Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Sensors and Actuators B 175 (2012) 246–254

Contents lists available at SciVerse ScienceDirect

Sensors and Actuators B: Chemical


journal homepage: www.elsevier.com/locate/snb

Detection limit improvement for NDIR ethylene gas detectors using


passive approaches
Adam Sklorz ∗ , Steffen Janßen, Walter Lang
Institute for Microsensors, -Actuators and -Systems (IMSAS), University of Bremen, Otto-Hahn-Allee NW1, 28359 Bremen, Germany

a r t i c l e i n f o a b s t r a c t

Article history: The influence of passive construction improvements of a NDIR ethylene gas detector on the signal inten-
Received 30 September 2011 sity and the detection limits were examined. Rectangularly and conically shaped measurement chambers
Received in revised form 11 July 2012 with and without low-cost Fresnel lenses were used. Also the influence of the choice of the applied broad-
Accepted 26 September 2012
band IR-sources was investigated and an equation for detection limit estimation is described. In addition
Available online 5 October 2012
three digital signal processing methods were compared with results provided by a lock-in amplifier.
It was shown that the best signal intensity can be achieved using a combination of a conically shaped
Keywords:
chamber and Fresnel lenses. The IR-radiation measurement efficiency was improved from 0.8% to 4.2% in
NDIR-spectroscopy
Non dispersive infrared
the best case. The achieved detection limits were 34 ppmv without and 25 ppmv with the lock in ampli-
NDIR-detector fier, the noise equivalent values 6 ppmv and 4 ppmv respectively. These limits match roughly with the
Ethylene sensor calculation results and are comparable with results obtained with white cell systems.
Ethylene detector © 2012 Elsevier B.V. All rights reserved.
Detection limit
Broadband IR-sources
Fresnel lenses
Measurement chamber
Infrared gas detection

1. Introduction and/or portable gas detectors for gases which absorb infrared (IR)
radiation. For example, detectors for CO2 are widely using this
The ability to measure ethylene gas (C2 H4 ) in ambience selec- method. The main advantage of NDIR-measurement systems is the
tively with a ppmv and ppbv resolution is of interest for the fruit good selectivity which is achieved by measuring the IR-absorption
logistic sector. Ethylene gas is emitted by fruits depending on within a gas specific IR-band. The assembly of NDIR-detectors
their ripeness state on one hand and it accelerates the ripeness holds the system costs low and provides robustness. In contrast to
of neighbour fruits on the other hand [1]. Knowing the ethylene these benefits three disadvantages limit the applicability of NDIR-
concentration and other parameters like temperature and humid- systems. For one thing, the selectivity of NDIR-systems depends on
ity in the transportation ambience provides the estimation of the the wavelength range at which the absorption is measured. This
current and future fruit ripeness and quality by so called shelf life is given by the wavelength dependant emission intensity of black-
models and autonomous control hardware. The application of gas body radiators. The second point is, that the sensitivity of NDIR-
measurement systems which provide the mentioned performance systems is strongly influenced by the optical pathway between the
would help to establish new logistic approaches like dynamic FEFO IR-source and the measurement detector. To enhance the sensitiv-
(First Expired–First Out) to reduce transport losses, costs for waste ity the optical pathway of the system has to be increased at the cost
management and limit unnecessary transports [2,3]. Since proper of system size and/or application of additionally optical devices.
solutions are not available for this application, different efforts are After all the sensitivity of NDIR-systems is limited by the applied
made to develop such systems [4–6]. The main challenge is thereby thermopile or pyroelectric detectors, especially by the physical lim-
to achieve good selectivity and sensitivity to C2 H4 . For this task itations due to noise and sensitivity to IR-radiation. These three
different measurement approaches are applicable; one is the non- disadvantages become an issue, if NDIR-systems for ethylene gas
dispersive infrared (NDIR) spectroscopy. The application of NDIR- detection are in focus. With ethylene as target gas, the absorption
spectroscopy is a common method for realisation of miniaturised band used for detection is at 10.5 ␮m. The radiation intensity in
this wavelength region is smaller by the factor of 4 compared to the
radiation in the 4 ␮m which is used for CO2 detection beside others
∗ Corresponding author. [7]. This necessitates different constructive approaches to utilise
E-mail address: asklorz@imsas.uni-bremen.de (A. Sklorz). the available performance of the system components in the most

0925-4005/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.snb.2012.09.085
A. Sklorz et al. / Sensors and Actuators B 175 (2012) 246–254 247

These methods do not affect the used active devices in the system,
Nomenclature so they can be called passive. Beside these passive methods the
influence of the IR-source choice on the detector signal is examined.
F bandwidth of the detector’s optical broadband filter
 gas specific extinction coefficient 1.1. State of the art
CW centre wavelength of the detector’s optical broad-
band filter As far as known two NDIR ethylene gas detectors were described
S efficiency related to Prad,C2 H4 and PTP in literature. One was a system based on a white cell with three con-
fe bandwidth of the used signal processing electronics cave gold mirrors for increasing of the optical pathway. The optical
c0 speed of light pathway of the system was 1.6 m provided by 20 reflections. The
dref reflector diameter of the IR-source entire measurement chamber had a size of 11 × 5 ×6 cm3 . A ther-
fQ,CLK IR-source clock signal frequency mopile array with integrated micro machined Fresnel lenses and
h Plank’s constant optical filters for different target gases was applied in the system.
href reflector height of the IR-source A preamplifier and a microcontroller based lock-in amplifier were
k Boltzmann constant used as measurement electronics. A new developed broadband IR-
l optical path length source with a 3D-emitter and a temperature of 800◦ C was applied in
tHI,0 , tHI,1 relevant times for digitally calculated high-state the system [9,20]. The IR-source was powered with 2.5 W and oper-
measurement signal ated in continuous mode because of its large time conctant. Thus
vnoise,RMS RMS value of noise voltage it was chopped mechanically at 8 Hz using a chopper blade. The
vnoise ] noise achieved ethylene detection limit is about 20 ppmv [8]. The second
vp−p,noise peak-to-peak noise voltage system, which was used for the measurements in this paper, con-
SENC2 H4 ethylene sensitivity sists of a conically shaped aluminium measurement chamber with
SNR100ppmv signal-to-noise ratio related to an ethylene con- an overall size of 22 × 2.5 × 4 cm3 . Instead of mirrors a conically
centration of 100 ppmv shaped measurement chamber was used to increase the efficiency
ADET active area of the thermopile detector of IR-radiation. As IR-source the standard device HSL-EMIRS with a
AIR emission area of the IR-source blackbody temperature of 460◦ C was applied. Using TPS 434 ther-
Cmin concentration detection limit mopile detectors the system sensitivity was about 40 ppmv [10].
I0 reference radiation intensity
IDET radiation intensity at the thermopile detector
2. Material and methods
IIR,On IR-source active high supply current
IMIN radiation intensity detection limit
2.1. Estimation of detection limit
L,TK spectral emission density of a blackbody radiator
NEC noise equivalent concentration
For the estimation of the detection limit Cmin of a NDIR-system
PDET radiation power measured by the thermopile detec-
the first aspect which has to be taken into account is the intensity
tor
of IR-radiation provided by the used IR-source within the wave-
PIR mean value of the IR-source input power
length region of interest. Beginning with the general equation for
Prad,C2 H4 calculated IR-source emission power within the
the spectral density
measurement wavelength region
Prad power emitted by the IR-source 2c02 h
L (, TK ) =   (1)
RIR heater resistor of the IR-source 5 exp(hc0 /kTK ) − 1
SDET sensitivity of the thermopile detector
TK emission temperature of the IR-source the emitted intensity is calculated using
TTP thermopile temperature  
CW + 2 F

VDD supply voltage I0 = L (, T )d (2)



VHI digitally calculated high-state measurement signal CW − 2 F

VIR,On IR-source active high supply voltage 


within the wavelength transmission band CW ± 2 F of the mea-
VLI lock-in measurement signal
surement detector. The output voltage of the thermopile detector
Vp−p digitally calculated peak-to-peak measurement sig-
is calculated using
nal
VQ,CLK IR-source clock signal VTP = PDET SDET = IDET SDET ADET
VRMS digitally calculated RMS measurement signal (3)
= s Prad SDET ADET
VTP,BSL thermopile signal at an ethylene concentration of
0 ppmv with the sensitivity SDET , the active area ADet of the detector and
VTP thermopile output voltage the efficiency s which describes all losses in the systems related
to the input power for the IR-source and the power measured by
the detector. The bottleneck concerning the detectability of IR-light
absorbed by ethylene is the noise vnoise of the used thermopile
effective way. This paper handles different approaches for the detectors. The RMS-value of the noise voltage can be calculated
improvement of the sensitivity of NDIR-measurement systems for by
ethylene: 
vnoise,RMS = vnoise fe (4)
• Usage of low-cost Fresnel lenses to focus the IR-radiation to the 
where fe is the bandwidth of the used signal processing elec-
active detector area, tronics. The related peak-to-peak value of the noise voltage can be
• application of a proper shaped measurement chamber to limit
estimated using
reflection loses and
• a signal processing technique to reduce measurement noise. vp−p,noise ≈ 7vnoise,RMS (5)
248 A. Sklorz et al. / Sensors and Actuators B 175 (2012) 246–254

Fig. 1. 3D exploded assembly drawing of the used NDIR-detector.

and Eq. (4) [17]. The detection limit Cmin = C1 − C2 can be described
by the lowest measurable thermopile signal change defined as
VTP,MIN = VTP (C1 ) − VTP (C2 ) (6)
With Eq. (3) the lowest detectable radiation intensity
VTP,MIN (I) (II) (III) (IV)
IMIN = (7)
s SDET ADET
can be calculated by setting VTP,MIN = 2vSS,noise . From the angle
of the Lambert–Beer’s law (see Eq. (A.1)) the lowest detectable
radiation intensity can be scribed by
IMIN = I0 − I = I0 (1 − exp (−lCMIN )) (8)
Transforming Eq. (8) leads finally to the related detection limit Fig. 2. Detector setups used in the measurements. Three different IR-sources (I) and

1
 I0
 two different chamber shapes (II). All assemblies were examined with and without
lenses (III) placed in front of the used thermopile detectors (IV).
Cmin (l) = ln (9)
()l I0 − IMIN
as a function of the optical pathway l and to In case of applied lenses the IR-radiation was focused on the

1 I0 (TK ) active area. In the opposite case the detector front side was flush-
Cmin (l, Tk ) = ln ) (10) mounted with the internal wall of the detector socket. Beside the
()l I0 (TK ) − (VTP,MIN /s SADET
passive system variations itself the three commercially available
with the temperature TK of the blackbody radiator. broadband IR-sources IR-55, HSL-EMIRS and MIRL-17 were used in
the measurements (see Fig. 3). Some important parameters of the
2.2. Detector design

The NDIR-detector used for the experiments described in this


paper consists of four main components: a detector socket for inte-
gration of two thermopile detectors, an IR-source adapter and a
measurement chamber broken up into a top and a bottom cover
(see Fig. 1).
The assembly with the dimensions of 22 × 2.5 × 4 cm3 was
screw-mounted. The distance between the IR-source and the detec-
tors is 20 cm. A Teflon lens holder with glued Fresnel lenses [11]
is applied in front of the detector socket. The IR-source adapter
provides the integration of IR-sources of different sizes into the
system. Two thermopile devices TPS 434 with optical absorption
filters were used as IR-detectors [12]. The transmission region
of the filters was 11 ␮m ± 1.5 ␮m for the measurement detector
and 4 ␮m ± 0.1 ␮m as reference [13]. The detectors had a sensi-
tivity of 35 V W−1 and a noise of 21 nV Hz−0.5 and were integrated
into the detector socket. In the socket the detectors were mov-
able in one direction to focus the IR-radiation onto the active area
ADET = 0.510−6 m2 of the detectors. All in all, four different system
setups were examined which are outlined in Fig. 2.
Two different chamber shapes were used: a rectangular and a
conical chamber, both with and without Fresnel lenses. The internal Fig. 3. IR-sources IR-55 (I), HSL-EMIRS (II), MIRL-17 (II), adapters (IV) and driver-
walls of both chambers were polished. electronics (V).
A. Sklorz et al. / Sensors and Actuators B 175 (2012) 246–254 249

Table 1 It consists of the operation amplifier OPA 547 and the resistances
Some specific values of the three applied IR-sources [14–16].
R1 to R3 and is powered by a supply voltage VDD2 of 8 V. At the
IR-55 MIRL 17 HSL EMIRS input of the non-inverting amplifier the rectangular driving signal
RIR () 50 50 45 VQ,CLK is applied with the clock frequency fQ,CLK and a duty cycle
UIR,On (V) 6.4 6.6–7.7 6 of 50%. VQ,CLK is 5 V for the active-high state and 0 V for active-
IIR,On (mA) 135 130–150 90 low state. The resistances R1 and R2 adjust the amplification of the
AIR (mm2 ) 2.25 2.89 3.78 circuit to provide voltages with the needed amplitudes between
IR (ms) 11.5 5 11
6.5 V and 7.5 V at RIR (see Table 1). Resistor R3 is used to limit the
Tk (◦ C) 750 750 450
dref (mm) 11.56 9.13 9 output current of the circuit to protect the IR-source. The E/S signal
href (mm) 16.10 9.75 6.52 is used to activate the output signal. The thermopile temperature
measurement electronics use a thermistor integrated in the TPS 434
device as a temperature sensing element and a LM334-device to
VDD,2 inject a constant current of 17.5 ␮A into the thermistor. All circuits
were realised using four layer PCBs. The internal layers of the boards
E/S R3 were used as pure analogue supply layers to prevent digital ripple
OPA 547
influence on the measurement signals. Additionally coaxial cables
VQ,CLK + were used for all measurements.

-
R1 2.3. Measurement setups and signal processing

RIR All experiments described in this paper were made using the
measurement setup drafted in Fig. 5. The output signals of the ther-
R2 mopile measurement detector VTP and the reference detector were
amplified using the electronics described above. The output signals
were connected to a NI USB-6212 DAQ-device for data acquisition.
A 5 V digital counter output of the DAQ-device was connected to
Fig. 4. IR-source driver circuit schematic with the input driving signal VQ,CLK and
the input VQ,CLK of the driver electronics and to the reference input
the Enable/Select input E/S. RIR symbolises the IR-source.
of a lock-in amplifier (Femto GmbH, model LIA-MV-200-S). The pre-
amplified output signal was connected to the measurement input
sources are listed in Table 1. All three sources were equipped with of the lock-in amplifier.
a reflector for radiation focusing. The output signal of the lock-in amplifier and the output sig-
nals of the measurement and reference electronics were connected
2.2.1. Electronics to the analogous inputs of the DAQ-device. A bottle with ethylene
The electronics developed for the NDIR detector consist of an gas in synthetic air with concentrations of 100 ppmv and 500 ppmv
amplifier circuit for the thermopile detectors and driver electron- respectively was connected to the mass flow controller MFC1. Addi-
ics for the IR-source as well as electronics for measurement of the tionally pure synthetic air was used on the input of the MFC2. The
detector temperature. The amplifier was realised as a state of the art output pressure of both bottles was limited to 2 bar. The pneu-
two stage circuit using the AD8628 and AD627 operational ampli- matic outputs of both MFCs were connected to the gas inlet of the
fiers. The DC-amplification of the preamplifier was set to 90.9, the NDIR-detector. The MFCs (MKS GmbH, model 1259CJ), the DAQ-
entire amplification was 3021 for the measurement channel. The device and the IR-source of the NDIR-detector as well as digital
cut-off frequency of the circuits was set to 10 Hz. The supply voltage data processing and logging was provided by a standard computer
VDD used for the signal measurement circuits was 5 V. The driver and a LabView implementation. In total two different measure-
circuit used to control the IR-source is shown in Fig. 4. ment series were made. The first measurement series was applied

Fig. 5. Draft of the used measurement setup. The ethylene gas and synthetic air are applied to the measurement chamber using the mass flow controllers MFC 1 and MFC 2.
Data acquisition and IR-source control are provides by a DAQ-Device. The measurement control, data logging and processing is provided by a standard computer.
250 A. Sklorz et al. / Sensors and Actuators B 175 (2012) 246–254

Fig. 6. Raw measurement signal VO2 in case of use of the MIRL-17 source and a
chopping frequency of 2 Hz. The Signal VTH is the mean value of the active-high
signal state between the times tHI,0 and tHI,0 which were taken at 99% of the signal
maximum.

to compare the influence of the different chamber shapes, the effect


of Fresnel lenses, the consequence of the IR-source choice and the
chopping frequency on the output measurement signal intensity.
Here the three IR-sources described above were integrated one
by one into each detector setup and were driven with frequen-
cies from 2 Hz up to 50 Hz. Clean synthetic air with a flow rate
of 200 sccm was flowing through the measurement chamber dur-
ing this measurements. The measurement signals were logged at
each frequency for 10 min. In the second measurement series the
best assembly identified in the previous measurements was used
to examine the detection limit to ethylene gas. Therefore the MFCs
were used to mix different ethylene concentrations and to apply
them to the detector chamber. During the measurements the sum
of both flows adjusted by MFC1 and MFC2 was hold constant at
200 sccm. The ethylene concentration was varied between 0 ppmv
and 500 ppmv in 100 ppmv (500 ppmv bottle) steps and between
0 ppmv and 100 ppmv in 20 ppmv (100 ppmv bottle).
In all cases the measurements were started with a 30 min system
warm-up time to ensure temperature equilibrium in the system.
The row measurement signals (see Fig. 6) were processed
to measure the signal change caused by the ethylene absorbed
IR-radiation. Three signals were calculated from the amplified ther-
mopile signals: the RMS-voltage VRMS , the peak-to-peak voltage
Vp−p and finally the high-state voltage VHI . The first two mentioned
signals were calculated using common digital algorithms provided
by LabView. For VHI calculation the raw-signal shown in Fig. 6
was processed using equation (A.2) between the times tHI,0 and
tHI,1 . The measurement signal VLI provided by the lock-in amplifier
was used unchanged. All signals were measured over 16 periods of
VQ,CLK , the mean of the processed values was calculated and taken
as measurement sample. Additionally the thermopile temperature
was investigated during all measurements. The temperature infor-
Fig. 8. Output measurement signal Vp−p as a function of the chomping frequency,
mation were used to compensate the temperature influence using f = 100 Hz. (a) Rectangular chamber without Fresnel lenses, (b) rectangular cham-
temperature calibration curves (see Fig. 7). ber with lenses, (c) conical chamber without lenses, (d) Conical chamber with lenses.

3. Results

3.1. Influence of chamber shape, lenses and IR-source

The results obtained from the first measurement series plotted


in Fig. 8 show the influence application of Fresnel lenses, differ-
ent chamber shapes and different IR-sources on the intensity of
measurement signal Vp−p which is taken here as example. The qual-
itative results concerning the other measurement signals show the
same behaviour.
The subplot (a) in Fig. 8 shows the signal intensity for application
Fig. 7. Comparison of the measurement signal Vp−p with and without temperature of the three IR-sources in a rectangular chamber without Fresnel
compensation. lenses.
A. Sklorz et al. / Sensors and Actuators B 175 (2012) 246–254 251

Table 2
Signal intensities at fQ,CLK = 2 Hz resulting from Fig. 8.

Shape Lenses IR-55 MIRL-17 (Vp−p ) HSL-EMIRS

Rectangular 0.313 0.466 0.603


Conical 0.347 0.509 0.733
Rectangular X 0.725 1.195 0.875
Conical X 2.549 3.712 3.225

The best signal intensity of about 0.6 V for chopping frequen-


cies up to 5 Hz is caused by the source HSL-EMIRS, followed by the
MIRL-17 and the IR55. In contrary Fig. 8(b) demonstrates the signal
intensity for the same chamber with integrated Fresnel lenses. Here
the best results are given by the application of the MIRL-17 source
Fig. 9. Estimated detection Limit CMIN as a function of optical path length l for NDIR-
with a value of about 1.2 V, followed by the HSL-EMIRS and the ethylene measurement systems using TPS 434 thermopile detectors and the MIRL-
IR55. In case of the conically shaped chamber without lenses, the 17 broadband IR-source with two different efficiencies S .
maximum signal intensity of 0.7 V appears when the HSL-EMIRS is
used (see Fig. 8(c)). Finally, the application of the conically shaped
chamber with lenses results in the best signal intensity of about
4 V for the MIRL-17, followed by the HSL-EMIRS and the IR55 (see
Fig. 8(d)). These results and its analysis are concluded in Table 2.
The related peak-to-peak output voltage of the thermopile
detectors VTP are listed in Table 3. Additionally the mean IR-source
input power PIR and the emitted source radiation power Prad,C2 H4
within the optical transmission band of the measurement ther-
mopile are indicated in the table. The given values for Prad,C2 H4
are theoretical values calculated using Eqs. (1), (2) and (3) for a TK
of 750◦ C. The thermopile voltage VTP was calculated by division of
Vp−p by the amplification factor of the used measurement electron-
ics, PTP was calculated using Eq. (3), the efficiency S is defined by Fig. 10. Measured partial sensitivity related to VTP plotted against ethylene concen-
tration for three different IR-sources.
Eq. (A.3). In case of application of the rectangularly shaped cham-
ber without lenses, the thermopile voltage VTP is below 200␮V and
the measured radiation power PTP is below 6 ␮W for all IR-sources. The IR-source temperature TK was set to 750◦ C, the extinction
The change of the chamber shape to the conical one improves this coefficient  was set to 5.672 × 10−5 mol ␮mol−1 m−1 (mean value
values marginally. The application of Fresnel lenses in the rectan- estimated from () for the used measurement bandwidth based
gularly shaped chamber improves the thermopile signal strength on data provided by [18]). The resulting relations between Cmin and
by a factor of about two. The best signal gain is given by applica- the optical path length l for both efficiencies are shown in Fig. 9.
tion of the conical chamber together with the Fresnel lenses. The As expected, the detection limit CMIN decreases with increasing
signal intensity is improved by a factor of eight for the IR-55 and optical pathway. The detection limit is about 1.2 ppmv at 0.2 m,
the MIRL-17 and by a factor of 5 for the HSL-EMIRS. 0.3 ppmv at 1 m and 0.15 ppmv at 2 m in the ideal case S = 100%. In
comparison CMIN changes to 46 ppmv at 0.2 m, 12 ppmv at 1 m and
3.2. Detection limit 6 ppmv at 2 m in the more realistic case of S = 2.4%. This estima-
tion correlates with measurement results achieved from sensitivity
The estimation of the detection limit Cmin described by Eq. (10) measurements.
was made with system parameter given above, an assumed effi- The change of the thermopile output voltage related to
ciency S of 100% and 2.4% (best case, see Table 3). the baseline as a function of the applied ethylene con-
centration CC2 H4 is shown in Fig. 10. The best behaviour
Table 3 provides the MIRL-17 source (SENTP,C2 H4 = 21 nV ppmv−1 ), fol-
Efficiency results for the three IR-sources. lowed by the HSL-EMIRS (SENTP,C2 H4 = 15nV ppmv−1 ) and the
IR-55 MIRL-17 HSL-EMIRS
IR-55 (SENTP,C2 H4 = 13 nV ppmv−1 ). The analysis results for the best
system setup (conical chamber, Fresnel lenses and the MIRL-17
PIR (W) 0.45 0.56 0.23
as IR source) are given in Table 4 for different signal processing
Prad,C2 H4 (mW) 1.5 1.5 0.73
methods.
Shape Lenses VTP (␮V) The SNR100ppmv and NEC values were calculated using Eqs. (A.5)
Rectangular 103.6 155.6 198.6
Conical 123.8 168.5 242.6
and (A.6). The best SNR100ppmv behaviour is provided by the Lock-
Rectangular X 240 395.6 290 In amplifier, followed by the peak-to-peak approach. The RMS and
Conical X 843.8 1228.7 1067.5 HI approaches show poor results. Since the detectable signal was
PTP (␮W)
Rectangular 3 4.4 5.7 Table 4
Conical 3.5 4.8 6.9 Analysis of the results for conically shaped chamber with lenses, MIRL-17 as IR-
Rectangular X 6.9 11.3 8.3 source (STP = 21 V ppmv−1 ).
Conical X 24.1 35.1 30.5
Vp−p VHI VRMS VLI
S
Rectangular 0.2% 0.3% 0.8% SNR100ppmv 305 83,5 145 480
Conical 0.24% 0.32% 0.95% vnoise,RMS (␮V) 372 807 630 330
Rectangular X 0.46% 0.75% 1.13% SENVO2 ,C2 H4 (mV) 65.3
Conical X 1.61% 2.34% 4.16% NEC(ppmv) 5.7 10.9 8.3 4.1
252 A. Sklorz et al. / Sensors and Actuators B 175 (2012) 246–254

4. Discussion

Based on the results presented it can be pointed out, that


the proposed passive approaches for sensitivity improvement of
NDIR-detectors provide a good performance gain (see Table 3).
A rectangular chamber give poor signal intensities, the efficiency
S is in this case below 1% for all IR-sources. This poor value can
be explained by the radiation losses caused by diffuse reflection
of the emitted IR-light in the measurement chamber and by the
small ratio of the thermopile active area to the surrounding area of
the detector housing and the internal wall of the detector socket.
The diffuse reflections in the measurement chamber result in
a reduction of the radiation intensity. Since the measurement
chamber was made of polished aluminium, the maximum possible
reflexion value in the 10 ␮m wavelength region is 98.7 [19] but
it can be assumed that the reflexion is lower because of flatness
imperfection. If the reflection is assumed to 95%, the intensity of
a radiation beam is reduced by more than 50% after 15 reflections,
and thus, the sensitivity is reduced before the beam can reach the
detectors. In contrary to the rectangularly shaped chamber the
conically shaped chamber guides the emitted IR-beams through
the measurement volume to the front surface of the detector
socket where the thermopiles can measure the intensity. This
effect gets clear most notably in case of lens application. While S
is below 1.2% for the rectangular chamber this value rises to 4.2%
for the conically shaped chamber using the HSL-EMIRS source. The
estimated detection limit using Eq. (10) led to 11 ppmv for optical
path lengths of 1 m. This is obviously a roughly estimation, but the
result shown in Fig. 9 match with the sensitivity measurements
(2NEC). Thus, Eq. (10) provides an approximately calculation
of the detection limit of NDIR-systems. The limitations of the
equations are given for longer optical path lengths which results
in a significant increase of reflections in the chamber and thus,
more reflection losses. These losses are not considered in Eq. (10).
Another point is, that the optical path length of the NDIR-detector
used for the measurements is not equal to the distance between
the IR-source and the detectors. Since the IR-beams are reflected on
the side walls during their way to the detectors, it can be assumed
at about 1 m. However, this will be proved in further works. The
detection limits related to ethylene gas achieved match the results
of other works based on white-cell systems [8,9]. The best result of
25 ppmv was achieved using a lock-in amplifier. This was expected
because of the noise reduction advantages a laboratory based lock-
in amplifier provides. In comparison the 34 ppmv detection limit
achieved using a standard peak-to-peak digital signal processing
method and additional temperature compensation is also an
acceptable result. The other two methods give quite poor results.
In general it is noticeable that the ethylene detection limits of the
Fig. 11. Long-term measurement results for 20 ppmv ethylene concentration steps. NDIR-detector described in this paper and the detection limits of
(a)–(c) detector signals, (d) thermopile temperature, (e) concentration profile. the white-cell NDIR-systems [8,9] are in the same double-digit
ppmv range. Because of some unknown characteristics of the
white-cell system a consideration like in Table 3 could not be done
here. Alternatively the Concentration Measurement Efficiency
said to be at least two times the noise voltage (see Eq. (6)), the
CC2 H4 ,MIN
theoretical detection limit CMIN can be written here as 12 ppmv for CME = (11)
PIR F
the peak-to-peak approach and 8.5 ppmv for the lock-in amplifier.
More realistic limits are given for ±3 which results in CMIN of is applied to compare the white-cell with the conically shaped
34 ppmv for Vp−p and 25 ppmv for VLI . This is clarified in Fig. 11 chamber which is related to the adsorption band of ethylene. The
where the reaction of the signals Vp−p , VLI and VRMS to 20 ppmv lower the CME-value, the better the system efficiency related
ethylene concentration steps is shown. to ethylene detection. To get comparable results for two sys-
It is in evidence that the signal levels of Vp−p can be barley tems using Eq. (11) the bandwidth F has to be referenced
assigned to the 20 ppmv steps, whereas this information disap- approximately to the same centre wavelength. For the white cell
pears in noise for VRMS . The signal VLI of the lock-in amplifier system with PIR = 2.5 W, CC2 H4 ,MIN=20 ppmv and F = 0.424 ␮m
shows the best behaviour related to the concentration, followed a value of CME = 19 ppmv W−1 ␮m−1 was calculated. For the
by Vp−p . Signal VRMS provides only poor information about the conically shaped measurement setup with the MIRL-17 IR-source
concentration. a CME of 20 ppmv W−1 ␮m−1 is obtained with PIR = 0.56 W,
A. Sklorz et al. / Sensors and Actuators B 175 (2012) 246–254 253

CC2 H4 ,MIN = 34 ppmv and F = 3 ␮m. This shows that the conical bandwidth F at CW . The sensitivity of the signal VY calculated by
chamber based system used in this paper achieves CME values the signal processing method Y to ethylene gas is calculated using
which are a just above the values of the white cell system. But it
has also taken into account that this result was achieved without (VY )
SENY,C2 H4 = (A.4)
a gilded chamber, without a lock-in amplifier and with electrically CC2 H4
chopped IR-sources. For the detection limit of 25 ppmv which
was achieved with the laboratory lock-in amplifier a CME of The signal-to-noise ratio is defined as
15 ppmv W−1 ␮m−1 is achieved. This shows that there is still
a optimisation potential related to the conically shaped NDIR- (VY,RMS (C))2
system. However it has to be assumed that the low two ppm-digit SNRC = (A.5)
u2Y,noise,RMS
detection region is the basic ethylene sensitivity limitation of
NDIR-systems with applied miniaturised IR-sources and ther-
where C is the concentration change which affects the change
mopile devices. This assumption is confirmed by the calculation
of the signal VY,RMS (C) and the signal processing method Y. The
results. Probably the detection limit could be lowered by a few
noise equivalent concentration was calculated using
ppmv improving the reflectivity of the internal chamber walls
and optimisation of the optical path length, but on balance the uY,noise,RMS
efforts needed for that would override the advantages provided by NECY = (A.6)
SENY,TP,C2 H4
NDIR-detectors. Additionally the sensitivity of NDIR-systems can
be improved indirectly by preconcentrator application [21].
for the signal processing method Y.

5. Conclusions
References
In this paper passive approaches for the improvement of the
signal intensity and detection limit of NDIR-detectors for ethyl- [1] Y. Wild, R. Scharnow, M. Rühmann, Containerhandbuch, in Fachinformatio-
nen der Deutschen Transportversicherer. vol. Band III: Gesamtverband der
ene gas detection were suggested. By applying conically shaped
Deutschen Versicherungswirtschaft e.V., 2005.
measurement chambers and low-cost Fresnel lenses the radia- [2] W. Lang, et al., The “Intelligent Container”-a cognitive sensor network for trans-
tion measurement efficiency was improved from 0.8% to 4.16% in port management, Sensors Journal, IEEE 11 (3) (2011 March) 688–698.
the best case. In experiments the influence of three digital sig- [3] R. Jedermann, et al., Linking keeping quality models and sensor systems to an
autonomous transport supervision system, presented at the Kreyenschmidt, J
nal processing methods and different IR-sources on the detection & Petersen, B. (ed.) Cold Chain-Management, Proceedings of the 2nd Intern.
limit were examined and compared to results provided by a lock- Workshop Cold Chain-Management, University Bonn, Germany, 2006.
in amplifier. A standard peak-to-peak signal processing method [4] R. Shekarriz, A high sensitivity continuous ethylene monitoring device for
postharvest applications, in 2007 Joint annual meeting of the Florida State Hor-
provided a good detection limit of 34 ppmv. The detection limit ticulture Society and the Soil and Crop Science Society of Florida, Palm Beach
achieved with the lock-in amplifier was 25 ppmv. This is compara- Gardens, Florida, 2007, pp. 251–255.
ble with results obtained with white-cell systems in other works. [5] S. Benedetti, et al., Electronic nose as a non-destructive tool to characterise
peach cultivars and to monitor their ripening stage during shelf-life, Posthar-
Beside the experiments an equation for detection limit estimation vest Biology and Technology 47 (2008) 181–188.
was derived. The calculation results using this equation match with [6] A. Sklorz, et al., Low level ethylene detection using preconcentrator/sensor
the measurement results and predict a achievable ethylene detec- combinations, Sensors, 2010 IEEE (2010) 2494–2499, 1–4 November.
[7] A. Sklorz, A. Schafer, W. Lang, Merging ethylene NDIR gas sensors with
tion limit of NDIR-systems within the two digit ppmv-range.
preconcentrator-devices for sensitivity enhancement, Sensors and Actuators
B: Chemical 170 (2012) 21–27.
Acknowledgement [8] J. Fonollosa, et al., Ethylene optical spectrometer for apple ripening monitoring
in controlled atmosphere store-houses, Sensors and Actuators B: Chemical 136
(March (2)) (2009) 546–554.
This research project (’The Intelligent Container’) is supported [9] J. Hildenbrand, et al., A compact optical multichannel system for ethylene mon-
by the Federal Ministry of Education and Research (BMBF), itoring, Microsystem Technologies 14 (2008) 637–644.
[10] A. Sklorz, D. Mrugala, W. Lang, IR-ethylene concentration measurement in
Germany, under reference number 01IA10001.
fruit logistics, in: Smart Systems Integration 2009 Brussels, Belgium, 2009, pp.
286–574.
Appendix A. Equations [11] Edmund optics GmbH, Infrared Fresnel Lense, In: Catalog: Optics and Optical
Instruments, 2006, Stock Number: E43-794.
[12] Perkin Elmer Optoelectronics, TPS 434: Thermopile Detector, Perkin Elmer Cor-
The Lambert–Beer’s law is defined as poration, Boston, MA, 2003.
[13] Perkin Elmer Optoelectronics, G9 and G20: optical filter, Perkin Elmer Corpo-
IDET = I0 exp(−X lCX ) (A.1) ration, Boston, MA, 2007.
[14] HawkEye Technology LLC, IR-55 Datasheet (URL: http://www.boselec.com/
with the radiation intensity IDET affected by IR-absorption, the products/documents/IR-55and-50 000.pdf), Distributed by Boston Electronics
unaffected intensity I0 , the optical path length l, the gas specific Corporation, Brookline MA (last access 30.09.11).
extinction coefficient X and the gas concentration CX of the gas [15] Electro Optical Technolgies, MIRL-17-900 Pulsed Broadband Infrared Light
Source Datasheet, POLCENIGO (Pordenone), Italy, 2006.
X. The high-state measurement signal VHI is defined as the mean [16] Datasheet, HSL EMIRS Series IR Sources for broad range Gas Analysis, Heimann
value of the measurement signal VO2 between the points in time Sensors and Leister Microsystems, 2004.
tHI,0 and tHI,0 [17] H.-R. Tränkler, E. Obermeier, Sensortechnik - Handbuch für Praxis und Wis-
senschaft 1st edition, Springer, Berlin, 1998.
VHI = V O2 (t)|ttHI,1
HI,0
(A.2) [18] P.J. Linstrom, W.G. Mallard (Eds.) NIST Mass Spec Data Center, S.E. Stein,
director, Infrared Spectra, National Institute of Standards and Technology,
Gaithersburg MD, 20899, http://webbook.nist.gov, (retrieved 31.07.09).
The values of tHI,0 and tHI,0 are taken at VO2 (tHI,0 ) = VO2 (tHI,0 ) =
[19] H. Naumann, G. Schröder, Bauelemente der Optik, 6th ed., Hanser Verlag,
0.99max {VO2 (t)}. The radiation measurement efficiency is defined München, 1992.
as [20] W. Konz, J. Hildenbrand, M. Bauersfeld, S. Hartwig, A. Lambrecht, V. Lehmann,
J. Wollenstein, Micromachined IR-source with excellent blackbody like
PTP
S = (A.3) behaviour, Proceedings of SPIE 5836 540 (2005).
Prad [21] A. Sklorz, A. Schafer, W. Lang, Merging ethylene NDIR gas sensors
with preconcentrator-devices for sensitivity enhancement, Sensors
with the theoretical radiated power Prad and the measured radia- and Actuators B: Chemical, available online 1.12.10, ISSN 0925-4005,
tion power PTP . Both values are assigned to the same optical filter 10.1016/j.snb.2010.11.049.
254 A. Sklorz et al. / Sensors and Actuators B 175 (2012) 246–254

Biographies the middle of 2011. Currently he is a research associate and responsible for the
research topic sensors, sensor nodes and RFID-systems in the logistics research project
‘The Intelligent Container’ at the Institute for Microsensors, -Actuators and -Systems
Adam Sklorz received the Dipl.-Ing. degree in electrical and information engineering (IMSAS) in Bremen.
at the University of Bremen in 2005 on mixed-signal ASIC design. He is currently
a research associate at the Institute for Microsensors, -Actuators and -Systems Walter Lang studied physics at Munich University and received his diploma in 1982
(IMSAS) in Bremen and coordinates the logistics research group at IMSAS and the on Raman spectroscopy of crystals with low symmetry. His Ph.D. in engineering at
cluster project ‘The Intelligent Container’ (http://www.intelligentcontainer.com). Munich Technical University was on flame-induced vibrations. In 1987 he joined
From 2007 till 2010 he was involved into research activities of the FWBI the Fraunhofer Institute for Solid State Technology in Munich, where he worked on
Friedrich-Wilhelm-Bessel-Forschungsgesellschaft mbH. His main research focus is microsystems technology. In 1995 he became the head of the sensors department at
the development and integration of complex autonomous sensor and measurement the Institute of Micromachining and Information Technology of the Hahn-Schickard
systems, in particular on the development of mobile ethylene gas detection systems Gesellschaft (HSG-IMIT) in Villingen-Schwenningen, Germany, working on sensors
for logistic applications. for flow and angular rate, sensor test and modelling. In 2003 Walter Lang joined the
University of Bremen. He is heading the Institute for Microsensors, -Actuators and
Steffen Janßen studied electrical engineering at the University of Bremen, Germany -Systems (IMSAS) and he is the speaker of the Microsystems Center Bremen (MCB).
and received his diploma during his work at the Fraunhofer Institute for Silicon His projects cover sensors and microfluidic systems, sensor networks for logistic
Technology (ISIT) in 2010. He worked at Plan Optik AG as a research engineer until applications and the embedding of sensors in sensorial materials.

You might also like