of Ae104aHomework#4
Development an indirect counterbalanced pendulum opticallever thrust balance for micro to
millinewtonthrustmeasurement.ByA.N.GrubiiandS.B.Gabriel:
Meas.Sci.Technol.21105101(2010).
AsummarybyGarrettLewisandDustinSummy.
A device is developed to indirectly measure thrust produced by microthrusters similar to those
used in spacecraft. The thrust balance is designed to provide a lowcost and reliable thrust
measurement for microthrusters with a thrust range of 103 to 105 Newtons. These measurements can
then be compared with calibrating measurements performed with a more expensive and accurate
system to estimate the absolute thrust. By using the simple system and correcting with comparisons to
the more complex system, measurements can be made much more quickly and cheaply than by relying
onthecomplexsystemalone.
The most accurate measurements using this kind of technique come from devices in which the
thruster is mounted directly to the pendulum itself. This avoids the uncertainties in the iontarget
interaction, but requires careful and expensive measures to mount the thruster to the pendulum
without causing undesired torques due to the wiring and fuel lines that must run to the thruster. To
avoid these expenses, the authors chose an indirect thrust balance setup as shown in Figure 1. The
advantages in simplicity are obvious the thruster remains fixed during the test negating the concern
over countertorques developed in connections, the hanging pendulum is in contact only at the pivot
point,andthecounterweightallowsforgreatersensitivitywithoutextendingthependulumarm(limited
byspaceinavacuumchamber).Thependulumreflectsalaserbeam,andthedeflectionismeasuredby
anopticalsensorwhichreportsthepositionofthebeamtothedatacollectionsuite.
The response of the resulting system is that of an underdamped harmonic oscillator to good
precision. An acceptable block diagram for the system is illustrated in Figure 2. The naturally occurring
value of is measured by the logarithmic decrement as shown in Figure 3a. In order to shorten the
settling time and decrease noise sensitivity, the authors added a ring magnet of 0.1T mounted a
distance behind the pendulum to induce magnetic damping. By varying the distance between the
magnet and the target, the authors were able to increase by up to 3 orders of magnitude, shortening
the settling time (authors definition) from hours to about 12 seconds. An example of the damped
responseisshowninFigure3b.
Finally, the authors characterized certain sources of noise inherent to the design. A significant
zero drift due to thermal heating was found immediately after thruster shutoff, which the authors
simply subtract from the steady state thrust measurement. Additionally, the authors measured the
power spectrum of noise in the system and attempted to relate resonances to environmental causes.
Certainlowfrequencynoisesourcessuchasthenaturalfrequencyofthetestfacilitysstructureandthe
frequencyofoperationofthevacuumpumpwereobserved.Todiminishtheinfluenceofthepump(the
50Hzpumpwasclosetothe56Hznaturalfrequencyofthependuluminonecase)theauthorsphysically
restrainedthemidpointofthetubingconnectingthepumptothechamber,forcingastandingnodeand
doublingthefrequencyofthetransmittednoiseto100Hz.
Ultimately, the authors are able to measure thrusts ranges from 1739 with no counterweight to
12.66 uN under a 500.00 g counterweight. The maximum resolution for the system is 3.80 nN at the
heavierweight.ThethrustrangeisshowngraphicallyinFigure4.Errorsintheresponseareprimarilyin
theformofuncertaintyrelatedtomassflowfromthethruster.Othersignificanterrorsareroughlyone
orderofmagnitudeweakerandincludethrusterlocation,balanceweight,opticalresolution,andoptical
pathlength.Thetotalquadratureuncertaintyisroughly4.01percent.
The accuracy of the balance is restricted by uncertainty over the nature of particle collisions
with the pendulum. The accommodation coefficient, which describes the relationship between the
actual and measured thrust, is very difficult to characterize and is therefore a great source of
uncertainty. A perfectly elastic interaction would yield an accommodation coefficient of that
increases as the collisions become more diffusive. The best estimates for such coefficients are
determinedusingdirectthrustmeasurementsforidenticalthrustersandarelistedinTable1.
The system was initially verified using a T5FO hollow cathode with argon fuel. Over the studied
rangesofthrust,astandarddeviationof0.135mNwasidentified overtheentiresystem. Theresultsof
these tests are displayed in Figures 5 through 7. Based on the measurements discussed above, the
authors estimate an accommodation factor of roughly 20 percent is appropriate, particularly for low
flow rates used in their analysis, and should be used to scale the raw data presented. The thrust
measurements obtained provide appropriate linear thrust increases with flow rate, but demonstrate a
specific impulse that falls off rapidly prior to stabilizing at flow rates of around 0.2 mg/s. In transient
studies, the magnetic damping provides a sufficiently rapid response; however, there is nonetheless a
slight lag of approximately eight seconds in system response due to the stabilization of fuel lines during
changesinmassflowrate.
Figure1.Thrustbalancesetup.
T
0
Dispersion
Effects
Tmeas
Pendulum System including
magneticfeedback
Optical
System
dX
Electrical
System
V0
Figure2.BlockDiagramofSystem
=Tmeas[LThrust/(MgLcogI2+j)]
Figures3aand3b.Systemdamping
Figure4.ThrustRangeandResolution
Table 1. Measured values of thrust compared to the theoretically calculated thrust
Gas (mg s1)
Cold xenon (298 K) 0.717
1.263
1.585
Cold argon (298 K) 0.2
0.6
1
Hot xenon (1500 K) 0.936
0.533
0.261
Hot argon (1500 K) 0.25
0.5
1
meas (mN) total (mN) factor
0.216
0.2
1.079
0.35
0.353
0.99
0.427
0.443
0.963
0.141
0.12
1.17
0.391
0.361
1.082
0.577
0.602
0.958
0.146
0.515
0.284
0.092
0.293
0.313
0.048
0.144
0.332
0.194
0.272
0.712
0.365
0.545
0.668
0.631
1.091
0.577
Figure5.Thrustmeasurements
Figures6and7.Specificimpulseandtransientresponse.
Ae104a Homework #4
A Direct-Measurement Thin-Film Heat
Flux Sensor Array by J. Ewing, A. Gifford, D.
Hubble, P. Vlachos, A. Wicks, and T. Diller: Meas. Sci.
Technol. 21 105201 (2010).
A Summary by Karen Oren and Stacy Levine
November 11, 2010
A heat flux sensor is a device used to measure convective, conductive, or radiative heat transfer.
These measurements are crucial for thermodynamic research in a broad range of applications,
such as monitoring material stresses in a jet engine. There are several types of existing sensors
in this field, though none are ideal. The Schmidt-Boelter design is a traditional single point
gauge, but it is often better to have an array of sensors that provide a distribution of heat flux
measurements. Schultz and Jones designed a thin film heat flux array (HFA) that utilizes a semi
infinite conduction model shown in figure 1, and though it is useful for extremely high
frequencies (>200kHz), it is limited to short time events. Inconsistencies in the glue layers that
fasten gauges such as this to the test objects provide a significant source of uncertainty. A
double sided sensor with an internal thermal resistance layer, shown in figure 2, eliminates the
effects of the glue and allows for a direct calculation of heat flux using = ( ) , where k
and are the thermal conductivity and thickness of the thermal resistance, respectively. Epstein
et al has an existing double sided gauge produced with nickel RTDs on either side of a Kapton
film. These devices have calibration problems and the internal resistances vary over time,
distorting the RTD temperature measurements and therefore the heat flux values.
Figure 1: Semi-infinite geometry model.
Figure 2: Double-sided heat flux gauge.
A new HFA has been developed at Virginia Tech using nickel/copper differential thermocouples
that are vacuum-deposited onto either side of a thin Kapton polymide film. When the
thermocouples are connected, the resulting voltage is directly related to the heat flux:
=
1
,
where Se is the Seebeck coefficient of the thermocouple materials. The HFA has a voltage
output that is self-generated, so the sensor passively measures the heat flux from a heat source,
through the gauge, and out to a sink. Copper and nickel are selected for the design because the
combination provides a high Seebeck number, resulting in a high sensitivity to changes in
temperature. Similarly, Kapton is selected for its low thermal conductivity which maximizes
the change in temperature. Kapton films are readily available, can be formed to a curved
surface, and have a good adherence with copper. Although thinner films have faster response
times, size and shape limitations are set by the machining of the masks. No flux (acid) is used in
the soldering process to prevent dissolution of the films, and the gauges are kept in argon to
prevent oxidation. The final encapsulation in Mylar film thermally and electrically insulates the
gauge without affecting the calibration.
Measurements were taken with National Instruments DAQ boards and LabVIEW. Figure 3
shows custom circuitry that was designed with an amplifier to provide ample resolution and a
low pass filter to prevent aliasing. The first prototype was mounted on a metal plate that
2
unintentionally created a capacitor with the metal films of the sensor, resulting in a voltage
offset. This was corrected with low impedance resistors placed between the HFA output and the
analog DAQ ground. This allowed the charge built up in the capacitor to bleed off, while not
affecting the output results.
Figure 3: Diagram of the HFA capacitance and the instrumentation amplifier circuit.
The convection calibration facility at Virginia Tech was used to calibrate the HFA. A single
point heat flux microsensor (HFM) was used as a reference for determining the convection
coefficient, h. Using the known h value in conjunction with the temperatures measured in the
thermocouples, the heat flux through the test gauge was calculated using = ( ),
where T is the free stream air temperature, and Ts is the surface temperature of the gauge. The
sensitivity of the HFA was determined by comparing voltage output, Vo, to heat flux using:
=
=
.
( )
This convection calibration was performed at seven different pressures, each of which resulted in
a different convection coefficient. The average sensitivity for the seven tests was calculated to be
423V (W cm-2)-1. To validate these results, a conduction calibration that determined average
sensitivity by comparing the HFA output to a HFM output was performed at different junctions
in the array. These tests resulted in an average sensitivity of 41.82V (W cm-2)-1 which is
similar to the average value from the convection calibration.
The sensitivity of the array is dependent on the material properties of the resistance layer as well
as the thermoelectric response of the thermocouple pairs. Thus it was important to determine the
Seebeck coefficient of the thin-film nickel/copper thermocouples since it was likely that the thin
films would behave differently than typical bulk metal wire. Various temperature differentials
were applied to the metals and the coefficient was calculated by = /, where Vo
represents the thermocouple voltage output. The average value for the coefficient was 23 V C-1
which is close to the value for bulk metal wire. The values for the sensitivity and the Seebeck
coefficient were then used to calculate the thermal conductivity of Kapton:
=
,
where is the thickness of the Kapton layer. The calculated conductivity was 0.28 W (m K)-1
for the type E Kapton that is used in the array.
3
The time response for the array was
determined by removing and replacing a
mechanical shutter in front of a laser aimed at
the array. Figure 4 shows the array output.
The symmetry between the signal rise and fall
suggest that the array has a linear first order
response while the return to the baseline value
is characteristic of the differential
thermocouple. The rise and fall times were
calculated for 5 trials and the average response
time was 323 ms with the uncertainty
including the effects of the partial heat flux
experienced by the array.
Figure 4: Response of the HFA to a step input
The effectiveness of the HFA was
demonstrated in an experiment that
looked at the effects of free stream
turbulence on heat transfer in a
stagnating flow over time. Velocity was
measured with a laser that tracked the
motion of reflective particles in the
flow. The turbulence had a time scale of
1 s which was suitable for the response
time of the array. The resulting heat flux
and velocity measurements for a single
point in the flow are shown in figure 5.
In general, the heat flux measurements
were well correlated with the velocity
measurements, as expected.
Figure 5: Heat flux and normal velocity unsteadiness for turbulent flow
stagnating on a flat plate
The new thin-film heat flux array improves upon previous models in several aspects. The Mylar
encapsulation enables use in most fluids, and the direct measurement technique allows the array
to be used on any type of surface. The array allows for the direct measurement of the heat flux at
multiple locations simultaneously with the only source of noise being degradation within the
nickel thin film connections. This should be the focus for future work on this sensor. The time
constant and sensitivity of the array make it suitable for a variety of experiments that seek to
determine the time dependent heat flux of some surface or flow and thus it is a valuable new
technology for thermodynamic research.
4
Ae104a Homework #4
Direct Model Torque Sensor for Model Wind Turbines by Hyung Suk Kang and Charles Meneveau
Meas. Sci. Technol. 21 105206 (2010).
A summary by Paul Anzel and Thomas Vezin
As society has moved towards generating more of its energy renewably there has been a great deal
of interest in getting the optimal performance out of wind farms. To get this requires much
experimental testing, commonly done by creating of scale models in a wind tunnel; avoiding the cost
and logistical issues of building large arrays outside and allowing for easy testing of a variety of wind
speeds and other atmospheric conditions.
As we know from fluid dynamics, the motion of fluids and solids can be precisely scaled provided
various dimensionless values (e.g. the Reynolds and Peclet number) remain constant, so the
mechanical motion of a wind farm can be easily scaled up. However, to measure the output power
for a given turbine the usual approach taken is to find the output power of an internal generator.
This is easily done (P=IV) but is problematicthere is no reason for the generator to scale accordingly
as well and in fact it often doesnt. Energy losses due to resistive heating, magnetic induction, and
friction will all depend on the construction of the dynamo, and so all confound efforts to find the real
output power. In order to get a more accurate understanding of how well a wind farm works it would
be better to look at something mechanical.
The approach proposed by Kang and Meneveau does just that. From the conservation of angular
momentum we know that for a given torque provided by the turbine blades, the generator will apply
the same torque to its supporting structure if its to keep from spinning as a whole. Kang and
Meneveau use a system of strain gages (see Figure 1) to measure this torque T, from which the
mechanical power can be determined by the formula PWT = TWT where is the angular frequency of
the rotor.
A block diagram of the full system is shown in Figure 2. The wind drives the turbine, generating a
torque TWT (minus some internal friction). This drives a DC motor (acting as a dynamo) generating
some electric power (Pelec). The motor would be free to rotate but is held in place by the torque
measuring system (seen in Figure 1) with two strain gages on either side of a metal plate. These
strain gages are placed on either side of a Wheatstone bridge (positions 1 and 2) with two more
unattached strain gages placed on the turbine to cancel out any temperature effects. The output
voltage is fed into an amplifier and the torque is inferred based on calibration with known torques
(by hanging dead weights on a horizontal rod attached to the shaft, see Figure 3) giving a linear
interpolation. Meanwhile, is determined by shining a laser at the blades and counting how many
times the beam is deflected (using a TTL system) over a given unit of time. The measured torque is
multiplied by to determine PWT. We thus have two modes of measuring turbineswe can find the
mechanical torque by direct measurement or infer it from the electrical power given off.
The interpolation which transforms the voltage measured in the sensor or current in the motor to a
value for the torque is also the main source of error among the following undesirable effects:
-
Temperature variations introduce errors in the bronze plate deformation value measured by
the strain gages. However, this error is easily corrected by the two other unattached gages.
-
-
-
The stick slip effect: one of the most important parts of the torque sensor is the contact
between the rod and the bronze plate, where the applied torque is converted into a
mechanical deformation which can be measured. This contact is not perfect, though, and the
rod can slip and therefore produce a non-negligible uncertainty which is very hard to
quantify. This effect can be reduced during calibration by loading and unloading the
reference torque several times, but its still present during actual measurements.
Another mechanical source of error is the friction inside the ball bearings. It prevents the
motor from rotating for torques bellow 0.127mN.m, therefore introducing a lower limit to
the range of possible measurement.
The calibration uncertainty.
These calibration errors need to be further studied. Indeed, as we said, they are the main sources of
uncertainty through the interpolation process as shown in table 1. Several points have to be taken
into account when evaluating the value of calibration uncertainty. The most important ones are:
-
The value of the different reference torques used to plot the interpolation curve (UT in table
1).
The measurement of the voltage in the torque sensor (UAD and Urep in table 1).
The linear fitting (Ufit in table 1).
The two first points are minor and are mostly negligible compared to the error introduced by the
linear fitting. Though the relation between the voltage measured in the torque sensor and the actual
applied torque appears linear in a good approximation, a slight oscillation was observed, probably
due to the stick-slip effect. Consequently the relative uncertainty of direct measurement of torque
is non-negligible for small torques whereas it decreases quickly as soon as we increase this torque
and the measurement process becomes very precise for values bigger than 1.5mN.m. This lack of
precision for small torques introduces another limit beyond the mechanical one (0.127mN.m): it does
not seem reasonable to use the sensor if we need a good precision for torques smaller than
0.4mN.m.
This is all the more true when we want to use the indirect operating mode (measuring Pelec) of the
sensor. In determining the mechanical torque from the electrical power a second linear interpolation
was taken, further increasing the lack of precision (table 2).
If the device is used within its range of efficiency with torque greater than 0.3-0.4mN.m (but not
enough to break the system!), the results appear to be very accurate and precise. And as shown in
Figure 5, there is a noticeable qualitative difference in the output power observed by finding the
torque versus reading out the electrical currentthe peak value for mechanical power is at a lower
frequency than the peak value of electrical power. If used in the mode where we extrapolated the
mechanical torque from the current produced by the DC motor, the sensor becomes very interesting
by its simplicity and by the possibility to follow the evolution of the power extracted from the wind.
Fig 1 (left) Diagram of DC motor and torque sensor
Fig 3 (right) - Torque sensor calibration as a function of the output from the bridge voltage. Note the small oscillations.
Fig 2 Block diagram of experiment
Table 1 (left) - Relative uncertainty in the torque calibration using the torque sensor
Table 2 (right) - Relative uncertainty in the torque using current measurement
Fig 5 Output power measurement at different wind speeds. The lines with symbols are from mechanical torque
measurement, the lines without are from electrical current measurement.
Ae104a Homework #4
Inuence of crack length on crack depth measurement by an
alternating current potential drop technique
Manoj K Raja, S Mahadevan, B P C Rao, S P Behera, T Jayakumar and Baldev Raj
Meas. Sci. Technol. 21 105702 (2010)
A summary by
Monica Martinez-Ortiz and Duvvuri Subrahmanyam
Non-destructive evaluation (NDE) techniques play an important role in ensuring manufacturing quality and making in-situ assessments of structural components. They are mainly
used to identify structural defects such as cracks due to fatigue or corrosion. Early detection
of these material discontinuities is essential to maintain manufacturing quality and structural
integrity. A commonly used NDE technique for sizing surface cracks in metallic structures is
the potential drop technique. Crack depths are estimated by driving electric current through a
specimen surface and measuring the voltage drop across the crack (Figure 1a). This technique
is classied into two types: Direct Current Potential Drop (DCPD) and Alternating Current
Potential Drop (ACPD). Although both types are employed in various elds, the ACPD technique has proved to be a more reliable depth measurement method. This is mainly because
direct current penetrates much below the material surface, and hence surface crack depth measurements are aected by the specimen geometry. Whereas in the case of alternating current,
the penetration is only up to a depth ( ) known as the skin depth, which is of the order of
a mm for frequencies in the range of a few KHz. A lower current in ACPD measurements is
sucient to achieve a given sensitivity, as compared to the DCPD technique. Thus, the risk of
ohmic overheating of the specimen is signicantly reduced. Due to its accuracy and low-cost
eectiveness, there have been several experimental and theoretical studies on its applications
and possible improvement (see article for references).
The main objective of this study was to analyze the inuence of crack length (L) and crack
depth ratio (L/d0 ) on crack depth estimates using ACPD. The experimental set-up consists of
3 SS plates of lengths 10 mm, 20 mm and 40 mm, each of which was introduced with a series
of electro-discharge machined notches of various depths (from 1 mm to 12 mm) to simulate
cracks. The notch width was set to 0.5 mm to ensure electrical insulation between the notch
faces. The actual notch depth d0 after machining was measured optically by using a vernier
caliper with a needle attachment. Notch depth measurements were then carried out using an
ACPD crack depth meter (RMG 4015 - Karldeutsch Germany) with a driving current of 0.3
A at a frequency of 3.5 KHz. The instrument was calibrated on a 100 mm long notch with a
continuously varying depth from 0 mm to 10 mm. AC was driven between electrodes A1 and
A2 (Figure 1a). The dierence of potential between the electrodes B1 and B2, separated by a
constant distance s of 2 mm was measured. Assuming an uniform electric eld and thin-skin
regime (/d << 1), the potential drop in a crack-free region is given by the following relation:
V0 = ks
(1)
where k = I/A with , I and A denoting the materials resistivity, the current and the crosssectional area through which the current ows, respectively. Across a crack of depth d, the
voltage drop can be written as
Vc = k (2d + s)
(2)
1
By combining equations (1) and (2), the crack depth d is determined to be
d=
s Vc
1
2 V0
(3)
The measurements were classied into two classes: shallow notches (d0 > 3mm) and deep
notches (d0 < 3mm). Ten depth measurements were made for each notch and the mean was
taken to be the measured notch depth. For deep notches, the error in depth estimation using
ACPD varied from 6.5% to 45% as the aspect ratio decreased from 12.9 to 1.1 (Figure 1b). For
shallow notches, the reported error was consistently high (18.2% 20.3%) and it was attributed
to relatively large /d ratios. In general, the deviation from the actual depth d0 increased as
the notch length decreased. Therefore, a lower aspect ratio translated into a higher error in the
measurements. The estimated depths for all the notches are presented in Figure 1c.
A new model for current ow was proposed to explain the inuence of the crack aspect
ration on the measurements. For nite crack lengths, the current has an additional parallel
resistive path of length 2L between A1 and A2 as shown in Figure 1d. Accounting for this, the
voltage drop across the crack can be modied as
Vc,l
1
1
+
2d0 + s 2L
1
=
2L(2d0 + s)
2d0 (2L s) s2
=2
+ s = 2def f + s
2L + 2d0 + s
2(2d0 + (2L + s))
(4)
where def f is the depth estimated by the device. The actual crack length is then given by
d0 =
s
2
Vc,l
V0
1 +
1
sVc,l
2LV0
sVc,l
2LV0
(5)
Note that as L , the above expression is the same as (3). Figure 1e shows the plots of ACPD
measurements against d0 and def f against d0 . Both follow a similar trend. The measurement
device is in reality measuring def f , from which the actual crack length can be obtained from
(6). By taking the dierence between d0 and def f to be an error with reference to d0 , we have
s
(1 + 2d0 )2
d0 def f
=
s
L
d0
(1 + 2d0 ) + d0
(6)
The theoretical and experimental variation of error with aspect ratio is shown in Figure 1f.
A correlation coecient of 0.85 between the experimental data t and the theoretical curve
was reported. The authors attribute the observed deviation to the fact that ACPD instrument
calibration was not done on an innite length notch (bias error). Also, the eective cross
sectional area A diers between the cases.
The experimental and theoretical results were consistent and lead to the conclusion that
the error in crack depth estimation using the ACPD technique increases with the decrease in
aspect ratio. We agree that the new model provides a better prediction of the crack depth for
nite surface cracks. However, the assumption of an uniform electric eld (Equations 1 and
2) is crude. A better model can be developed by considering the current density distribution
across the length of the crack. Also, the authors do not consider the precision associated
with the ACPDs current source and voltage measurement units in trying to understand the
observed errors. There might also be considerable inductive impedance that aects the voltage
measurements, partly explaining the deviation of observed notch depths from the actual values.
2
(a) Sketch of the ACPD technique
(b) Error in depth estimation for dierent notch
lengths (Raja et al )
(Raja et al )
(c) Comparison of the notch depth estimated using (d) Additional current path (current path 2) available
ACPD for 10 mm, 20 mm and 40 mm long notches for a nite length notch in a component (Raja et al )
in SS plates (Raja et al )
(e) Notch depth estimated using Equation 4
(Raja et al )
(f) Variation of error in def f with the aspect ratio for
experimental and theoretical cases (Raja et al l)
3
Ae104a Homework #4
Measurement of Liquid Crystal Film Thickness Using Interferometry
by F. Kossivas and A. Kyprianou:
Meas. Sci. Technol. 21 105707 (2010)
A summary by Mike Rauls and Charles (Stan) Wojnar
Measuring the thickness and the surface topography of transparent thin films is important
for materials used in nanotechnology. In particular, liquid crystals (LCs) used in this work are
useful in display devices, sensors, detecting surface abnormalities, and discriminating healthy
from malignant biological tissue. LCs were used in this work due the ability to manufacture LC
films that are close to being isotropic with respect to their index of refraction.
Using inteferometry to measure thicknesses less than 1 m is difficult using current
methods. So instead of using interferometry to measure thicknesses less than 1 m, atomic force
microscopy (AFM) is used. AFM works by having a probe attached to a cantilever traverse the
surface the of film. Using AFM in tapping mode, which was used in this work, involves
oscillating the cantilever at a constant frequency. When the cantilever is close to the surface of
the film, the intermolecular forces between the probe and the film become strong enough to
change the frequency of the cantilever. The AFM machine constantly adjusts the height of the
cantilever above the film in order to maintain a constant frequency of oscillation. Since the
cantilever remains oscillating at a constant frequency, the cantilever will always be at a constant
height above the film. As the probe traverses the surface of the film, the height is recorded and
then the film thickness and surface topography can be extracted. The problem with AFM is that
it is slower and more costly than interferometry and you also run the risk of damaging the film if
the AFM probe penetrates the film.
Interferometry thickness measurements of thin films are based on the interference
patterns of reflected light after it traverses a distance or passes through a sample specimen. For
films thicker than 1 m, the reflections from the top surface of the specimen and the surface of
the substrate are sufficiently separated to measure the fringe patterns. Optically transparent films
do not reflect much light back to the optical sensor, so the interference pattern describing the
surface topography is difficult to resolve, contributing to errors in image processing. The issue of
fringe resolution is further complicated by the fact that the LC films in the experiments are of
sub micron thickness, and the fringe patterns of the top surface of the film and the substrate
overlap (in Figure 1, points F and D converge). The intensity of the light beam reflected from the
substrate dominates the intensity of the beam reflected from the top of the specimen making
direct computation of the thickness impossible.
The measurement process described in Kossivas et al. provides a method to measure the
thickness of thin films optically as opposed to the previously necessary AFM methods listed
above. Kossivas et al. propose that since the optical sensor can only register the retardation of the
speed of light as a perceived increase in optical path, so that the application of a thin film on the
surface of the substrate will appear as a step down that the coated region will appear lower than
the uncoated region, which is clearly not the case.
It is known that the LC film possesses a specific index of refraction that can be computed
theoretically, making it a good candidate for the confirmation of this experimental technique.
Individual LCs are optically anisotropic depending on the light direction and propagation with
respect to the optical axis. In the experiments in the paper, the LCs are randomly oriented in the
film. Thus, the overall index of refraction takes on a value in between the minimum and
maximum values here it is 1.58. No analysis of molecule alignment was performed, but the
LCs were estimated to be parallel to the substrate with no preference of horizontal orientation
with the alignment becoming more decoupled with increasing thickness (Kimura et al. energy
arguments). Therefore, the situation is similar to the isotropic case since the light impinges on
LC molecules in all directions.
With the optical properties of the film known, the appropriate corrections can be made.
For a an approximately constant substrate roughness (mean height), the height of the film above
the substrate, Z, can be computed as
Z = (n-1)-1(zmean-zmeasured).
As a consequence, the thickness of the film can be measured from the reflection of the substrate
through the film. Figure 2 illustrates the correction process.
Depending on the properties of the film, the surface topography can mimic the substrate
(Figure 3) or the upper boundary can smoothen since the surface molecules of the film can
decouple from the influence of the substrate (Figure 4). In the decoupled case, the roughness of
the substrate appears to be increased by a factor of n. This difference in measured roughness is
an indicator to the data analyst that the molecules on the top surface of the film have become
decoupled from the influence of the substrate.
As mentioned above, this measurement technique is suitable for films thinner than 1 m.
When the values recorded with the interferometer were compared with data taken by AFM,
which was assumed to be correct, a deviation on the order of 4.4% was computed. A thin film
was analyzed and measured to be 581 nm thick, and it was shown to follow the topography of
the substrate. Sources of error for this measurement include variation of substrate roughness,
variation in the index of refraction of the material, vibration of the interference mirror and the
sample, and resolution errors of the camera. A linear block diagram is presented in Figure 5. This
linear approach may not be appropriate since Kossivas et al. did not completely describe the
experimental set up used to complete the measurements.
The method developed in this work is useful for applications in quality control for
transparent thin film devices, where measuring film thickness and surface topography quickly
and inexpensively is important. Even though the technique in this paper is limited to transparent
materials, requires assumptions about material indices of refraction, and requires assumptions
about substrate surface topography underneath the film, this technique is still useful because it is
significantly less expensive than AFM and still results in comparable measurements with AFM.
Figure 1. Schematic representation of operation of inteferometry.
Figure 2. Image correction methodology.
Figure 3. Substrate and film exhibit same
roughness.
Vibration of
apparatus
Fringe
pattern
Variation of
substrate
roughness
Figure 4. Substrate and film exhibit different
roughness.
Variation
of index of
refraction
Index of refraction
Discretization
(resolution) error
Figure 5. Block diagram.
Image correction
Height of
surface
Ae104a Homework #4
Development of a micro liquid-level sensor for harsh environments using a periodic heating technique
by J. Hong, Y. S. Chang & D. Kim
Meas. Sci. Technol. 21 105408 (2010).
A summary by Cheikh Mbengue and Gina Olson
The goal of the authors of this paper was to create a new, easily commercialized sensor to determine
the amount of oil in an industrial compressor, as a certain amount is required for the compressor to
function properly. More generally, their goal was to determine liquid level under harsh conditions, which
they classified as environments with high temperature, high pressure, liquid flow, mechanical vibration
and limited space. Liquid level detection is a mature field, but the harsh conditions within a compressor
make currently available options undesirable or inappropriate. The authors listed six existing sensors, and
gave a reason why each is unsuitable; a summary of that list is given below.
1. Sight window: a simple and effective approach, but difficult to automate.
2. Optical sensors: films of viscous liquids form on the surface, negatively impacting performance.
3. Electrical sensors: sensitive to environmental noise and contaminants.
4. Acoustic sensors: sensitive to gas bubbles in the fluid.
5. Conventional hotwire: easily affected by environmental flow.
6. AC heater: closest to design proposed in paper, but is too bulky for a limited space.
The sensor design proposed within the paper is constructed of thin films of gold (thickness of 300
nm) and chromium (thickness 30 nm) patterned on a glass substrate (dimensions 1.5 cm by 1.5 cm by 500
m). Four electrodes are attached to the sensing portion to allow the application or current and
measurement of voltage.
The authors applied an alternating current with frequency , and measured the voltage drop across
the sensor. The authors used the 3 measurement technique, which states that a current of frequency
produces a temperature oscillation of frequency 2 (though the amplitude is determined by the thermal
properties of the material and surrounding fluid or gas). Moreover, it states that the third harmonic of the
voltage oscillation is useful in determining the thermal properties of the system; accordingly, the authors
used a lock-in amplifier to filter out all but the third harmonic of the voltage signal. The amplitude of the
temperature oscillation is given by:
The sensor design takes advantage of the large difference in the thermal properties between liquids
and gases, e.g. liquids typically have much higher thermal conductivities. Rather than continuously
tracking the liquid level then, the authors are conducting point sensing, determining the phase of the
material outside of the sensor. Instead of looking for a particular temperature oscillation then, the authors
simply look for a large jump in the voltage oscillation, which would indicate a phase transition.
The authors carried out an analytical sensitivity analysis, which revealed that the sensitivity of their
measurements was most affected by the thermal properties of the fluids in question. The sensitivity was
calculated as the thermal conductivity and specific heat capacity varied. Figure 1 shows the amplitude
and phase ratio of the temperature change in the gas and in the liquid (essentially the sensitivity). In
general, the sensitivity increased when the thermal conductivity and the heat capacity of the fluid
increased. At low viscosity, the sensitivity increases with increased current frequency, and the response
time decreases, but the amplitude of the temperature (and thus voltage oscillation) also decreases.
However, at high viscosity there was a considerable increase in the response time of the system.
There were two main experiments conducted by the authors. The first was a flow test which
consisted of a magnetic stirrer in the fluid that spun at 1100 rpm to induce circular flow in the liquid. The
sensor was repeatedly lowered into and raised from the fluid and the voltage was read by the lock-in
amplifier. The fluids used for this experiment were de-ionized water, ethanol, and ethylene glycol.
The results of the flow tests show near insensitivity to the flow conditions (Figure 3). Indeed, in deionized water, the average change of the sensor signal due to the flow effect was 0.42mV, which was
approximately 0.3% of that of the no flow condition. The results with ethanol paralleled that of the deionized water. The average change of the sensor signal due to the flow effect was 0.53mV, which was
again approximately 0.3% of that of the no flow condition. The authors cite boundary layer theory to
explain the apparent insensitivity of their sensor to flow conditions. They attribute this insensitivity to the
short thermal penetration depth, which is always less than the thickness of the momentum boundary layer.
In these experiments, the authors recorded fast response times around 27ms. The operating range in
which there is this notable insensitivity to flow conditions is within a current frequency in the range of 5 5000Hz. The de-ionized water, for example, was operated at 1KHz. Operation outside this range would
lead to increased sensitivity to noise in the signal.
During the presentation, the question was raised about how the geometry of the flow would affect the
level reading. The authors did not directly address this issue; however, it stands to reason that the design
is for pseudo one dimensional-type flows. Since, the experimental apparatus allowed for visual
observation of the sensor being dipped in the fluid, all the authors needed to do was to ensure that the
sensor was fully immerse thus negating the issue of the geometry of the flow.
The second experiment was the high temperature/high pressure experiment (Figure 4). Here, the
authors used the experimental set-up shown in Figure 2 to generate the high temperatures and high
pressures that one would find in a typical compressor. Polyvinyl ether (PVE) oil was pressurized with
Refrigerant R410-A and the tests were conducted at 80 C and 3.82MPa and 2.35MPa. A low temperature
and low pressure test was also conducted for comparison purposes, as well as to simulate compressor
conditions at start-up.
The higher viscosity and thermal properties of PVE had a limiting effect on the results of the high
pressure, high temperature tests. For example, the signal fluctuations were approximately 10% and the
response time was now 2.7s. Nonetheless, the changes in the amplitude of the signal were still marked
enough to differentiate when the sensor was in the fluid and when it was not, thus fulfilling the
requirements as a point level sensor.
At low temperatures, there is a further increase in the viscosity of PVE. This had the effect of leaving
a residue on the surface of the sensor, reducing the amplitude of the signal. Additionally, the rise time
increased significantly reaching as high as 14s. Notwithstanding, the authors believe that there is still
some practical application of the point level sensor in this environment.
In conclusion, the authors did that which the set out to do. They developed a point level sensor, using
the 3 method; that is simple, small and can be used in their definition of a harsh environment. However,
we found that they still have some of the same issues that they claimed to be faults in their competitors.
One example is the effect of residue on the sensor. Another is the fact that the sensor will need to be
complemented by a lock-in amplifier. This means that whereas the sensor itself maybe cheap, the
requisite components of the entire system may by expensive and difficult to develop or set up.
Figure 1. Showing the results of the sensitivity analysis.
Figure 2. Left: Flow test apparatus Right: Harsh environment apparatus.
Figure 3. Left: Signal Amplitude of various liquids. Right: Results of Flow test for DI water.
Figure 4. Left: Results of high temperature/high pressure test. Right: Low Temp/Pressure tests.
Ae104a Homework #4
Infrared Thermography of Solid Surfaces in a Fire by Melendez et al.
Meas. Sci. Technol. 21 105504 (2010).
A summary by Eddy Seetho and Neal Bitter
This paper discusses infrared thermography of solid surfaces in a fire, a technique which
is useful for evaluating the fire resistance of materials for Aerospace and Mechanical
Engineering applications. In this method, traditional infrared camera technologies are paired with
computer filtering to generate the temperature contours of a solid behind a fire, as demonstrated
in Figure 1.
In the past, thermocouples were used to measure temperatures on such surfaces; they
interface directly with the surface allowing measurements without any significant filtering or
post-processing involved. That being said, thermocouple systems have several major technical
limitations. Because thermocouples can only measure at discrete points, it is difficult to
determine the spatial temperature variation unless many thermocouples are used. Thermocouples
can also create a heat conductive interface that affects the surface temperature where they are
installed. Finally, there is a tradeoff between the rigidity of a thermocouple and its response time;
larger thermocouples are more rugged in harsh conditions, but have a greater thermal inertia that
hinders responsiveness to rapid temperature change. On the other hand, smaller thermocouples
respond quickly but may separate from the surface during a heat test. To avoid these problems,
the authors of this paper have turned to infrared thermography.
Infrared thermography takes advantage of the fact that all surfaces emit radiation, and the
intensity of this radiation at each wavelength is uniquely determined by the surfaces emissivity
and temperature. Infrared cameras characterize a surfaces emission spectrum by various
methods; in some cases cameras use photodiodes that are sensitive only to certain wavelength
bands, while other versions use solid state sensors that measure radiation power over all
wavelengths [1]. For either method, the measured quantity is entered into a correlation function
that produces an estimate of surface temperature.
Several errors are inherent in ordinary infrared thermography. First, unwanted heat
sources may reflect radiation off of the surface under observation and affect the measurements.
Second, the medium between the camera and the observed surface both emits and absorbs
radiation, which also influences measurements. Third, the emissivity of the surface under
observation must be known, for it is a parameter in the calibration functions by which the IR
signature is related to the surface temperature. These errors are generally mitigated through
correction factors; most infrared cameras allow users to supply parameters such as ambient
temperature, surface emissivity, and distance from camera to observed surface and these
parameters are used to generate correction factors. The accuracy and sensitivity of measurements
then depend on the cameras radiation receptors, the users inputs, and the mathematical nature
of the correlation functions and correction factors.
The paper under consideration seeks to take this mature technology of infrared
thermography and apply it to a new situation with a flame in front of an observed surface. This
complicates the analysis because the infrared camera is unable to distinguish between the
radiation from the surface and the radiation from the flame. Furthermore, the radiance,
transmittance, and thickness of the flame are neither known nor easily predicted, and the
emissivity of the surface may change in time as soot accumulates. The crux of this new method
is finding ways to solve these problems.
To determine the flames radiance, the authors suggest that fire resistance tests be
restricted so that the flame is steady and is started or stopped very quickly relative to the thermal
time constant of the plate being observed. The radiation field can then be measured immediately
before and after flame startup or extinguishment, and the difference between these two radiation
fields can be attributed to the flame alone. By subtracting this flame contribution from the
remainder of the IR fields, the IR signature of the plate alone can be estimated. Note again that
this flame contribution can be estimated by looking at flame startup or extinguishment. Ideally,
both measures should be the same since the flame is assumed to be steady. However, the two
methods were found to produce significantly different results. The reason for the difference is
not explained in the paper, nor is it obvious from an analysis of the method, and hence is a
subject that requires additional research.
The authors were able to experimentally verify that the flame was approximately constant
and that the flame could be turned on and off very quickly relative to the plates thermal time
constant, as shown in Figure 2. The flames transmittance was also measured and found to be
very close to 1.0 (the measurement was 0.96). However, this report does not address the fact that
emissivity may change with time; it is assumed in this paper to be constant.
It is of interest to compare the measurement range and accuracy of infrared thermography
with the existing thermocouple technique. Both thermocouples and infrared cameras are capable
of measuring temperatures in excess of 1500C, though special thermocouple materials are
needed and special range-extending options are needed for cameras.
Thermocouples can generally provide very accurate temperature measurements.
However, the authors of this paper point out that thermocouples often have to be mounted from
the back of the plate, meaning that the flame-side temperature of the plate must be inferred. This
reduces the accuracy of the measurements. However, the thermographic technique is also subject
to inherent errors. Several major approximations were made to find the radiosity of the flame,
and in this paper these approximations were not justified by quantitative comparisons of size of
terms. Consequently, the effects of these approximations on the accuracy of temperature
measurement are of unknown size. The authors attempted to quantify the error by numerically
calculating the plate radiance and applying the measurement technique to the result. The
temperature error they obtained is plotted in Figure 3. However, that error analysis failed to
account for other errors including error of the solid-state radiation measurement devices,
uncertainty in the calibration function, and uncertainty in emissivity of the plate.
Overall, the new method trades versatility of allowable flame conditions for spatial
resolution and non-intrusiveness. However, the restrictions that the flame be steady and have
short start-up transients are not overly prohibitive; such tests are already frequently used to
evaluate the performance of flame resistant materials. Though the basic principles of this method
are not complicated, this report is worthwhile because it experimentally demonstrates that
infrared thermography can obtain good surface temperature measurements even in the presence
of a flame.
Figure 1: Left: The experimental setup of a typical fire test. Right: Same test viewed using false-color infrared imaging
Figure 2: Left: Radiance measured as a function of time at four points during a fire resistance test. Right: Location of
points on an IR image prior to the fire resistance test. Note that point 1 represents the contribution of the flame alone.
Figure 3: Comparison of errors using two estimates of flame radiance. Tini uses a flame radiance estimate based on the
flame startup, and Tfin uses a flame radiance estimate based on the flame shutoff.
Reference:
[1]
FLIR Systems, T-Series Users Manual. Rev. A460, July 1, 2010.
Ae104a Homework #4
A 3D vision system for the measurement of the rate of spread and the height of fire
fronts by L Rossi, T Moliner, M Akhloufi, Y Tison and A Pieri:
Meas. Sci. Technol. 21 105501 (2010)
A summary by Esteban Hufstedler, Antoine Mathurin,and Yuan Xuan
Many countries are concerned every year by forest fires that represent a substantial
economic loss and a serious danger for population. The technique presented in this article was
developed to improve fire propagation understanding and to provide firemen on the ground
with an efficient tool that can predict fire front propagation. In order to foresee the fire
propagation, the relevant fire front parameters are its position, height and rate of spread
(ROS). This paper presents a new technique using a stereo vision camera to obtain a 3D
model of the fire front from which the authors have extracted these crucial parameters. The
main difference with current techniques is that it involves new image processing algorithms
that permits deep 3D reconstruction even for the most complex fire fronts. Among the exiting
tools to study fire front propagation there are two main types of tool: techniques using
telemetry sensors and GPS or methods involving standard and infrared cameras. Because of
the important time scale and low resolution, the first ones are still difficult to carry out
nowadays and these techniques were developed specifically for wild land fires and are not
adapted for laboratory experiments. The second ones cannot measure ROS and height for a
significant number of points, are not precise enough to study complex fire fronts and are not
suitable for large and wild land fires.
The device of the experiment is a pre-calibrated multiple baseline stereo camera from
Point Grey. The multiple baseline choice of 12 cm or 24 cm baseline for stereo processing
makes this new technique suitable for fires of different scales. The fire region is reconstructed
by doing data processing to the left and right image obtained by the stereo cameras. To extract
the 3D coordinates of salient fire points using stereo image processing they adapted the
following steps:
(1) Segmentation of the regions
They conducted a color space processing in order to select the best color space and
image area of the fire. The method used for indoor scenarios is the RGB space, and the
YUV color space is employed for outdoor fires.
(2) Detection of the feature points
A modified Harris corner detection algorithm is employed to extract the feature points
along the fire contour.
(3) Automatic computation of the disparity over time.
The disparity and the depth of the detected object in an image are related, and the depth
and disparity of a moving object change. The disparity is estimated by
, as
shown in Figure 1. And we use the grey level of the points to define a normalized crosscorrelation between the point of the images and a selected point.
As the 3D fire points are irregularly distributed in space, it is necessary to differentiate
between the points of the base and the points of the upper part of the fire front in order to
estimate different characteristics of the fire region such as the ROS and height. The quarters
split strategy is used, and the result is shown in figure 2. To construct a continuous base plane
contour, they used the spline method with polynomials of order 3. The ROS of the fire front
can be easily estimated by doing the previous process at different time steps and measuring
the distance between corresponding points of successive back front lines. However in real
cases, some local variations can occur due to small variations in load. The ROS during the
whole propagation can be thus estimated by computing the mean value and standard deviation
of ROS, and averaging those values that are in the interval
.The height of
the fire front is simply estimated to be the distance between the base plane and each point of
the upper part.
The efficacy of this technique was proved with a laboratory experiment. The
experiment was carried out on a 3m x 3m combustion table to which was added a thermal flux
sensor to measure the maximum Z coordinate of the fire whose derivative is the ROS. Several
images were captured with a constant time step of 4s: the mean ROS value generated by the
article method was 0.0035 ms-1. The evolution of the distance between the fire and the flux
sensor is plotted on figure 3. As we can see the evolution is linear which implies that the ROS
is constant over time. The straight line slope was estimated at 0.0037 ms-1 which means the
method is relatively precise with an error of 5.5 %. The mean height of the fire front (whose
3D representation can be seen of figure 4), was measured at 0.13m while it was estimated to
be 0.126m through another method. Therefore given the complexity of the measurement,
notably due to the environment unsteadiness, the measurement results provided by this
technique are quite satisfying.
This technique is valid for outdoor fires that are up to 10m high, 20m wide, and 15m
away from the cameras, as well as for indoor experimental fires that are 30cm high, 3m wide,
and 50cm away. The vailidity of this device was proven as described above. This approach is
suitable for steady fires, where the ROS and height can be determined from a time-average of
the estimated values.
A linear systems approach does not apply to this system, as the errors that come into
the system are primarily due to computer programming and image analysis. The few physical
errors are misalignment of the cameras (negligible according to the manufacturer,) objects
such as ember occluding the fire, resolution of the cameras (negligible at the scales and
resolution they consider,) and parallax error due to the spacing of the cameras. During image
processing, errors may enter initially through the methods of determining the fire area, and
determining and matching feature points. These may result in erroneously placed 3D points,
leading to a flawed estimation of the base plane and thus a flawed estimation of the height and
ROS of the fire.
This experiment can be connected to Ae104 primarily through its use of statistical
averaging. In determining the height and ROS of the fire, they sample each value multiple
times over a run, then fit a curve to those sample points. Since it's an unsteady system, this fit
curve should provide a reasonable time-average value. The image processing also assumes
that the fire is a continuous surface to match points, so it eliminates points with computed
disparities that are too far from average, based on a computed standard deviation.
Figure 1. Estimation of disparity and corresponding point-correlation window in the left and
right images.
Figure 2. 3D fire points.
Figure 3. Time evolution of the distance between the fire front and the flux sensor.
Figure 4. Height estimation of a linear fire front.