You are on page 1of 4

Proceedings of the 42nd European Microwave Conference

Wireless Temperature Sensor for Condition


Monitoring of Mechanical Seals
Lokesh A. Gupta and Dimitrios Peroulis
Electrical and Computer Engineering, Birck Nanotechnology Center,Purdue University, West Lafayette, Indiana, USA
Email: lgupta@purdue.edu,dperouli@purdue.edu

Abstract—This paper presents the first RF temperature sensor implementing efficient cooling techniques to maintain tempera-
for condition monitoring of mechanical face seals. The sensor tures below satisfactory limits. Various methods such as acous-
comprises a resonant inductor-capacitor circuit that is inductively tic measurements of lubrication film thickness using ultrasonic
coupled to a separate interrogating coil. The resonant circuit
is mounted on the stationary face of the mechanical seal. A sensors for seal condition monitoring [2], or wired thermocouple
temperature rise is sensed as a resonant frequency increase, sensors to measure temperatures close to the seal interface have
caused by a decrease in the sensor capacitance value. The been implemented to give an indication of seal performance.
resonant sensor is implemented using commercially-available Acoustic methods are sub-optimal due to external noise in in-
components and has a footprint of less than 5.12 mm2 . Despite dustrial environments. Wired sensors are difficult to use in small
the very close proximity to the steel shaft (0.79 mm) and the
stainless steel flange (2 mm), successful interrogation is achieved spaces (few millimeters depending on pump size) that exist in
and the sensor’s frequency is increased linearly from 41.1 MHz mechanical seal enclosures. In addition, mechanical seals are
to 41.6 MHz for a temperature increase from 25◦ C to 80◦ C. replaced after their useful life. As a result, in such cases, wired
The measurements are verified by simulating the system using sensors attached to the stationary face of a seal are difficult to
a combination of HFSS-ADS EM-circuit co-simulation model. replace without disconnecting the signal processing circuitry.
The uncertainty in the measurements with respect to calibration
equation was less than ±5%. This research provides a solution to the above application
by implementing the first RF sensor using commercially avail-
I. I NTRODUCTION able passive, surface-mount components [3]. Passive compo-
nents such as inductors and capacitors are typically more robust
Mechanical seals are critical subsystems of pumps. The pur-
for sensing applications in high-temperature industrial environ-
pose of a seal is to prevent or minimize leakage from a pump
ments, compared to active components such as transistors. Fig. 2
where the shaft exits the pump housing [1]. A typical mechanical
shows a simplified equivalent diagram of the complete sensing
seal consists of several parts, but for this discussion the two
topology. The temperature sensor is based on the principle of
primary ones are the stationary and rotating faces as shown in
Fig. 1. The contacting surfaces are highly polished to a surface
roughness of 2-6 µin Ra with flatness of 10-20 µin taper across
Seal stationary Flange
face. These faces slide relative to each other on a thin fluid film.
face
Fig. 1 shows a typical mechanical seal and the complete assem-
Seal rotating
bly. Mechanical seals are typically used in harsh environments.
face
Due to friction between the rotating and the stationary faces, heat
is generated that leads to deterioration of the contact surfaces,
which may eventually lead to seal failure. Typical failure condi- Shaft
tions due to high temperatures are listed below [1];
1) The seal tapering in radial direction due to heat generation
from friction can lead to the whole system instability, lead-
ing to complete failure of machine.
2) Due to high temperature, the seal can become wavy in
structure and may lead to increased leakage. Interface Location
3) For fluids like oil, coke deposits may form that can seri- of heat source
ously impair the seal performance.
4) Constant high temperatures may also lead to formation Fig. 1. Pump sub-system with mechanical seals, shaft and flange [4]
of crystals if system fluids contain dissolved salts. These
crystals can bind to the contacting surfaces and lead to electromagnetic induction from one inductor to another. The
excessive wear in the seals. primary coil represents the interrogating inductor while the sec-
These conditions make it is essential to monitor the temperature ondary coil represents the sensor inductor. The shorted tertiary
of the seal interface. Measured data can be analyzed for esti- coil represents the eddy current losses in metallic housing/flange
mating seal lifetime and identifying failures. It can also help in enclosing the mechanical seal. A temperature sensitive capacitor

978-2-87487-027-9 © 2012 EuMA 424 29 Oct -1 Nov 2012, Amsterdam, The Netherlands
Authorized licensed use limited to: Ruhr Universitat Bochum. Downloaded on March 25,2022 at 10:16:41 UTC from IEEE Xplore. Restrictions apply.
0.79 Seal interface/heat
is connected to the secondary coil. The sensor inductor and the
source

Seal stationary
part
Sensor Seal lock key in a
inductor & seal slot
capacitor Shaft
Interrogator inductor
2 on PCB

4 Flange

Fig. 2. This plot shows the sensor equivalent circuit diagram. C1, C2 along with Hole for coax. cable
50 Ω transmission lines (thick black line) form the matching network. The third connection to PCB
All dimensions in mm
coil which is shorted, represents eddy current losses in the metallic housing
Fig. 3. Simplified cross section diagram showing location of sensor and inter-
capacitor form a resonant circuit. When the interrogator inductor rogator inductors, sensor capacitor with respect to stationary part of the seal,
is excited by a frequency sweep, resonance can be detected by flange and metallic shaft
measuring S11 at the source terminals. This resonant frequency
can also be expected as
B. Physical mounting
1
fr = √ (1) The inductor and the sensor capacitor were soldered together
2π Lsen Csen as shown in Fig. 4a. During normal pump operation, a lock-
As the temperature changes, the sensor capacitance changes pro- ing key shown in Fig. 3 prevents rotation of the seal station-
portionally resulting in a shift in resonant frequency measured ary face by latching in to a 3-mm deep, 5-mm wide slot. The
at the source terminals. A similar sensing topology has been sensor inductor and the capacitor were mounted in the slot
presented for bearing cage temperature measurement by [5], [6] without any obstruction to the locking metal post using high-
& [7]. temperature, electrically non-conductive and thermally conduc-
This paper is organized as follows; Section II demonstrates tive epoxy [10]. The seal was made up of high thermally and
the sensor implementation in a mechanical seal with physical electrically conductive material. A thin layer of kapton tape
assembly, Section III discusses the HFSS-ADS EM-circuit co- was used as shown in Fig. 4a to isolate the metallic contacts of
simulation model implementation along with the measurement the sensor inductor and capacitor from the seal. The location
test setup. Lastly, Section IV presents simulation and measure- of sensor inductor and capacitor was 10 mm away from the
ment results. actual heat source at seal interface. This introduced a gradient
II. S ENSOR IMPLEMENTATION of 4◦ C between the temperatures at the interface and location
where temperature was measured. This gradient remains ap-
A. Selection of inductors and capacitors proximately constant due to high thermal conduction of the seal.
In this case, the physical space available for the sensor imple- Hence, any lag introduced by this arrangement can be nullified
mentation was restricted by the mechanical seal housing/flange by implementing correction factors in data acquisition system.
and the shaft. The critical dimensions are shown in Fig. 3. The
gap between the inner-circumference of stationary seal face and B
A
shaft was 0.79 mm, while the gap between the metal housing in- Sensor
Sensor inductor (100nH) inductor & Stationary
ner circumference and shaft was 2 mm. For this application, the & capacitor (150pF) capacitor seal face
interrogator and the sensor inductor should have low series re- (Silicon
sistance to maintain sufficient quality factor, high self-resonant Epoxy Stationary
carbide)
frequency and suitable dimensions to fit in the small gaps shown seal
in Fig. 3. Based on these requirements, a 100 nH 1008HT (2.5 face
Interrogating PCB
mm × 2.1 mm × 1.1 mm) low profile inductor from Coilcraft, inductor
Kapton Flange
Inc. was used for this sensor implementation [8]. (100nH)
tape
Ideal sensor capacitor for this application should have linear Sensor mounted in stationary face Mechanical seal in flange
monotonic response with temperature and dimensions similar
to the 100 nH inductor. [5] has presented a few capacitors for Fig. 4. Sensor inductor and capacitor mounting
wireless temperature measurement of bearing cage. A 150 pF,
1206 package, Murata U2J-series capacitor with a temperature The interrogator inductor was mounted on a small 0.3-mm
coefficient of −750±120 ppm/◦ C was selected as temperature thick printed circuit board as shown in Fig. 4b. The sensor and
sensor capacitor[9]. the interrogator inductor were aligned for maximum signal cou-

425
Authorized licensed use limited to: Ruhr Universitat Bochum. Downloaded on March 25,2022 at 10:16:41 UTC from IEEE Xplore. Restrictions apply.
pling. The gap between the two inductors was approximately 1 issue was mitigated by using combination of HFSS-ADS EM-
mm. RG178B/U coaxial cable interfaced the interrogator induc- circuit co-simulation [13]. A 3-dimensional model of inductors
tor to the measurement setup. The interrogator inductor value with a seal, flange and shaft was implemented in HFSS as shown
was 100 nH, coaxial cable inductance was 297 nH/m and ca- in Fig. 5. Circular shapes of these components were substituted
pacitance was 96 pF/m [11]. Hence, the length of coaxial cable by rectangular blocks as shown in the figure for a simplified
between measurement circuit and interrogator inductor was kept CAD diagram. The electrical properties of the silicon carbide
at 25 cm to minimize any uncertainties introduced by the cable. seal such as conductivity and permittivity were unknown hence
graphite was used as seal material for HFSS simulations. The
C. Matching network dimensions, number of windings and wire diameter of the induc-
The amplitude of the S11 signal at the resonant frequency was tors were physically measured using a microscope for creating
extremely small being in the order of -1 dB. To improve the the 3-dimensional inductor CAD model. The complete model
measured signal, a matching network was designed to match the was simulated using S-parameter analysis from 20 MHz to 60
50 Ω source and coaxial cable to the low impedance interrogator MHz. S-parameter analysis results were then used to generate
inductor. The matching network equivalent circuit is shown in an equivalent coupled inductor model using the HFSS built-
Fig. 2. The capacitors C1, C2 and the three 50 Ω transmission in SPICE model export function. The three coupled inductor
lines form the matching network. The values of the capacitors model shown in Fig. 2 was replaced by this equivalent model
were calculated by equations 2 and 3 [5] and the circuit was simulated (S-parameter analysis) for various
1 sensor capacitance values based on −750 ppm/◦ C drift. The
C1 = (2) matching network capacitor values were fine tuned in simula-
ω RRin (R2 −RRin + L2 ω 2 )
tions to achieve maximum signal at the output close to resonant
Rin −R frequency of 41.09 MHz calculated for 100 nH inductor and 150
C2 = R
(3) pF sensor capacitor.
LRin ω 2 + C1
where Rin is the DC resistance of the interrogator inductor, L is
the inductance(interrogator inductance plus cable inductance),
R is the source impedance (50 Ω) and ω is the resonant fre-
quency in rads/sec calculated using equation 1. For fine tuning
the impedance matching for maximum signal, 4 pF - 40 pF vari-
able capacitors were soldered on a PCB with a 50 Ω mircostrip
line.

III. S IMULATION AND TEST SETUP

Shaft
Seal

Sensor
inductor Fig. 6. The measurement test setup. The matching network PCB was kept
outside the oven to prevent errors due to drift in capacitance of matching network
capacitors.

Fig. 6 shows the measurement test setup. The entire mechan-


ical assembly was placed in a oven and temperature was in-
Interrogator
inductor
Flange creased by 2◦ C/min. A thermocouple was placed close to the
sensor inductor in the gap between the stationary seal face and
shaft for reference temperature measurement. The measurement
circuit was connected to Agilent network analyzer for single port
Fig. 5. This image shows the simplified 3D model of the complete seal mechan- S-parameter measurements [14]. The S11 readings were cap-
ical assembly. For CAD and simulation simplification, rectangular blocks were
implemented instead of cylindrical shaped seal, flange and shaft.
tured after every 5◦ C rise up to 80◦ C and resonant frequencies
were extracted for corresponding temperatures.
The circuit shown in Fig. 2 was simulated using Agilent ADS
IV. R ESULTS AND DISCUSSION
[12]. The coupling factors between the sensor inductor, the in-
terrogator inductor and the flange cannot be measured directly The resonant frequency is inversely proportional to square
because of space constraints and small inductor values. This root of the capacitance, as given by equation 1. Fig. 7(a) shows

426
Authorized licensed use limited to: Ruhr Universitat Bochum. Downloaded on March 25,2022 at 10:16:41 UTC from IEEE Xplore. Restrictions apply.
this response and resonant frequency increased with tempera- performed for −750 ppm/◦ C and if the sensor capacitor during
ture. The resonant frequency at 25◦ C from simulation and mea- measurements operated with drift value of −630 ppm/◦ C due to
surement was 41.3 MHz and 41.1 MHz respectively, indicating piece-to-piece variations, then measurements trend would have a
good accuracy of the simulation model. The simulation results lower slope. Fig. 7(b) shows uncertainty in the measured temper-
showed excellent linear response with respect to temperature. ature results with respect to calibration equation. All measured
The measurement results were linear up to 45◦ C. Beyond this data points were confined within uncertainty of ±4.5%. Table I
summarizes the matching network capacitor values from simu-
lation and measurements. A good agreement was observed for
capacitor values obtained from measurements and simulation.
TABLE I
M ATCHING NETWORK CAPACITOR VALUES FROM SIMULATION AND
MEASUREMENT

Capacitor C1 C2
Simulation 38.6 pF 34.1 pF
Measurement 32.6 pF 38.1 pF

V. C ONCLUSION
In this research, a wireless temperature sensor was imple-
mented and tested for temperature monitoring of mechanical
face seals. The topology was implemented using commercially
available inductors and capacitors. Furthermore, the system was
validated using a HFSS-ADS EM-circuit co-simulation model.
(a)
Excellent congruity was observed between simulated and mea-
sured resonant frequencies at room temperature. Measurement
trend followed a power equation with respect to temperature
from 25-80◦ C with temperature uncertainty of less that ±5%.
R EFERENCES
[1] A. Lebeck, Principles and design of mechanical face seals, ser. Wiley-
Interscience publication. Wiley, 1991.
[2] Y. E. Fan, F. Gu, and A. Ball, “A review of the condition monitoring of
mechanical seals,” ASME Conference Proceedings, vol. 2004, no. 41758,
pp. 179–184, 2004.
[3] L. A. Gupta, “Rectennas and temperature sensors for wireless sensing ap-
plications,” Master’s Thesis, Purdue University, West Lafayette, Indiana,
Aug 2010.
[4] Silberwolf, “Rotating mechanical seal-3D,” http://commons.wikimedia.
org/wiki/File:Rotating mechanical seal-3D 135.png, 2012, ”[Online;
accessed 17-February-2012]”.
[5] A. S. Kovacs, “Early-warning wireless telemeter for harsh-environment
bearing,” Master’s Thesis, Purdue University, West Lafayette, Indiana, Dec
2008.
(b) [6] A. Kovacs, D. Peroulis, and F. Sadeghi, “Early-warning wireless telemeter
for harsh-environment bearings,” in Sensors, 2007 IEEE, Oct. 2007, pp.
Fig. 7. (a) This plot shows resonant frequency vs. temperature for measurement 946 –949.
and simulations. Simulation results followed a linear trend while measurement [7] S. Scott, A. Kovacs, L. Gupta, J. Katz, F. Sadeghi, and D. Peroulis, “Wire-
results followed a non-linear trend given by power equation.(b) This plot shows less temperature microsensors integrated on bearings for health monitoring
temperature uncertainty in the measurements with respect to power equation applications,” in Micro Electro Mechanical Systems (MEMS), 2011 IEEE
shown in Fig. 7(a). 24th International Conference on, Jan. 2011, pp. 660 –663.
[8] “Coilcraft, inc.” http://www.coilcraft.com/, 2012.
temperature, the trend became non-linear as indicated by the [9] “Murata manufacturing co.,ltd,” http://www.murata.com, 2012.
[10] “Ultra temperature epoxies,” http://www.cotronics.com, 2012.
power equation from the curve fit shown in the plot. This non- [11] “Belden, inc.” http://www.belden.com/, 2012.
linearity is thought to have been caused by very small expansion [12] “Advanced design system,” http://www.agilent.com/find/ads, 2012.
of the epoxy used to attach the sensor capacitor to the seal. The [13] “Ansoft hfss,” http://www.ansoft.com/products/hf/hfss/, 2012.
[14] “Agilent technologies,” http://www.home.agilent.com, 2012.
expansion of the epoxy modified the capacitance of the sensor
capacitor at higher temperatures resulting in reduction of slope
of the measurement curve. The epoxy was not modeled in the
simulation model hence simulation results showed linear trends.
Another reason for a lower slope of measurement trend as com-
pared to simulation, was the temperature drift specification of the
sensor capacitor at −750 ±120 ppm/◦ C. The simulations were

427
Authorized licensed use limited to: Ruhr Universitat Bochum. Downloaded on March 25,2022 at 10:16:41 UTC from IEEE Xplore. Restrictions apply.

You might also like