysis of the mode of this
rcctly. circulation is based on observing particle
The record in Fig. 5 shows the inward movcmcnt in the water. The analysis of the
water movement in the channel when the circulation regime from the record itself
research boat Asterias passed out to the would bc much more difficult. However
ocean. The deflection indicates an average WC have gradually learned to trust the sen-
flow of 15 cm set-l for 3.5 set, which rep- sor when it generates such unexpected rec-
resents a 0.53-m displacement of the water ords. In every case whcrc we have con-
in the channel. This means then 14.6 m3 tinuously monitored flow we find features
of water moved in to replace the boat, more complicated than WC expected.
which thus must weigh about 15 tons, We We plan to USCthis electromagnetic sen-
present this not as a serious method of sor to monitor flow in the mouth of nets.
weighing boats but as an example of the This will tell the filtering rate and also give
rapid dynamic response of the sensor, an idea of the ship’s speed, a needed datum
By the direct approach of more power to not now available on Woods Hole ships.
the coil one can improve the signal-to-noise We arc also building sensors to do rout&re
and get a more rapid response from the monitoring of the flow in salt marsh creeks.
sensor. We propose it as a means of moni- WC are also combining the sensor with a
toring the .turbulent nature of a flow where magnctomctcr to give an output propor-
the cddics are larger than the sensor. As an tional to the windward component, starting
cxamplc we show the decay of circulation from the speed and heading, of a sailboat.
182 Kanwisher and Lawson

Many physical oceanographers presently NEUMANN, G., AND W. J. PIU~SON. 1966. Prin-
involved in the construction and use of cur- ciples of physical oceanography. Prentice-
Hall.
rent meters must receive a part of the credit OLSON, J. R. 1972. Two-component electromag-
for the work described here. Their con- netic flowmeter. Mar. Technol. Sot. J. 6:
tinued lack of interest in other than Sa- 19-24.
vonius rotor current meters has finally PILLSBURY, R. D., J, S. BOTTERO, R. E. STILL, AND
stimulated us to the present effort. We hope W.E. GILBERT. 1974. A compilation of ob-
scrvations from moored current meters.
their current agonizing reappraisal, growing School of Oceanogr. Data Rep. 57. Oregon
out of the disappointing MODE results, State Univ. Ref. 74-2.
will make them more receptive to alternate TUCKER, M. J. 1972. Electromagnetic current
methods such as e.m. sensors and magne- meters. Proc. Sot. Underwater Technol. 2:
53-58.
tome tcrs. N. D. SMITH, F. E. PIERCE, AND E. P.
GdLLINS. 1970. A two-component electro-
References magnetic ships’ log. J. Inst. Navig. 23: 302-
BOWDEN, K. F., AND L. A. FAIRBAIRN. 1956. 316.
Measurement of turbulent fluctuations and VON ARX, W. S. 1950. An electromagnetic
Reynolds stresses in a tidal current. method for measuring the velocities of ocean
Proc. R.
Sot. Land. Ser. A 237: 442-438. currents from a ship underway. Pap. Phys.
DREVER. R. G., AND T. SANFORD. 1970. A free- Oceanogr. Mctcorol. 11: l-60.
fall electromagnetic current meter. IEEE
( Inst. Elec. Electron. Eng. ) Conf. Proc. 19 : Submitted: 11 March 1974
353-370. Accepted: 12 November 1974