RoboGas

Inspector
A Mobile Robotic System for Remote Leak Sensing and
Localization in Large Industrial Environments: Overview and First Results

Samuel Soldan. Gero Bonow. Andreas Kroll

University of Kassel, Faculty of Mechanical Engineering, Department of
Measurement and Control, Mnchebergstrae 7, 34125 Kassel, Germany
(e-mail: {samuel.soldan, gero.bonow, andreas.kroll}@mrt.uni-kassel.de)
Abstract: In order to automate the routine inspections in large industrial environments a mobile robotic
system is being developed in the project RoboGas
Inspector
. The robots sensor-head consists of different
instruments for remote sensing using multiple measurement principles. First results show that passive IR-
thermographic imaging can be used to detect leaks in pipes as well as liquid spills on the ground. The
measurable effects are the temperature profile disturbance due to expansion of pressured gas and
evaporative cooling, respectively. Tunable Diode Laser Absorption Spectroscopy (TDLAS) measurement
systems provide for quantitative gas concentration measurements and feature a high sensitivity due to the
active measurement principle. On the contrary, TDLAS systems measure just the concentration along a
path at a time and require a diffusely reflecting background. Using a Pan-Tilt Unit, objects/areas can be
scanned and abnormal gas concentrations can be identified. This contribution introduces the project
RoboGas
Inspector
and presents the used measurement technology as well as first results.
Keywords: Service robotics, industrial inspection, remote leak detection, TDLAS, thermography.

1. INTRODUCTION
Within the oil and gas industry, remote operation of offshore
production facilities is of key interest due to the extremely
demanding environments. Of particular importance is tail-end
production due to the technically and commercially difficult
situation (Vatland et al. 2007). During normal operation,
many platforms are already remotely operated, while
specialized tasks such as inspection and maintenance still
require humans on site (Skourup and Pretlove 2011). This has
driven research in mobile robotics to enable remote operation
of such tasks. Proposed approaches include ground-based
autonomous mobile robots for inspection (Graf and Pfeiffer
2008) and portal robots for inspection and intervention
(Skourup and Pretlove 2011).
Aforementioned work addresses majorly mobility and
accessibility problems. This contribution addresses the
problem of detecting and localizing fluid leakages in plants
and industrial infrastructure using mobile robots. The
detection of leaks in fluid transportation and processing
equipment is of high relevance in order to prevent harm to
humans, nature and assets or just to prevent financial losses.
Stationary installed in-situ sensors are used in plants to
monitor high-risk areas on a sample base. An area-wide
coverage of extended facilities is, however, a costly
undertaking. Therefore, non-high-risk areas; that are still
subject to possible leakages, are in general not monitored by
gas sensors. In fact, routine inspections are commonly carried
out using smell, hearing, and eyesight. Formal inspections,
particularly of pressurized equipment, are carried out during
major revisions, e. g. each 3rd or 5th year using manually
operated measurement equipment and manual data analysis.
This article provides an overview of the project
RoboGas
Inspector
(section 2). This targets the development of a
human-machine system with autonomous mobile inspection
robots equipped with remote sensing technology. The aim is
to automate routine inspections for fluid leakages in plants
and infrastructure. The use of remote sensing technology with
mobile robots instead of in-situ (commonly semiconductor)
gas sensors is a key differentiator to related work, see the
overview in (Kowadlo and Russel 2008) and oil and gas
specific work (Graf and Pfeiffer 2008, Persson and Anisi
2011). Details on research problems, principles and first
experimental results for remote sensing of gas leakages and
liquid spills using active spectroscopic (TDLAS) and passive
infrared (IR) thermographic measurement systems are
provided in sections 3, 4 and 5, respectively. The paper
concludes with a summary and an outlook.
2. PROJECT OVERVIEW
A review of the state of the art and perspectives of robotic
inspections of industrial sites (Kroll 2008) inspired the
project RoboGas
Inspector
(2011), which started in Dec. 2009. It
is a joint research venture with four research, three industrial
development and two application partners (a petroleum
refinery and a gas transportation & distribution provider).
The project consortium is led by the University of Kassel.
The advantages of the planned system addresses
Quality of process and results: improved inspection due
to mobile remote gas sensing technology, central
knowledge base increases locally available knowledge,
increased coverage of extended inspection areas and
increased inspection frequency with available staff
Proceedings of the 2012 IFAC Workshop on Automatic
Control in Offshore Oil and Gas Production, Norwegian
University of Science and Technology, Trondheim,
Norway, May 31 - June 1, 2012
ThA2.1
Copyright held by the International Federation of
Automatic Control
33




Operator relief: easier inspection of remote areas by
remote sensing technology, automation of repetitive,
monotonous routine tasks, automated measurement
evaluation and results documentation
Profitability: more efficient and effective deployment of
specialist staff, more time for human operators for
activities that create higher value such as planning,
supervision, and optimization, efficient deployment of
sophisticated measurement devices


Fig. 1. Prototypic remote sensing head: A thermal camera (2),
TDLAS device (3), laser scanner (4), video camera (5) and a
thermographic camera for gas visualization are mounted on a
pan-tilt unit (1). The computer and other electronics are
housed in a control cabinet (7).

Fig. 2. Robot prototype with remote sensing technology.
The project activities have been grouped to the major
working areas of Gas propagation and measurement,
Mobile robots and Human-robot-interaction.
Gas propagation and gas measurement: To better understand
the behavior of gas propagation from leaks, typical
elementary leakage scenarios have been defined and are
simulated using CFD (Computational Fluid Dynamics) tools.
The simulations are validated with large scale outdoor gas
dissemination experiments. The remote sensing is carried out
by active and passive IR spectroscopic and thermographic
technology. A prototypic sensor head has been designed and
realized (Fig. 1). Measuring, processing and pattern
recognition strategies as well as gas-detection/leak-
localization algorithms have been developed.
Mobile robot platform: This working area covers the
assembly of robot prototypes (see Fig. 2 for an impression of
a prototype of the robotic system). Besides, mobility- related
functions such as path and task planning for single-/multi-
robot systems, autonomous path following, obstacle and
collision avoidance are developed.
Human-robot Interaction: To increase the operators
acceptance of the overall system, the human is always in
charge of supervisory control of the system. This means not
only mission planning and supervision, but also user
intervention if problems occur as well as enhanced human-
robot interactions with the system (like using gesture control
or augmented reality).
3. RESEARCH PROBLEMS AND APPROACH
The main research area for the authors is remote sensing of
gas. With integral directional measurement devices (e.g.
TDLAS in section 4) the problem is a) the detection of an
abnormal gas concentration given varying background
concentrations as well as varying path lengths and b) the
localization of the leak source as a point in space and not as a
direction. Ideas for possible strategies have been adopted
from nature (bionics) as well as from engineering/statistics
and have been extensively tested in simulations.
Thermal imaging devices on the other hand provide for an
intensity matrix where the temperature disturbance due to a
leak can be detected (see section 5) and also devices exist that
can visualize the escaping gas (e.g. FLIR GF320). Here the
questions are related to image processing, pattern recognition
and machine learning. Also adding data from other sensors
(e.g. depth or video) can provide helpful information and
support the classification task. The algorithms are developed
and tested in controlled lab environments and then transferred
to testing in real industrial environments.
4. TDLAS: PRINCIPLE AND FIRST EXPERIMENTAL
RESULTS
Optical remote gas measurement devices based on the
tunable diode laser absorption spectroscopy (TDLAS) have,
in contrast to classical in-situ gas sensors, the advantage that
they dont need to enter a gas plume to make a concentration
measurement. Cycle times of TDLAS measurement devices
are shorter (t
In-Situ
[1, 90] s vs. t
TDLAS
[0.01; 0.1] s,
(Bonow, Kroll 2011)), permitting to observe larger areas in a
given period. The disadvantage however is that they provide
for integral directional gas concentrations (typically in
ppmm: parts per million meter). Therefore, the position of a
gas plume on the measurement path cannot be located from a
single measurement.
4.1 Principle of measurement
Many technical gases like volatile hydrocarbons (e.g.
Methane (CH
4
) or Butane (C
4
H
10
)) absorb radiation at
different wavelengths in the IR spectrum. The absorption
coefficient ( ) is unique for every gas and can be used to
identify the gas (finger print). If it is known, the average gas
concentration C within the optical path length L can be
determined by using the Lambert-Beer Law:
Copyright held by the International Federation of
Automatic Control
34




( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) L C I
dx x C I I
A
L
A M
=
=



exp
exp
0
(1)
whereas
A
I is the measured light intensity without gas,
M
I
the measured light intensity with gas and ( ) x C the local gas
concentration at position x on the measurement path of the
length L. Reassembling (1) provides for
( ) ( ) ( ) ( )
1 1 1
ln

= L I I C
A M
. (2)
When using a remote gas sensor, L is in general unknown,
and only the integral gas concentration C
~
can be determined:
( ) ( ) ( ) ( )
1 1
ln

= =
A M
I I C L C
~
(3)
4.2 Remote gas measurement device RMLD
The RMLD emits an IR-laser beam and analyzes the light
backscattered from any diffusely reflecting surface in the
measurement path (e.g. walls, components, plants, (Frish et
al. 2005)), thus
R
L L = 2 where
R
L is the geometric path
length of the measurement setup. The IR-laser diode of the
RMLD can be tuned within a narrow spectral range. To
determine the integral gas concentration, the laser is at first
tuned to a reference wavelength
R
, where methane has only
a negligible influence on the measurement ( ( )
R A
I ). For the
second measurement, the laser is tuned to the wavelength of
an absorption line of methane ( ( )
4 CH M
I ). Based on these
two measurements and assuming that the absorption
characteristic of the atmospheric gases at the two
wavelengths is constant or negligible ( ) ( ) (
4 CH A R A
),
the integral gas concentration can be calculated using (3).
As already mentioned, the RMLD can be used with most
surfaces under different viewing angles. Own experiments
have shown that most surfaces can be used up to a viewing
angle of 70 (angle between surface normal and optical
axis of the RMLD). Smooth or polished surfaces, e.g. made
of metals like aluminum or stainless steel, that are observed
normal to the surface (viewing angle of 0 to 5 and
distance < 5 m), can reflect too much of the laser radiation
and saturate the detector. In outdoor industrial environments,
such surfaces are rare because most components (pipes,
vents, flanges, etc.) are painted, coated or have a thin patina
due to oxidation (rust).
Another problem occurs in case of measuring while the
device is moving: The panning velocity has a direct
influence on the minimal detectable gas concentration (Frish
et al. 2007). This is because the RMLD averages the
measured concentration in a time window of
M
t 100 ms
(update frequency of device is
M
f = 10 Hz), thus the scanned
volume at s / > 0 is larger than in the static case (note
that the laser has a cone shape with an opening angle of
RMLD
1, so always a volume is measured). Conversely,
the RMLD reports a smaller gas concentration if it is panned
and the dimension of the plume orthogonal to the optical axis
is smaller than the scanned volume. Fig. 3 shows the
estimation of the minimal gas concentration within a 90
mm cylinder (e.g. petri dish) that can be distinguished from
natural methane in the atmosphere for different panning
velocities and path lengths. In this test a gas concentration
has been marked as distinguished if the RMLD readings
exceed 130% of the background concentration. The results in
Fig. 3 base on a mathematical sensor model with
experimentally identified parameters.
Simple field tests and tests in the lab with the RMLD have
shown that gas leaks with a mass flow of m&

2 mg/s
(methane) can be distinguished from the natural methane
concentration (
A
C 2-4 ppm) by an operator up to a path
length of
R
L =30 m (wind speed v 2-3 m/s orthogonal to
the measurement path). The displayed values during the tests
from the RMLD varied between ( ) m L C
~
R
30 =

= 70-110
ppmm and ( ) m L C
~
R
2 = = 220-320 ppmm, where the
measured average atmospheric methane was C 2 ppm.

Fig. 3. Minimal methane concentrations in a 90 mm
cylinder dish that can be distinguished from the natural
methane concentration (RMLD reading > 130%
A
C
~
) of
A
C = 2 ppm in the atmosphere for different measurement
path lengths and panning velocities.
In this context, false alarms can arise if the alarm threshold
used by the gas detection strategy is too low. On the contrary,
with a high threshold, small leaks can be missed. This
false/missed alarms are predominantly the result of the
different path lengths and panning velocities of the sensor in
combination with the measurement uncertainty of the RMLD
(1 = 3.7 ppmm).
4.3 Using path length information for TDLAS-based
measurements: First results
For robust automatic leak localization, it is necessary to
determine the direction of the maximal gas concentration
discernible from current location (Baetz et al., 2009). It is
assumed that a gas leak is always located next to a surface.
With the known position and orientation of the measurement
device and path length to the maximum gas concentration, a
potential leak position can be estimated. For this, a 2D laser
scanner (Sick LMS 151) on a pan-tilt unit (Schunk PW90) is
used to get 3D-depth information of the environment. These
data can be used to calculate the average gas concentration
for every RMLD measurement using (2).
Fig. 5 shows the result from a 3D depth scan of an
outstretching utility corridor with depth information and Fig.
6 shows the measured integral methane in ppmm produced
Copyright held by the International Federation of
Automatic Control
35




with a RMLD raster scan with reduced pan velocity
( = 10/s) of the area of interest. After the raster scan, the
depth information of the 3D-depth scan is used to calculate
the average gas concentrations (Fig. 7). In this step, the
lateral shift between the optical axis of the measurement
devices are taken into account.
Because of the considered path length and adapted panning
velocity, the alarm threshold for the strategies can be
reduced. Due to this approach, it is more likely to distinguish
small gas leaks from atmospheric methane by using the path-
length-independent average gas concentration than integral
gas concentration.


Fig. 4. Photography of an outstretched utility corridor.

Fig. 5. Depth scan of the utility shaft from Fig. 4.

Fig. 6. Integral gas concentration measured from the
viewpoint.

Fig. 7. Averaged gas concentration calculated from Fig. 6.
4.4 Discussion
By using the path length information, the RMLD
measurements can be interpreted more easily than the integral
gas concentration. In this context, the recommended fixed
threshold strategy with a threshold of
A
C
~
= 160 ppmm (Frish
et al, 2005) can theoretically be lowered to the maximal
natural atmosphere methane concentration ( 4 ppm L).
However, due to the natural variation and due to the
measurement uncertainties of the PTU angles, laser scanner
readings and RMLD readings a higher value is to be
expected. In addition, the search for the maximum gas
concentration becomes easier, because variations in the
measured concentrations due to a changing path length have
now only a minor influence on the measurement. Up to now
the compensation strategy has been tested under ideal
laboratory conditions. A field test is planned to assess the
robustness of the strategy. Moreover, the choice of a
threshold for reliable automatic detection of small leaks
under real world conditions has to be examined.

5. IR-THERMOGRAPHY: PRINCIPLES AND FIRST
EXPERIMENTAL RESULTS
5.1 Basic principles and measurement effect
Every surface with a temperature above absolute zero (0 K)
radiates energy. Its electromagnetic spectrum is in the IR-
wavelength region from 1-100 m. The spectral distribution
of the ideal black body is given by Plancks law as the
radiance per unit wavelength:
1
2
5
0
2
exp 1

( | |
=
| (

\
h c h c
L
k T

, (4)
with h: Planck's constant, c: speed of light in vacuum, :
wavelength,
0
: solid angle, and T: absolute temperature of
the target surface. The total power radiated per unit area of a
body for the entire spectrum is given by the Stefan-
Boltzmann law as the radiant exitance:

4
T M = (5)
with : Stefan-Boltzmann constant, : emissivity ( := 1 for
black body target). With increasing object temperature, not
only the total emitted energy changes, but also the maximum
of the wavelength distribution shifts to shorter wavelengths.
The position of the maximum of the distribution is given by
Wiens displacement law:
T b =
max
(6)
with Wiens displacement law constant b = 2898 mmK. In
processing plants, typical temperatures of pipes and vessels
range from ambient temperature to 300 C and therefore, the
maximum of the radiation lies in the range of 5-10 m.
Due to the absorption of mainly water and carbon dioxide,
the atmosphere has a high transmission only for some IR-
spectral regions, referred to as atmospheric windows. The
8-14 m window fits well for the targeted temperature range.
Moreover, long wavelengths are less disturbed by the sun in
Copyright held by the International Federation of
Automatic Control
36




case of outdoor application. The used IR camera (Infratec
VarioCAM hr head) works in the 7.5-14 m-spectral
region.
At thermal equilibrium, the power of any incident radiation at
the objects surface has to match the sum of absorbed,
emitted and transmitted power. A black body target absorbs
100% of the incident radiation (i.e. 1 = ), a real body only a
fraction ( 1 <
real
). The emission coefficient of a real
body is given as the ratio of absorption of the real body vs.
the black body:
real
= . A resulting problem for
thermography is that a real body does not only radiate
corresponding to its surface temperature, but reflects thermal
radiation of surrounding heat sources. The sensor measures
the totally emitted radiation. For instance, polished metals
have emission coefficients in the range of 15 0 01 0 . . ,
meaning they specularly reflect thermal radiation of
surrounding objects. Corroded or unfinished iron has 7 0.
and paint typically 99 0 95 0 . . , which is almost
independent of color. Therefore, thermography is difficult to
apply to uncolored, polished metal surfaces.
If gas escapes from a higher to a lower pressure reservoir
without significant heat exchange with the environment, it
expands, cools down and cooling the leaking object in the
vicinity of the leak. Due to heat conduction, the temperature
field disturbance propagates dependent on the thermal
conductivity k of the material. Therefore the thermograpically
sensed object temperature profile at the surface is disturbed
by the leak. Metal conducts heat well (e. g. at 0 C:
copper k = 386 W/(mK), carbon steel with C = 0.5%
k = 55 W/(mK), stainless steel (V2A) k = 16.3 W/(mK)
(Maldague, 2001)). For this reason, the temperature field
disturbance of leaking stainless steel pipes has spot type and
is more extended for copper objects.
A liquid (e.g. water, petrol) with the surface exposed to the
atmosphere will cause the liquid to vaporize without boiling.
This phenomenon is called evaporation. The required energy
causes cooling of the liquid or objects in the vicinity.
Thermography can be used again to detect liquid spills on a
surface based on the temperature profile.
5.2 Thermographic case study results
A measurement campaign has been designed and executed
with pressured air leaks in pipes and instruments. A test stand
was constructed to shield radiation from the environment
(Fig. 8). Different pipes and instruments that vary in material,
diameter, surface condition and leak type/size were tested
under lab conditions. Pressured air was chosen because it is
neither toxic nor explosive and easy to handle.
The measurements from the campaign clearly show a
temperature drop around the leak area. For a leak diameter of
0.5 mm and a pressure drop of 5 bar, the cooling effect is
around 0.5 K. Fig. 9 illustrates that the cold spot varies with
the material due to different heat conductivities and different
surface conditions resulting from different emissivity.
The data were used to develop automatic pattern matching
algorithms, which worked well in a lab environment. See
Kroll et al. (2009) and Baetz et al. (2010) for further details.


Fig. 8. Test stand for thermographic inspection (top), leaking
pipes (bottom left) and instruments (bottom right).

Fig. 9. Measurement results for different pipes with gas leaks.
The false color code measurement spans of: A: 0.6 K, B: 1.2
K, C: 0.5 K, D: 0.9 K.

Fig. 10. Lab inspection robot with first-gen. sensor head.
For the detection of liquid spills, a simple case study was
developed to test the feasibility. Water was poured on a
basement floor 1 h before the measuring. The location was
chosen to mimic a typical production environment. The
sensors were mounted on a lab robot (Fig. 10) and
measurements were taken with video and thermographic
cameras. The water spill is not easy to detect in the video
Copyright held by the International Federation of
Automatic Control
37




image but stands out in the thermographic image (Fig 10).
The temperature difference between spill and dry floor is 1.5
K. Fig. 10 also includes a gas leak in a pipe above the spill.



Fig. 11. Water spill on the floor and gas leak in the pipe in
video image (top) and thermogram with temperature range
from 20C to 22.1C (bottom).
5.3 Discussion
These first results on detection of small temperature effects
(T < 2 K) look promising. However, Maldague (2001)
states that in case of passive IR thermography T = 12 K
is a first vague indication and T > 4 K a clear indication for
an abnormality. Despite the technological advances (i.e.
temperature resolution <0.03 K), it is assumed that these
values are still valid for real-life unaided single-image
thermography. The reason is that the condition of the target is
unknown in general (unknown variations in emissivity vs.
real variations in temperature). For this reason, the authors
expect straight thermographic leak detection in an
industrial setting to be challenges. Therefore, strategies that
exploit further information (depth, color, multiple
measurements) are investigated.
6. SUMMARY AND OUTLOOK
The mission for the project RoboGas
Inspector
is the simulation-
based design and evaluation of a human-machine system with
autonomous mobile inspection robots for remote IR-optical
gas leakage detection and localization in plants. At the time
of writing, the project is half-finished and preliminary results
in the working area of remote fluid sensing are available.
Future work on the inspection module includes improvement
of the detection and localization algorithms as well as data
processing. A gas visualization camera will be added shortly.
Tests in a gas compressor station and a petroleum refinery are
scheduled to evaluate performance in an industrial setting.
ACKNOWLEDGMENT
The project RoboGas
Inspector
is funded by the Federal Ministry
of Economics and Technology due to a resolution of the
German Bundestag. The authors would like to thank the
project partners and Dr. W. Baetz for valuable discussions
and support.
REFERENCES
Baetz, W., Kroll, A. and Bonow, G. (2009). Mobile Robots
with Active IR-Optical Sensing for Remote Gas
Detection and Source Localization. IEEE Int. Conf. on
Robotics and Automation (ICRA 2009), Kobe, Japan,
pp. 2773-2778.
Baetz, W., Kroll, A. and Soldan, S. (2010). On Gas Leak
Detection of Pressurised Components by Using
Thermograms and Pattern Recognition Algorithms. 8th
Int. Conf. on NDE, Berlin, Germany, pp. 503-512.
Bonow, G. and Kroll, A. (2011). Zur automatisierten
Inspektion von Anlagen mittels Gasfernmesstechnik:
Technologien und Gerte. Proc. of the AUTOMATION
2011, Baden-Baden, Germany, pp. 381-384.
Frish, M.B. et al. (2007). The next generation of TDLAS
analyzers. Society of Photo-Optical Instrumentation
Engineers Optics East, vol. 6765, pp. 7650676517.
Frish, M.B. et al. (2005). Extended Performance Handheld
and Mobile Sensors for Remote Detection of Natural Gas
Leaks - Phase II: Final Report, [Online]
http://www.netl.doe.gov/technologies/oil-
gas/publications/td/ 41603_FinalReport.PDF, 11/2011.
Graf, B. and Pfeiffer, K. (2008). Mobile Robotics for
Offshore Automation. Proc. of the IART/EURON
Workshop on Robotics for Risky Interventions and
Environmental Surveillance. Benicassim, Spain.
Kowadlo, G. and Russell, R.A. (2008) Robot odor
localization: a taxonomy and survey. Int. Journal of
Robotics Research, vol. 27, no. 8, pp. 869-894.
Kroll, A. (2008). A survey on mobile robots for industrial
inspection. Int. Conf. on Intelligent Autonomous Systems
(IAS10), Baden-Baden, Germany, pp. 406-414.
Kroll, A., Baetz, W. and Peretzki, D. (2009). On
Autonomous Detection of Pressured Air and Gas Leaks
Using Passive IR-Thermography for Mobile Robot
Application. IEEE Int. Conf. on Robotics and
Automation (ICRA 2009), Kobe, Japan, pp. 921-926.
Maldague, X.P.V. (2001). Theory and Practice of Infrared
Technology for Nondestructive Testing. Wiley & Sons.
Persson, E. and Anisi, D. A. (2011). A Comparative study of
robotic gas source localization algorithms in industrial
environments. IFAC World Congress, Milano, Italy, pp.
899-904.
Skourup, C, and Pretlove, J. (2011). Remote inspection and
intervention. ABB Review, no. 2, pp. 50-55.
RoboGas
Inspector
. (2011). RoboGas
Inspector
- Project Website.
[Online] www.robogasinspector.de, 03/2012.
Vatland, S., Doyle P. and Andersen, T.M. (2007). Integrated
operations: Creating the oil company of the future. ABB
Review, no. 3, pp. 72-75.
Copyright held by the International Federation of
Automatic Control
38