Sensors and Actuators B 113 (2006) 830–836

Design of a portable optical sensor for methane gas detection

Crawford Massie a , George Stewart a,∗ , George McGregor b , John R. Gilchrist b
a Department of Electronic and Electrical Engineering, The University of Strathclyde, 204 George Street, Glasgow G1 1XW, UK
b Gas Measurement Instruments Ltd., Inchinnan Business Park, Renfrew PA4 9RG, UK

Received 18 January 2005; received in revised form 11 March 2005; accepted 14 March 2005
Available online 23 May 2005

Abstract

A detailed investigation has been carried out on the design of a low-cost portable optical sensor for methane detection with a sensitivity
of ∼1% of the Lower Explosive Level (LEL) for methane (500 ppm) and able to operate in harsh environments with temperature variation
between −20 and 50 ◦ C. The sensor design is based on the use of near-IR LEDs operating around the overtone absorption lines of methane at
1660 nm using a stainless steel tube to direct the light through the gas to the detectors. Various configurations of source/detector layout have
been examined to provide appropriate reference and signal paths in order to achieve reliable methane detection at LEL levels in the presence
of temperature variation. An optimum design has been identified using two detectors with appropriate optical filtering and with temperature
stabilisation of the source and detectors. Based on this design, a prototype instrument has been demonstrated with an ultimate sensitivity of
0.2% LEL methane (100 ppm).
© 2005 Elsevier B.V. All rights reserved.

Keywords: Methane sensor; Optical gas sensor; Portable gas sensor; Hydrocarbon gas sensor

1. Introduction based on semiconductor sources, have the potential advan-

tages of: (i) intrinsically safe, (ii) ability to detect a specific
Detection of methane gas is extremely important for safety gas by selection of appropriate wavelengths, (iii) able to
reasons in the oil and gas industries, in water treatment plants, operate in zero-oxygen environment (e.g. for purging of pipe
in landfill sites and in commercial or domestic environments, lines), and (iv) low cost of ownership, since the gas-detection
where methane gas may filter up through the ground and cre- principle is a physical process (not a chemical reaction), and
ate an explosion hazard (the lower explosive limit, LEL, for therefore, poisoning of the sensor is not an issue (although
methane is 5% by volume methane gas). Portable gas detec- dirt/contamination on the optics needs to be considered in
tors for locating gas leaks are clearly important in a number the sensor design). For hydrocarbon gases, such as methane,
of these applications, but must be of a low-cost design for the strongest optical absorption occurs in the mid-IR region,
widespread use and availability. The established method for around 3.3 ␮m, but the use of a broadband source in the
methane gas detection is the catalytic sensor [1–3] which, mid-IR to detect methane [5] has cost and performance
however, is not methane-specific, as any gas whose ignition limitations. For example, a number of hydrocarbon gases
is catalysed by the pellistor will be detected. Additionally, have absorption bands/lines in the same spectral region, so
catalytic sensors do not operate correctly in low-oxygen a mid-IR system will often respond to other hydrocarbons,
environments and can be poisoned. Consequently, such and hence, they are generic hydrocarbon detectors rather
detector systems may require frequent functional checks. than methane-specific. Mid-IR detectors are relatively
It is therefore of commercial interest to develop portable expensive and are often cooled [6] in order to achieve an
gas sensors, based on optical techniques [4]. Optical sensors, enhanced signal-to-noise ratio, which increases the power
consumption—a disadvantage for portable sensors.
∗ Corresponding author. Tel.: +44 141 548 2887; fax: +44 141 548 2926. One way to overcome some of these problems is to op-
E-mail address: g.stewart@eee.strath.ac.uk (G. Stewart). erate the sensor at overtone absorption lines in the near-IR,

0925-4005/$ – see front matter © 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.snb.2005.03.105
 C. Massie et al. / Sensors and Actuators B 113 (2006) 830–836 831

around 1665 nm for methane [7–9]. In this spectral region,

inexpensive LEDs and detectors are widely available, and
detectors operate with low noise at room temperature (DFB
laser sources are preferred [7] but are expensive for portable
sensors). The chief drawback is that the absorption strength
of second harmonic lines in the near-IR is at least two orders
of magnitude weaker than fundamental mid-IR absorption
lines. This presents a challenge in sensor design, especially
when a considerable range of environmental temperature is
involved.
The objective of this work was therefore designing and
development of an inexpensive, rugged, portable, near-IR
methane sensor, with sufficient sensitivity for commercial
application (<1%LEL or equivalently <500 ppm1 ), and able
to operate over an environmental temperature range of −20
and +50 ◦ C. Hence, although the basic scientific principles
are well-known, the practical engineering design requires a
number of factors to be considered in the design including
instrument size, cost, power consumption and immunity to
environmental conditions.

2. Theoretical description of absorption from

near-IR LED sources

We assume that the LED has an intensity distribution I(ν),

with wavenumber ν, as illustrated in Fig. 1(a). If the total
intensity emitted by the LED is designated by I0 , then defining
a normalised f (v) = I(v/I0 ), we can write:

∞
∞
Fig. 1. (a) General intensity distribution of LED against wavenumber; (b)
I0 = I(ν)dν = I0 f (ν)dν (1) simple “block” approximation for the intensity distribution of the LED along
0 0
with multiple absorption lines; (c) effect of strong absorption on LED source
distribution.

∞
where f (ν)dν = 1. overlap integral between the source distribution and the ab-
0 sorption cross-section profile.
If the output of the LED is passed through an absorbing Absorption data and spectral line positions in the near-
gas in a cell of length l, and assuming that the absorption is IR are available on the Hitran database [10,11] for methane
weak and does not change the source distribution, the output (CH4 ), but not for other hydrocarbon gases, such as, ethane
intensity from the cell is given by: (C2 H6 ), propane (C3 H8 ) and butane (C4 H10 ), which are im-
I = I0 exp{−αeff Cl} (2) portant to consider in the design of a methane-specific sen-
sor, as discussed later. Spectral line positions for these gases
where the “effective” absorption coefficient is defined by: were, therefore, measured experimentally on a spectrome-

∞ ter, and are shown in Fig. 2. For methane, Fig. 2 shows
αeff = N0 σ(ν)f (ν)dν (3) the absorption lines around 1660 nm, with the P, Q and R
branches. The Q branch contains a number of individual lines
0
which are not resolved in this diagram; the strongest is the
here, σ(ν) is the absorption cross-section per molecule of Q6 line at 1665.5 nm with an absorption of α = 0.25 cm−1
the gas, N0 = 2.5 × 1019 molecules/cm3 is the atmospheric at the line centre [7]. The Hitran database [10,11] provides
concentration of molecules at 25 ◦ C, and C is the gas information on line strengths, and gives an integrated line
concentration fraction (C = N/N0 , 0 < C < 1, C = ppm/106 strength2 of S = 5.172 × 10−20 cm2 molecule−1 cm−1 for all
or Cl = (ppm m)/106 , with N as the concentration of gas the methane lines between 1637 nm (ν2 = 6107 cm−1 ) and
molecules per unit volume). Note that Eq. (3) represents the 1697 nm (ν1 = 5891 cm−1 ).

1 ∞
For methane, concentration in ppm is given by multiplying the percentage 2 The line strength is defined by S = σ(ν)dν.
of LEL by 500. 0
832 C. Massie et al. / Sensors and Actuators B 113 (2006) 830–836

Fig. 2. Measured absorption bands and line spectra in the near-IR for several hydrocarbon gases.
 C. Massie et al. / Sensors and Actuators B 113 (2006) 830–836 833

We can quickly estimate the effective absorption coeffi- is predicted to give ∼0.6% change in light intensity on pas-
cient with the above data by assuming an idealised spec- sage through a 12 cm cell length with 50%LEL methane and
tral distribution, as shown in Fig. 1(b), for an LED with similar values were obtained in practice.
∼60 nm spectral width and centred on 1667 nm. In this case, It should be noted that the above theoretical description
f (v) = 1/(v2 − v1 ), and the integral in Eq. (3) takes the sim- is based on the condition that αCl  1 for each absorption
ple form: line within the spectral bandwidth considered. For high
concentrations and/or long path lengths, the strongest

ν2
N0 N0 S absorption lines may deplete the available light intensity at
αeff = σ(ν)dν = (4)
(ν2 − ν1 ) (ν2 − ν1 ) their respective wavelengths, as shown in Fig. 3(c), and the
ν1 LED intensity distribution is no longer constant along the
Substituting values gives an effective absorption coeffi- cell; so, as a result, the absorption signal does not change
cient of 0.006 cm−1 or, from Eq. (2), approximately 0.15% linearly with concentration. However, the work here is con-
change in light intensity on passage through a 10 cm cell cerned with obtaining sufficient sensitivity for LEL levels
length with 50%LEL methane (approximately 40 times of methane, so the assumption of αCl  1 is justified, as for
weaker than using a narrow-linewidth laser source centred the strongest Q6 line with α = 0.25 cm−1 and 100%LEL (5%
on the Q6 line). methane) in a 12 cm length cell, we have αCl ∼ 0.1.
However, the effective absorption coefficient can be im-
proved by appropriate filtering of the LED, blocking spectral
regions that are not affected by gas while targeting strongly 3. Portable sensor design options
absorbing regions, such as the central Q-band lines. Ef-
fective absorption coefficients for filtered sources can be Various LED and detector configurations were considered
estimated as above, using the HITRAN database which in the development of the near-IR portable methane detector,
provides individual line strengths (hence, the integrated all based around the same gas detection cell, as shown in
line strength for all lines within the bandwidth of a filter Fig. 4. The cell consists of a 12 cm long polished stainless
through summation). For example, for 20 and 7 nm filters, steel tube with an internal diameter of 1 cm, providing
both with centre wavelength on the Q6 line at 1665.5 nm, a sealed volume through which gas can be drawn. The
the integrated line strengths are S20nm = 2.65 × 10−20 and multiple reflections from the polished walls of the tube cell
S7nm = 2.14 × 10−20 cm2 molecule−1 cm−1 , giving effective provide a light pipe that guides light from the source to the
absorption coefficients of 0.0092 cm−1 and 0.0212 cm−1 , re- detector, greatly increasing the amount of light reaching the
spectively. The theoretical plot of Fig. 3 for filters, centred detector and increasing the effective absorption path length
on 1665.5 nm, shows the marked effect on the effective ab- of the cell. For all cases, a phase-sensitive detection (PSD)
sorption coefficient, as the filter linewidth is reduced due to method was used for signal extraction to attain a small
the dense spacing of lines within the Q-band. detection noise-bandwidth of ∼1 Hz. By 1800 out-of-phase
In practice, the filter tolerances, angular alignment and pulsing of the LEDs (or inversion of one output signal
temperature effects, all become more critical, as the filter for the single LED/two-detector system), an active (gas)
linewidth is reduced, and hence, a compromise is often nec- signal and a reference signal is created consisting of pulses
essary between sensitivity, cost and stability of the final in- of similar magnitude but opposite phase. The two signals
strument. With these considerations in mind, a 7 nm width are appropriately scaled and summed, giving a fixed DC
was used in this work, as discussed later in Section 3. This level (with no target gas present in the cell). When gas is
present, the signals will be unbalanced, giving an output
at the pulse modulation frequency (∼1 kHz), and hence,
the PSD produces an output voltage related to the gas
concentration.
The first design considered was based on a dual LED sin-
gle detector system, as shown in Fig. 4(a). Here, one of the
LEDs is chosen to emit at wavelengths outside the absorp-
tion band of the target gas (methane), and provides the op-
tical reference signal to compensate for changes in optical
intensity due to dirt or condensation, etc., in the cell, since
both LEDs will be affected in a similar way. Experimental
tests on this configuration, however, revealed problems in
sensor operation over the required temperature range. This
arises from the different temperature coefficients of the two
LEDs (which are, of necessity, different models, since they
Fig. 3. Effective absorption coefficient as a function of filter line-width for operate at different wavelengths), which can lead to false sig-
filters centred on 1665.5 nm. nals from temperature change. Fig. 5 shows the temperature
834 C. Massie et al. / Sensors and Actuators B 113 (2006) 830–836

Fig. 6. (a) Photograph of a standard detector and the lensed detector; (b)
comparison of absorption from 100% methane using lensed and unlensed
detectors for single and double path length configurations.

shown in Fig. 4(b). This second detector, positioned at the

Fig. 4. Various configurations for source/detector layout in the sensor cell. LED end of the cell, monitors the output of both LEDs (in an
(a) Dual source, single detector; (b) dual source with stabilisation detector;
area shielded from gas). If there is any intensity difference
(c) folded path; (d) single source, dual detector.
between the LEDs, the stabilisation detector circuitry feeds
a signal back into the reference LED-drive circuitry to adjust
characteristics of the two LEDs used, namely, Hamamatsu the drive current and compensate for the intensity difference.
L8245 (1650 nm) and L7866 (1300 nm reference), giving By using this process, the output of both LEDs can be main-
temperature coefficients of approximately −0.0372 dB/◦ C tained at the same intensity. However, for this system to op-
for L8245 and −0.0242 dB/◦ C for L7866. One possible so- erate correctly, the two detectors will have to track each other
lution to this problem is to introduce a second detector, as with temperature, and because the detectors are at opposite
ends of the cell, this is not easily achievable. Also, the number
of components and cost is increased with this approach.
Because temperature drift may be a problem and absorp-
tion by the gas is very weak in the near-IR, it is clearly ben-
eficial if the absorption signal can be improved through an
increased path-length without having to change dimensions
of the stainless steel cell (and the final instrument). Hence,
the configuration shown in Fig. 4(c) was investigated, where
the detector and the source are positioned at the same end of
the cell, with a mirror at the opposite end of the cell, appar-
ently doubling the available path length. Experiments were
carried out on this layout, using a standard G8421-03 InGaAs
photodiode as the detector, and then repeated with a G8421-
03 photodiode that used a lens instead of a window on the
Fig. 5. Relative radiant output against ambient temperature for Hamamatsu photodiode can, as shown in Fig. 6(a). The mirror used was a
LEDs L8245 (1650 nm active LED) and L7866 (1300 nm reference LED). 15 × 60 mm concave mirror from Edmund Scientific. Results
 C. Massie et al. / Sensors and Actuators B 113 (2006) 830–836 835

from tests are shown in Fig. 6(b), where it can be seen that filter is manufactured at the edge of the central wavelength
very little improvement was observed for the standard detec- tolerance. Similar concerns apply for the reference filter, but
tor, but the lensed detector showed an increase by a factor of an additional point to be considered is the behaviour of the
∼1.7. The reason for these results lies in the acceptance an- LED spectral distribution with temperature. If the position of
gles of the detectors. The lensed detector mainly accepts light the reference filter is badly chosen, a situation similar to the
that has travelled directly down the cell and back again. How- dual LED set-up could arise, where the active and reference
ever, the unlensed detector will pick up a substantial amount channel behave very differently with temperature. To be safe,
of back-scattered light from inside the cell which does not the reference filter was chosen to have its centre wavelength
have a large path-length through the gas. close to that of the active filter (1670 ± 1.5 nm), but far away
The final configuration and the one considered to be the enough and with a sufficiently narrow bandwidth (3 nm) to
best option is illustrated in Fig. 4(d), consisting of a single avoid the Q-band absorption lines.
LED and two detectors, each with a narrowband spectral fil- In order to make the sensor methane-specific, and to avoid
ter to create reference and active channels. The active filter is cross-sensitivity to other gases, it is also important that the
chosen to have a spectral distribution over a strongly absorb- wavelengths chosen for both the active and reference LEDs
ing region of methane absorption lines, whereas the reference lie outside the absorption lines or bands of other hydrocarbon
filter has a spectral distribution outside the absorption lines. gases (ethane, propane and butane) that may be present when
As discussed in Section 2, the narrowband filter used for monitoring gas leaks in domestic or commercial environ-
the active channel will also increase the effective absorption ments. Careful examination of the experimentally-measured
coefficient, thereby increasing the measured absorption. spectral data of Fig. 2 shows that the choice of wavelengths
Moreover, this configuration uses two detectors of the same for the LEDs (1665.5 and 1670 ± 1.5 nm) does avoid ab-
model number and at the same position within the cell. sorption by the other gases and this was confirmed through
Common changes in detector performance with temperature independent tests performed during the design stages of the
will tend to cancel, when the reference and active signals are sensor.
compared, and hence, temperature sensitivity should be less As noted, temperature stability is important to minimise
of a problem than a dual LED system. We shall now discuss drift and false alarms. To improve the temperature character-
the experimental design and test of this system in some detail. istics of the sensor, the two detectors were housed in a single
aluminium block to increase the thermal mass and ensure the
same temperature at both detectors. The block was heated
4. Design and test of the dual detector system using a 5.6 , 0.5 W resistor to raise the temperature above
ambient to ∼35 ◦ C and was maintained at this temperature
The dual detector system was constructed, using a single by a control circuit using a thermistor sensor. Also the cell
Hamamatsu L8245 LED and two Hamamatsu InGaAs PIN was sealed using sapphire windows which also shield the
photodiodes, G8371-01. In the design, a number of factors LED and detectors from the cooling effect created when gas
need to be considered carefully, including: (i) the wavelength is drawn through the cell.
and bandwidth chosen for the filters to create the active Fig. 7 shows the fully assembled prototype portable sen-
and reference channels, and (ii) temperature stability of the sor. For the instrument, the electronic gain was set so that
sensor. For the active filter, the target wavelength should be 0–1 V range corresponded to 0% to 100%LEL methane.
within the dense Q-band absorption line structure, centred
on 1665.5 nm (see Fig. 2), with a narrow bandwidth of a few
nanometres or less (see Fig. 3). However, in practice, other
factors limit the choice in filter specifications, for example,
cost of manufacture, temperature effects on the centre
wavelength, tilt sensitivity, etc. The cost of manufacture
increases with narrowing bandwidth and tighter tolerances.
The manufacturing tolerances are important because if the
filter bandwidth is sufficiently narrow, for example 3 nm, and
the tolerance on the filter centre wavelength is sufficiently
large, for example ±1.5 nm, then some filters within a batch
production may include very little of the Q-band absorption
lines. For our sensor, the specification chosen for the active
filter had a centre wavelength of 1665.5 ± 1.5 nm with a
bandwidth of 7 ± 0.15 nm. This was found to provide a
reasonable compromise in terms of tolerances, cost, and
temperature sensitivity, and gives considerable improvement
in absorption over a non-filtered LED (see Fig. 3), but is Fig. 7. Photograph showing the fully-assembled prototype portable methane
still wide enough to have a satisfactory performance if the sensor.
836 C. Massie et al. / Sensors and Actuators B 113 (2006) 830–836

Johnstone of Strathclyde University are gratefully acknowl-

edged.

References

[1] P.T. Mosely, B.C. Tofield (Eds.), Solid-State Gas Sensors, Adam
Hilger, Bristol, 1987, pp. 17–50.
[2] P. Krebs, A. Grisel, A low-power integrated catalytic gas sensor,
Sens. Actuators B: Chem. 13–14 (1993) 155–158.
[3] H. Debeda, L. Dulau, P. Dondon, F. Menil, C. Lucat, P. Massok, De-
velopment of a reliable methane detector, Sens. Actuators B: Chem.
44 (1997) 248–256.
[4] P.T. Mosely, J.O.W. Norris, D.E. Williams (Eds.), Techniques and
Fig. 8. Time response of the instrument to 1.6%LEL and 50% LEL methane. Mechanisms in Gas Sensing, Adam Hilger, Bristol, 1991, pp.
234–259.
[5] A. Krier, V.V. Sherstnev, Powerful interface light emitting diodes for
To further improve the stability of the reading, output sig- methane gas detection, J. Phys. D: Appl. Phys. 33 (2000) 101–106.
nals were averaged over 32 samples through software fil- [6] C.H. Wang, J.G. Cowder, V. Mannheim, T. Ashley, D.T. Dutton,
tering. A time response trace of the instrument output for A.D. Johnson, G.J. Pryce, S.D. Smith, Detection of nitrogen dioxide
1.6%LEL (800 ppm) and 50%LEL methane is shown in using a room temperature mid-infrared InSb light emitting diode,
Electr.Lett. 34 (3) (1998) 300–301.
Fig. 8. The magnitude of change for 1.6%LEL is ∼40 mV, [7] B. Culshaw, G. Stewart, F. Dong, C. Tandy, D. Moodie, Fibre op-
and the change in signal for 50%LEL is ∼580 mV. The ul- tic techniques for remote spectroscopic methane detection—from
timate sensitivity of the detection system is limited by the concept to system realisation, Sens. Actuators B: Chem. 51 (1998)
signal to noise ratio and was determined to be ∼0.2%LEL 25–27.
(100 ppm). [8] G. Stewart, C. Tandy, D. Moodie, M.A. Morante, F. Dong, B. Cul-
shaw, Design of a fibre optic multi-point sensor for gas detection,
Sens.Actuators B: Chem. 51 (1998) 227–232.
[9] G. Whitenett, G. Stewart, K. Atherton, B. Culshaw, W. Johnstone,
5. Conclusion Optical fibre instrumentation for environmental monitoring applica-
tions, J. Opt. A: Pure Appl. Opt. 5 (2003) S140–S145.
In this paper, we have described in detail the design [10] L.S. Rothman, C.P. Rinsland, A. Goldman, S.T. Massie, D.P. Ed-
wards, J.-M. Flaud, A. Perrin, C. Camy-Peyret, V. Dana, J.-Y.
and construction of a low cost, rugged, portable, methane- Mandin, J. Schroeder, A. McCann, R.R. Gamache, R.B. Wattson,
specific sensor. The optical sensor has potentially a low cost K. Yoshino, K.V. Chance, K.W. Jucks, L.R. Brown, V. Nemtchino,
of ownership compared with the standard pellistor sensor and P. Varanasi, The HITRAN molecular spectroscopic database and
the component cost is relatively low through use of standard HAWKS (HITRAN atmospheric workstation, 1996 edition, J. Quant.
LEDs and detectors operating in the near-IR region around Spectrosc. Radiat. Transfer 60 (5) (1998) 665–710.
[11] L.S. Rothman, R.R. Gamache, A. Goldman, L.R. Brown, R.A. Toth,
1660 nm. However, due to the low absorption from overtone H.M. Pickett, R.L. Poynter, J.-M. Flaud, C. Camy-Peyret, A. Barbe,
absorption lines in the near-IR, the main challenge is to N. Husson, C.P. Rinsland, M.A.H. Smith, The HITRAN database,
achieve the required sensitivity over a wide temperature range 1986 edition, Appl. Opt. 26 (19) (1987) 4058–4097.
of −20 ◦ C to +50 ◦ C and the best design option for achieving
this has been identified. A prototype instrument has been
constructed with a sensitivity down to 0.2%LEL (100 ppm). Biographies
The knowledge gained from this investigation provides a
good basis for the development of portable sensors for other Crawford Massie graduated with a BSc(Hon) in Physics from the Uni-
hydrocarbon gases such as ethane, propane and butane and for versity of Paisley in 1998, and an MSc in Lasers & Optoelectronics
from the University of St. Andrews in 1999. He worked as a patent an-
the development of portable laser-based systems for higher alyst with Derwent Information before taking up an Associate Post with
sensitivity. Gas Measurement Instruments (GMI) Ltd. under the Knowledge Trans-
fer Partnership (KTP) programme with the University of Strathclyde, in
2002.
Acknowledgments George Stewart is currently a Reader at the University of Strathclyde,
Glasgow. He was awarded a BSc Degree with First Class Honours in
The authors gratefully acknowledge the financial sup- 1974, and a PhD degree for research on integrated optics in 1979, both
port provided through a Knowledge Transfer Partnership from the University of Glasgow. He was appointed as a Research Fellow
at the University of Glasgow until 1985, and then joined the Photonics
(KTP3876) between The University of Strathclyde and Gas Group at Strathclyde University to work on fibre optic components, sen-
Measurement Instruments Ltd. Assistance provided by Pro- sors and fibre lasers. Dr. Stewart is the primary author of five chapters
fessor G Duxbury in obtaining spectral data for gases and in various specialist textbooks and is author/co-author of more than 100
technical discussions with Professor B Culshaw and Dr W technical and scientific papers in the field of fibre and integrated optics.