Design of an Infrared Imaging System for Robotic ... - SAGE Journals

International Journal of Advanced Robotic Systems

ARTICLE

Design of an Infrared Imaging System for Robotic Inspection of Gas Leaks in Industrial Environments Regular Paper

Ramon Barber 1*, Miguel A. Rodriguez-Conejo2, Juan Melendez2 and Santiago Garrido3 1 University Carlos III of Madrid, Leganes, Madrid, Spain 2 LIR-Infrared Laboratory, Physics Department, University Carlos III of Madrid, Spain 3 RoboticsLab Systems Engineering and Automation, University Carlos III of Madrid, Spain *Corresponding author(s) E-mail: [email protected] Received 28 July 2014; Accepted 15 December 2014 DOI: 10.5772/60058 © 2015 The Author(s). Licensee InTech. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract Gas detection can become a critical task in dangerous environments that involve hazardous or contaminant gases, and the use of imaging sensors provides an impor‐ tant tool for leakage location. This paper presents a new design for remote sensing of gas leaks based on infrared (IR) imaging techniques. The inspection system uses an uncooled microbolometer detector, operating over a wide spectral bandwidth, that features both low size and low power consumption. This equipment is boarded on a robotic platform, so that wide objects or areas can be scanned. The detection principle is based on the use of active imaging techniques, where the use of external IR illumina‐ tion enhances the detection limit and allows the proposed system to operate in most cases independently from environmental conditions, unlike passive commercial approaches. To illustrate this concept, a fully radiometric description of the detection problem has been developed; CO2 detection has been demonstrated; and simulations of typical gas detection scenarios have been performed, showing that typical industrial leaks of CH4 are well within the detection limits.

The mobile platform where the gas sensing system is going to be implemented is a robot called TurtleBot. The control of the mobile base and of the inspection device is integrated in ROS architecture. The exploration system is based on the technique of Simultaneous Localization and Mapping (SLAM) that makes it possible to locate the gas leak in the map. Keywords gas remote sensing, infrared imaging, robotic inspection, robotic application

1. Introduction The research carried out in this paper has been mainly inspired by the project Robauco [1]. This project’s main objective is the development of technologies to obtain mobile robots capable of complex tasks that demand a high degree of autonomy and capacity of collaboration in the presence of human beings. One of the main applications of this line of work is the use of mobile robots in dangerous missions [2] where the environment can be risky for humans (e.g., rescue missions, etc.). In this case, an unman‐ Int J Adv Robot Syst, 2015, 12:0 | doi: 10.5772/60058

1

ned autonomous vehicle (UAV) is sent in advance to explore and obtain images of the area. A very desirable feature for these robots is a sensor capable of detecting dangerous gases in the environment. The same requisite appears within the context of gas leak detection, in a large number of applications in sectors such as the energy and chemical industries [3]. Since inspection tasks are often routine and may appear in dangerous or difficult-to-access places, the use of robotic systems has received considerable attention. Some typical gas detection problems are: • Inspection of pipelines: Pipes are used mainly for the transportation of natural gas (composed of 80 % meth‐ ane) and hydrocarbons in general. There are lines with lengths of hundreds of kilometres that must be regularly inspected. • Inspection of electrical transformers: Refrigerant and insulating gases with high dielectric constants are often used; for instance, sulphur hexafluoride (S F 6), which is also one of the main greenhouse gases.

• Leak detection in industrial processes. Both the chemical and the metallurgical industries use in many processes compounds that are toxic to one degree or another. Such industrial plants are constantly subjected to controls of their emissions, and have to inspect large stretches of facilities to locate and repair existing leaks. The main goal of this work is the design of an imaging system for the remote sensing of toxic or dangerous gases from a mobile robot, to be used in industrial environments. An imaging sensor has the obvious advantage of providing spatial resolution, making it possible to precizely locate the source of the gas. The challenge here is to develop a sensor that is able to provide a ’gas image’ in real time, at least with semi-quantitative concentration information, being small enough both in size and in power consumption to be placed on a mobile robot. This paper is organized as follows. Section 2 summarizes the state of the art in gas sensing from robotic platforms; section 3 reviews the principles of infrared remote sensing of gases; section 4 explains the measurement strategies proposed in this work and estimates the minimum detect‐ able concentrations for C H 4. Section 5 describes a specific

robotic implementation for an imaging gas sensor system operating along the lines proposed in this work. Finally, Section 6 shows the performance of the overall platform. 2. Gas sensing in robotic platforms: state of the art

Gas sensing devices that may be found in a robotic system can be classified into in-situ systems, where in order to obtain the measurement the gas must be in direct contact with the sensor, and remote sensing systems, that provide measurements without contact. Among remote sensing systems, those that provide images deserve specific treatment. 2

Int J Adv Robot Syst, 2015, 12:0 | doi: 10.5772/60058

2.1 In-situ systems To inspect an area, in-situ systems require the robot explore the whole region. Despite this limitation, these sensors are mostly used in experimental robotic systems, primarily due to their low cost and ease of integration with microcontrol‐ lers. They are typically encapsulated devices that transduce the variations in the concentration of a given gas into variations of electronic magnitudes (resistance, voltage, etc.), and can be based on a variety of technologies and operating principles [4]. The technology most commonly used for in-situ gas sensors is known as MOX (metal oxide) [5], since it provides high sensitivity for the gases of greater interest, short response time and good stability as compared to other alternatives. MOX sensors are also easily available and very inexpensive devices (although they often require a notinexpensive prior calibration). On the other hand, one of their major limitations is low selectivity. Although their sensitivity to a specific gas can be increased, they do not respond to a single compound but are sensitive to many gases. To try to overcome this drawback, arrays of MOX sensors are often constructed so that each element has a different sensitivity, although overlapped to some extent with others. The response is then analysed by array-based pattern recognition algorithms to identify and quantify the gas present. This concept was introduced in the 1980s and is known as electronic nose or e-nose [6]. Since the in-situ sensors themselves do not provide information about the position in which the measurement takes place, this information must be obtained by other means. Therefore, despite its widespread use, this type of sensor is not well fitted to the task of locating a leak of a particular gas with enough accuracy for most applications. However, algorithms have been developed to produce maps of concentrations using on-board sensors on mobile robots [7]. 2.2 Remote sensing systems Regarding remote sensing gas sensors for robotic systems, several devices have been used that are based on TDLAS technology (Tunable Diode Laser Absorption Spectrosco‐ py). The measurement principle is infrared (IR) absorption. A single absorption line of the target gas is selected, together with a reference band in which the gas has no absorption. The decrease of intensity at the absorption wavelength is directly related to the product concentration ⋅ optical path ( ppm ⋅ m), as will be explained in the next section. In contrast with MOX sensors this method is very selective, since absorption lines are extremely specific for each gas. Wavelength variation is achieved by using a tuneable laser, which has the advantage of having very low response times that allow the incorporation of detection strategies based on the modulation of both the wavelength and the power employed [8], thus increasing sensitivity.

A specific development for remote measurement based on TDLAS technology is the device known as a Remote Methane Leak Detector [9]. It incorporates lighting from a laser in the near infrared (NIR) next to the detector element; this radiation is reflected by the background, crosses the cloud of gas and is subsequently detected by the detector element. Although these are intrinsically non-imaging devices, when placed on robotic platforms they can be provided with a scanning system that allows the visuali‐ zation of the distribution of the product concentration⋅ optical path of a specific gas more easily, making them able to locate leaks with greater accuracy and efficiency than insitu systems. 2.3 Imaging systems Although it is possible to find different imaging systems for remote sensing of gases (e.g., the GasFindIR system manufactured by FLIR, operating in the absorption band of hydrocarbons [10]), they are still very uncommon in the field of robotics. Currently there are a number of proposals for the development of mobile robots and even unmanned aerial vehicles (UAVs) that incorporate this technology to be employed in inspection tasks. But in spite of this, today is not possible to find a device of this nature that is com‐ mercially available [11]. One of the most ambitious projects, developed by the University of Kassel (Germany), is called RoboGasInspector: A Mobile Robotic System for Remote Sensing Leak and Locali‐ zation in Large Industrial Environments [12]. The project aims at the automation of routine inspection tasks taking place in large industrial plants. A prototype of a mobile robot has been developed that, in addition to a laser scanner and a video camera used for navigation and location, has three sensor elements: a conventional thermal camera in the range of 8 − 12μm (FLIR model 320), a RMLD device for hydrocarbon detection equipped with a scanning system, and a FLIR GasFindIR camera for gas leak detection. The RMLD device and the GasFindIR camera are dedicated commercial instruments for gas leak detection, but the conventional IR camera can also be used for that applica‐ tion, since leaks frequently produce anomalies in temper‐ ature, either cold (due to adiabatic cooling on expansion) or hot (when hot gases escape to the environment). Although this system marks the current state of the art in the field of robotic systems for remote sensing of gases, no data are presented on the imaging of leaks, and improve‐ ment of algorithms both for remote sensing and for leak location are mentioned as future works.

each molecule, the IR absorption spectrum of a gas is a ’spectral signature’ that uniquely identifies each chemical species. Homonuclear molecules, however, such as O2, N 2, or Ar , do not absorb IR radiation, because vibrations or

rotations do not change their dipole moment. This is in fact an advantage for IR remote sensing of toxic or pollutant gases, since the major components of atmosphere are transparent in that spectral band. On the other hand, typical examples of compounds with infrared signatures [13] are carbon dioxide (CO2), carbon monoxide (CO ), hydrocar‐ bons (Cx H y ), nitrogen oxides ( N Ox ), sulphur oxides (SOx ),

VOC’s (Volatile Organic Compounds), and sulphur hexafluoride (S F 6), among many others. Infrared absorption makes possible not only the identifica‐ tion, but also the quantification, of gases. When an electro‐ magnetic wave strikes an object it can be, to a greater or lesser extent, absorbed, transmitted and reflected. If we represent by α , τ , and ρ the respective energy fractions (called absorptance, transmittance and reflectance), the following relation is fulfilled according to the principle of conservation of energy:

a +t + r = 1 If, instead of a solid or liquid object, we consider a region inside a gas (like, for instance, a pollutant leak), reflectance is zero (ρ = 0), leading to:

a +t = 1 Transmittance at a particular wavelength depends on the concentration of the gas and the optical path travelled, as given by the Beer-Lambert law [14],

t (T , l ) =

I - a T , l ×c ×l =e ( ) I0

(1)

where : • I 0 is the intensity incident on the medium. • I is the intensity remaining after being absorbed by the medium. • a(T ,λ) is a gas-specific parameter called absorptivity, that depends on wavelength and temperature. • c refers to the concentration of a specific gas.

3. Principles of infrared remote sensing of gases

• l is the distance the radiation travels through the medium, usually known as the optical path.

A photon, in order to be absorbed by a material medium, must have an energy that corresponds to the difference between two quantum levels. Transitions between vibra‐ tion and rotation levels of gas molecules have energies in the IR range and, since these levels are highly specific for

Values of absorptivity are well known for the main gases, and can be found tabulated in databases like HITRAN [15]. Figure 1 shows a typical transmittance spectrum obtained with HITRAN on the web, a web implementation of HI‐ TRAN [16]. The gas is CO2, but the overall pattern is similar

Ramon Barber , Miguel A. Rodriguez-Conejo, Juan Melendez and Santiago Garrido: Design of an Infrared Imaging System for Robotic Inspection of Gas Leaks in Industrial Environments

3

0 2260

for all gases: a large number of individual narrow lines (for ambient temperature and pressures, typical widths are less than a nanometre) grouped in bands of characteristic profiles, that may be several hundreds of nanometres wide. This extremely strong spectral structure can be used to an advantage in gas detection, because signal and reference spectral channels can be defined by using wavelengths with and (respectively) without IR absorption. This is done at the line level in TDLAs, but can be done also at the band level using interference filters, whose bandwidths in the medium IR band are typically in the range of a few tens of nanometers; i.e., wider than absorption lines but narrower than absorption bands. 1 0.9 0.8

Transmittance

2320 2340 2360 Wavenumber (cm−1)

2380

2400

Lnoleak − Ll

Figure 1. Transmittance spectrum for a path of 1 m, 0.1% CO2 , at 296 K and 1 atm total pressure. The x axis is in units of wavenumbers, ν = λ−1 . A vibration-rotation band, spanning several hundreds of nanometers (the scale ranges from 2250cm−1 = 4.44µm to 2400cm−1 = 4.16µm) is made up by several hundreds of very narrow individual absorption lines.

Clearly, if Tg = will be indistingu one-band, passiv work insofar as th background.

4. Measurement strategy

This is very often

The simplest approach is to provide the camera with aambient tempera Any gas with an absorption band at an atmospheric band-pass spectral filter centred at the band. Due toexpansion as they window can be, in principle, detected with an IR camera. usually is not co absorption, the leaking be seen as a dark The simplest approachgas is towill provide the camera withregion a thermal contrast band-pass spectral filterThis centered at the passive band. Due to against the background. one-band, approach it is weighted b absorption, the leakingsystems gas will be a dark regionFLIRfor small leaks ( is used in commercial asseen the aswell-known against the background. This one-band, passive approach, detect if their tem GasFind IR.in commercial systems as the well-known FLIR is used

background. In alarms, since an can be mistaken 4.1 Radiometric model 4.1. Radiometric model with a complex, an industrial env This This simple idea must be precize, however, since it focuses simple idea must be precised, however, since it on the radiance onlyfocuses on the IRonabsorption of the leak, only the IR absorption of the leak,but but ignores ignores its feature. Comm its emission. A simple radiometric model is necessary in emission. A simple radiometric model is necessary in ordertemporal domain order to understand the problem. the gas flow of to understand the problem. variations in app processing is a p Lleak = L BB ( Tb ) · τ +Gas leak L BB ( Tg ) · (1 − τ ) advisable not to r , Tg for gas leak ident BB is Planck’s blackbody radiance, T and T are, laminar stationar where LBlackbody g b GasFind IR.

Background

0.6 0.5 0.4 0.3 0.2 0.1 2280

2300

2320 2340 2360 Wavenumber (cm−1)

2380

2400

Figure 1. Transmittance spectrum for a path of 1 m, 0.1% Figure 1. Transmittance spectrum for a path of 1 m, 0.1% CO2, at 296 K CO2 , at 296 K and 1 atm total pressure. The x axis is in units

wavenumbers, ν The = λ− . A isvibration-rotation band, spanning ν = λ −1. A andof 1 atm x1 axis total pressure. in units of wavenumbers, several hundreds of nanometers (the scale ranges from 2250cm−1 = vibration-rotation band, spanning several hundreds of nanometres (the 4.44µm to 2400cm−1 = 4.16µm) is made up by several hundreds of −1 scalevery ranges fromindividual 2250cm −1 = 4.44μmlines. to 2400cm = 4.16μm) is made narrow absorption up of several hundreds of very narrow individual absorption lines.

4. Measurement strategy

Infrared gaswith detection need notband be based on radiation Any gas an absorption at an atmospheric absorption, since gases also emit radiation same window can be, in principle, detected with anatIRthe camera. The simplest approach is to provide the other camera with a wavelengths where they absorb. As in any substance, band-pass spectral filter centered at the band. Due to the emitted radiance (W / m 2sr ) is given by absorption, the leaking gas will be seen as a dark region against the background. This one-band, passive approach, is used in commercial systems as the well-known FLIR Lgas (l ) = LBB Tg , l × e l GasFind IR.

(

) ( )

4.1. Radiometric model

where ℒ BB (T g ) is the emission of a blackbody at tempera‐

This simple by ideaPlanck’s must be precised, since it ) is a gas-specific ture law) and ε (λhowever, T g (given focuses only on the IR absorption of the leak, but ignores fact, since Kirchoff’s parameter calledAemissivity, 0 ≤ ε ≤ 1. In its emission. simple radiometric model is necessary in laworder states that under the condition of thermodynamic to understand the problem. equilibrium the absorptance α equals the emissivity α = ε , it turns out that for a gas ε + τ = 1 and thus ε = 1 − τ . Therefore,

calculation of the transmittance spectrum provides also the Gas leak , Tgprovided its temperature is emission spectrum of the gas, Blackbody known. On the other hand, for a solid, τ = 0 and ε = 1 − ρ . For Background a blackbody, ρ = 0 and ε = 1 ; for a perfect reflector, ρIR = 1 and camera ε = 0. Tb 4. Measurement strategy Any gas with an absorption band at an atmospheric Figure can 2. A be, scheme of the measurement configuration, window in principle, detected with an IRassuming camera. blackbody background and no external illumination.

4

2300

respectively, temperatures of background and gas, andIR we camera have used Tthat e = 1 − τ. The radiance outside the leak b region, Lnoleak , is simply:

0.7

0 2260

2280

Measurement1

Int J Adv Robot Syst, 2015, 12:0 | doi: 10.5772/60058 Suppose the background is a blackbody and, at wavelength λ0 , the gas leak has a transmittance τ and the clean atmosphere is transparent. The radiance Lleak (W/m2 sr) incoming to a camera that looks at the leak against the background (Figure 2) is the sum of two terms:

As a first step to assumption of a more realistically a background ambient tempera noleak BB L = L ( Tb ) background is L Figure 2. A scheme of the measurement configuration, assuming blackbodyreduces to the pr Figurethe 2. Ano scheme of the measurement configuration, And thus contrast between the no-leak andassuming leak background and external illumination Tb , but otherwis background and no external illumination. Measurement1 regionsblackbody is: means that now L Suppose the background is a blackbody and, at wavelengththis does not mea Suppose the background is a blackbody and, at simply that, since noleak BBgas leak BB λ0, the gas transmittance and λ0= , a the τ Lwavelength −leak Lleakhas [L ( Tb ) − L has ( Tτga)] ·transmittance (1the − τclean ) (2)atmos‐ the condition of and the clean atmosphere is transparent. 2 / mradiance sr ) cominghas a temperatur phere is transparent. The radiance ℒleak (WThe leak 2 sr) incoming noleak leak L (W/m to a camera that looks at the leak Clearly, if Tg =that Tb ,looks L at − L leak = 0, i.e., the leak of the gas. into a camera the background the background (Figure 2) isagainst the sum the of two terms: will beagainst indistinguishable from the background. Thus, this Assuming a no background traveled through theradiation leak (Figure 2) is theradiation sum ofthat twohas terms: background one-band, passive approach to leak detection does only possible a strateg (and has been partially absorbed) and radiation emitted that has travelled through the leak (and has been partially work insofar as the gas shows a thermal contrast against the external source t by the leak: absorbed) and radiation emitted by the leak: background. the background

This is very often the case, since gases from conduction at BB considerably BB cooled by adiabatic ambient temperature LleakSy, = =are × t No, + LNo:2013 T × 1 -t (Tb )Vol. 4 Int J Adv Robotic 2013, expansion as they leak (whereas coolingg of very hot leaks usually is not complete near the leak point). However, thermal contrast is not guaranteed in all cases, and since ℒ BB isby Planck’s radiance, T b and it where is weighted a 1 − τblackbody factor, it may be very smallT g are, for small leaks temperatures (τ near 1), that can be very difficult respectively, of background and gas,to and we detect theirthat temperature is radiance not very different from theregion, haveifused ε = 1 − τ . The outside the leak background. In addition, this method is prone to false ℒnoleak , is simply: alarms, since any variation in background temperature can be mistaken for the presence of a leak. In fact, with a complex, non-uniform as is usual in Lnoleakbackground, = = BB Tb an industrial environment, there is no way based only on the radiance values to tell a leak from a background feature. Commercial systems use processing in the And thus the contrast between the no-leak and leak regions temporal domain, taking advantage of non-stationarity in is: gas flow of leaks, to discriminate them from other the variations in apparent temperature. But, although time processing is a possibility that is is always open, it seems Lnoleak - Lleak = éLBB Tb - LBB Tg ù × 1 - t û main criterion (2) advisable not to rely onë time variation as the for gas leak identification, since otherwise a gas leak with laminar stationary flow will be undetectable. Clearly, if T g = T b, ℒnoleak − ℒleak = 0, i.e., the leak will be Asindistinguishable a first step to a more study, let’sThus, relax this the onefrom in-depth the background. assumption of a blackbody background and suppose, band, passive approach to leak detection only works more realistically, a nonzero reflectance ρb (and thus as theemissivity gas shows1 a− thermal contrast against a insofar background ρb ). Now, for an the background. ambient temperature Ta , the radiation coming out of the background is L BB ( Tb ) · (1 − ρb ) + L BB ( Ta ) · ρb . This This istovery often theblackbody case, sincevalue gases from at reduces the previous L BB ( Tb )conduction if Ta = temperature are cooled considerably adiabatic Tbambient , but otherwise is (usually slightly) different, by which means that now Lnoleak − Lleak 6= 0 for Tg = Tb . However, this does not mean any improvement in gas detection, but simply that, since now the background is not a blackbody, the condition of no thermal contrast is achieved when it has a temperature different (but usually not much) to that of the gas.

( )(

)

( )

( )

( ) (

)

expansion as they leak (whereas cooling of very hot leaks is usually not complete near the leak point). However, thermal contrast is not guaranteed in all cases, and since it is weighted by a 1 − τ factor, it may be very small for small leaks (τ near 1), that can be very difficult to detect if their temperature is not very different from the background. In addition, this method is prone to false alarms, since any variation in background temperature can be mistaken for the presence of a leak. In fact, with a complex, non-uniform background, as is usual in an industrial environment, there is no way based only on the radiance values to tell a leak from a background feature. Commercial systems use processing in the temporal domain, taking advantage of non-stationarity in the gas flow of leaks, to discriminate them from other variations in apparent temperature. However, although time processing is a possibility that is always open, it seems advisable not to rely on time varia‐ tion as the main criterion for gas leak identification, since otherwise a gas leak with laminar stationary flow will be undetectable. As a first step to a more in-depth study, let us relax the assumption of a blackbody background and suppose, more realistically, a nonzero reflectance ρb (and thus a back‐ ground emissivity 1 − ρb). Now, for an ambient temperature

T a, the radiation coming out of the background is ℒ BB (T b) ⋅ (1 − ρb) + ℒ BB (T a) ⋅ ρb. This reduces to the previous

blackbody value ℒ (T b) if T a = T b, but otherwise is (usually BB

noleak

leak

−ℒ ≠0 slightly) different, which means that now ℒ for T g = T b. However, this does not mean any improvement

in gas detection but simply that, since the background is no longer a blackbody, the condition of no thermal contrast is achieved when it has a temperature that is different (but usually not much) to that of the gas. IR  source

Non‐ blackbody Background IR  camera

b , Tb

because the light of the source travels twice through the gas. It is clear that this term will increase the sensitivity to gas concentration, irrespective of its temperature. In addition, a term for reflected ambient radiance has to be included (Figure 3). Thus, according to the use or lack of use of external illumination, we have the following cases: Loff = éëLBB (Tb ) × (1 - r b ) + LBB (Ta ) × r b ùû × t + + LBB Tg × ( 1 - t )

(3)

Lon = éëLBB (Tb ) × ( 1 - r b ) + LBB (Ta ) × r b ùû × t + + LBB Tg × ( 1 - t ) + Ls × F × t 2 r b

(4)

( ) ( )

The contrast between the no-leak and leak regions calcu‐ lated previously (eq. 2) corresponds to the difference between values of eq. 3 for the cases τ = 1 (no leak) and τ ≠ 1 (leak), when T a = T b. 4.2 Minimum Detectable Concentrations In order to compare the passive and active approaches, values of ℒoff and ℒon in equations 3 and 4 have been calculated for a C H 4 leak. This gas has been chosen because

of its importance in industrial environments, although analogous calculations can be made for most pollutant gases. Concentrations have been varied from c = 0% to c = 100% , with an optical path of l = 10 cm and a temperature T g = 280K . Background temperatures ranging from T b = 250K to T b = 310K have been considered, and emissivity

has been fixed at ε = 0.8 (so that ρb = 0.2). This is a relatively high value, to prevent a bias against passive modes in our

on an Ab =50 × 50 cm2 area of the background. Thus, the simulation. ambient temperature has been fixed at −5 geometric factor is FThe = A s /Ab = 8 · 10 . The source T a = 280K although results are largely parameters are ,based on the the specifications of the IR-18 independent infrared source from it. manufactured by HawEye Technologies, with a power consumption of 18 watts [17], illuminating the background a distance of approximately 2 m. Planck’s from spectral blackbody equation has been used to

The transmittance spectrum using / m 2srcalculated W been T b, T g , and units ofhas ) for the different obtain ℒ BB (in eq. 1, with the absorptivity provided by the HITRAN T a temperatures. has been by a database. An interferenceThe filtersource was assumed, with modelled a Ambient radiation bandpass 375nm with wide, εcentered the CH absorption C4 and greybody = 0.75 atat1150 an area As =0.2 cm 2, band at λ0 =3.3 µm. The detected radiances Lo f f and that,calculated by means a parabolic reflector,3 focuses Figure 3. A of scheme of the measurement configuration, for a Lon were byof integration of equations and 4, its radiation Figure 3. A scheme the measurement configuration, for a non-blackbody non-blackbody background with external illumination. background with external illumination respectively, that Measurement2 A = 50spectral × 50cm 2channel. area of the background. Thus, the on anover Gas leak , Tg

b

scheme, instead of the passive approach described so far. If In thisgeometric measurement approach, as the source parame‐ factor is F = Asignal ⋅ 10−5. The / A =is8 defined If a the non-blackbody assumed, a strategy radiance of the background source is Ls , anisadditional term Ls · F · tovariation in Lo f f or Lon due to sthe bgas leak; i.e., the 2 ters are based on the specifications of the τ ρb will added to the radiance incomingto touse the camera, improve gasbe detection becomes possible: an externaldifference in L between values calculated, respectively,IR-18 infrared where F is a geometrical term that takes into account the leak gas concentration and gas concentration. source manufactured byzero HawEye Technologies, with a source to illuminate the scene, thereby changing thewith the solid angle of the source as viewed from the background, power consumption of 18 watts [17], the IR background radiation.ofThis usingand an active scheme,The measurement noise can be estimated from theilluminating ρb is the reflectance the means background, τ 2 appears camera data. The key parameter is NETD (noise background from a distance of approximately 2 m. because the light of theapproach source travels twice through instead of the passive described so far.theIf the equivalent temperature difference). A typical value for gas. It is clear that this term will increase the sensitivity s radiance of the source is ℒ , an additional term ℒs ⋅ F ⋅ τ 2ρba modern camera operating in thehas 3 to 5 µmcalculated band The MIR transmittance spectrum been using eq. to gas concentration, irrespective of its temperature. In with a1, quantum detector cryogenically cooledby (for instance will addition, be added to the radiance coming into has thetocamera, a term for reflected ambient radiance be with the absorptivity provided the HITRAN database. an InSb focal plane array) is 15 mK @300K. This means included 3). where F is a(Figure geometrical term that takes into account thethat the Annoise interference filter wasdifference assumed, with athe bandpass 375nm rms value is the between solidThus, angle of the source asorviewed from the background,radiances according to the use not of external illumination, wide, centred at the C Hband absorption band at λ0 = 3.3 μm. The emitted in the whole by two blackbodies 4 have the following cases: 300.015 Kℒ andand 300.000 K. We have ρb iswethe reflectance of the background, and τ 2 appearsat, respectively, ℒ detected radiances were calculated by integra‐ off on preferred, however, to be conservative and use a larger NETD value, namely 50 mK @300K, that is reasonable for , Miguel A. Rodriguez-Conejo, Juan Melendez Santiago Garrido: uncooled microbolometer arrays, cheaper and with and lower Lo f f =[L BB ( Tb ) · (1 − ρb ) + L BB ( Ta ) · ρb ] · τ + Ramon Barber Design of an Infrared Imaging System for Robotic Inspection of Gas Leaks in units Industrial Environments power consumption. Noise expressed in radiance is + L BB ( Tg ) · (1 − τ ) (3) then 3.4 · 10−3 W/m2 sr. It is assumed that this value is not changed by the use of the interference filter. Lon =[L BB ( Tb ) · (1 − ρb ) + L BB ( Ta ) · ρb ] · τ +

+ L BB ( Tg ) · (1 − τ ) + Ls · F · τ 2 ρb

(4)

Figure 4 shows (at the top) the graph of signal

5

tion of equations 3 and 4, respectively, over that spectral channel.

Tb = 280 K

In this measurement approach, the signal is defined as the variation in ℒoff or ℒon due to the gas leak; i.e., the difference

0.03

∆ L (W/m2·sr)

The measurement noise can be estimated from the IR camera data. The key parameter is NETD (noise equivalent temperature difference). A typical value for a modern MIR camera operating in the 3 to 5 μm band with a quantum detector cryogenically cooled (for instance an InSb focal plane array) is 15 mK @300K. This means that the noise rms value is the difference between the radiances emitted in the whole band by two blackbodies at, respectively, 300.015 K and 300.000 K. We have preferred, however, to be conser‐ vative and use a larger NETD value, namely 50 mK @300K, which is reasonable for uncooled microbolometer arrays, cheaper and with lower power consumption. Noise expressed in radiance units is then 3.4 ⋅ 10−3W / m 2sr . It is assumed that this value is not changed by the use of the interference filter.

0

0.69 %

0.01

26.82 %

0.62 %

310

0.005

1.35 %

0.43 %

∆ L (W/m2·sr)

Tb = 250 K Tb = 280 K Noise Equivalent Radiance

10

20

30

40

50

60

70

80

90

Figure 4. Signal ) vs. C Δℒ ≡ | ℒas ∆L− ℒ Figure 4. (defined Signal as(defined ≡ |L|noleak −H Lleak |) vs. CH4 4 concentra‐ concentration for three different background temperatures; tion for three different background temperatures; the gas temperature is the temperature is Tg = 280K in all cases. The noise equivalent Tgas g = 280K in all cases. The noise equivalent radiance is shown for noleak

leak

radiance is shown for comparison. b

MDC ActiveaMDC b ( K )a Passive illumination. TFor tilted background, cosine factor appears in the F factor, and plotted 250 18.08the %ΔL value 0.81 % in Figure 4 is reduced accordingly. MDC values are therefore 270 in figure 78.22 % 0.75 % increased, as shown 5. 280

-

0.69 %

290

26.82 %

0.62 %

310

1.35 %

0.43 %

2.5

1

0.5

−40 0 20 40 60 In order−60to get a −20 moreAngle graphical feeling of the IR (º) scene, a simple simulation has been performed using the 5. Minimum detectable concentration 10 cm of Figure 5.of Cpath H assume Minimum detectable concentration for follows. a 10 for cm apath ofWe as a dataFigure the previous section, as CH4 as a function of the tilt angle of the background, for Tb = Tg4 = ◦ function of the tilt angle of the background, for T = T = T = 280K . For the Tabackground is divided into horizontal regions = 280K. For a 45 angle, MDC increases to b 1.36%, g aas compared a of to 0.69% for normal illumination. temperatures T = 250, 270, 280, 290 and 310 K,normal ordered 45 angle, MDC increases to 1.36%, as compared to 0.69% for b √ illumination. from top to bottom. Emissivity is, as previously, (standard deviation of 2 · 8 pixels), that varies from c =0.8. 100% Between at the image to c = 0%and at the image e = thiscenter background the camera, in thecloud camera we border. supposeTransmittances there is a gas at spectral Tg =280channel K with a vary, correspondingly, from τ in=the 0.52 to τ = 1. As gaussian concentration profile horizontal dimension

b

Noise Equivalent Radiance

Int J Adv Robot Syst, 2015, 12:0 | doi: 10.5772/60058 0.005

previously, ambient temperature is fixed at Ta =280 K.

0

70

80

90

Tb = 3

4.3. Scene simulation

100

0.01

60

Tb = 2

Tb = 2

b

50

Tb = 2

(Top) Without external

no thermal betweenand gas and and the gas is undetecta‐ gas and contrast background thebackground gas is undetectable. (Bottom) ble. (Bottom) With external illumination, can be detected at much With external illumination, gas can gas be detected at much smaller smaller concentrations background temperatures. temperatures. concentrations andand forfor allall background

Tb = 250 K

40

Tb = 2

= 280K there is comparison. (Top) Without external illumination. For T b contrast illumination. For T = 280K there is no thermal between

It must be pointed out that MDC values have been calcu‐ 0.02 T = 280 K T =background 310 K lated assuming 0.015 a normal incidence for the

30

100

Concentration (%) along a 10 cm optical path

0.03

20

(stan c = borde vary, previ

Tb = 310 K

0.015

0

0.025

∆ L (W/m2·sr)

Figur CH4 a Ta = 2 to 0.6

−0.005 0

0.035 375 nm spectral centred at the cm = λ 3.3 μm interference filter with 0

10

100

0.005

temperatures. In all cases, calculations assume that the optical path is 10 T g = 280K and the0.04MIR camera has a 50 mK @300K NETD and used an

−0.005 0

90

2 Table 1. Minimum Detectable Concentrations (MDC) using passive and active methods for CH4 at Tg = 280K and for several background temperatures. In all cases, calculations assume that 1.5 the optical path is 10 cm and the MIR camera has a 50 mK @300K NETD and used an interference filter with 375 nm spectral centered at the λ0 =3.3 µm CH4 absorption band.

Table 1. Minimum Detectable Concentrations (MDC) using passive and −0.005 20 30 40several 50 background 60 70 80 90 active methods for T b0( K ) 10at C H 4 and for(%) Concentration along a 10 cm optical path

6

80

0.01

0

absorption band.

70

L (W/m2·sr)

-

290

280

60

0.02

Minimum Detectable Concentration of CH4 (%)

0.75 %

0.015

∆ L (W/m2·sr)

0.81 %

78.22 %

0.02

50

0.025

Active MDC

18.08 %

270

40

0.03

Tb = 250 K

Noise Equivalent Radiance

30

0.04

Tb = 280 K

0.025

20

0.035

Results can be neatly summarized if the minimum detect‐ able concentration (MDC) is defined as the gas leak concentration that gives a SNR = 1 value. Values of MDC in active and passive modes for different background temperatures, obtained from Figure 4, are compared in table 1 (always0.04 for the 10 cm path).

250

10

Concentration (%) along a 10 cm optical path

ble. In comparison the graph at the bottom in the same figure shows that introducing illumination greatly im‐ proves gas detectability.

Passive MDC

0.015

−0.005 0

horizontal line. It is clear that signal-to-noise ratio (SNR) depends heavily on the background temperature, and, as expected, when there is no thermal contrast between gas and background (i.e., for T b = 280K ), the leak is undetecta‐

Tb = 310 K

0.02

0.005

concentration for three different background temperatures, without external illumination (in all cases l = 10 cm, T g = 280K , and T a = 280K ). Noise level is represented by the

0.03

Noise Equivalent Radiance

0.01

Figure 4 shows (at the top) the graph of signal vs. C H 4

C H4

Tb = 310 K

0.025

in ℒ between values calculated, respectively, with the leak gas concentration and zero gas concentration.

0.035

Tb = 250 K

0.035

Minimum Detectable Concentration of CH4 (%)

0.04

100

Concentration (%) along a 10 cm optical path

0

6

0.01

1.83

Concentration (%) 36.78 100 36.78

1.83

0.01

Int J Adv Robotic Sy, 2013, Vol. No, No:2013 Tb = 250 K

0 0.08

Figur mode Tb = 2 appre with a profile exten Horizo tempe

2

4.3 Scene simulation

90

100

0

1.5 to get a more graphical feeling of the IR scene, a In order simple simulation has been performed using the data of the previous1 section, as follows. We assume the background is divided into horizontal regions of temperatures T b = 250,

270, 280, 0.5 290 and 310 K, ordered from top to bottom. −60 −40 −20 0 20 40 60 Angleε(º) = 0.8. Between this back‐ Emissivity is, as previously, ground and the camera, we suppose there is a gas cloud at Figure 5. Minimum detectable concentration for a 10 cm path of T gCH=280 K with a Gaussian concentration profile in the 4 as a function of the tilt angle of the background, for Tb = Tg = T = 280K. For a 45◦ angle, MDC increases to 1.36%, as compared

a horizontal dimension (standard deviation of 2 ⋅ 8 pixels), to 0.69% for normal illumination. that varies from c = 100% at√the image centre to c = 0% at the (standard deviation of 2 · 8 pixels), varies from image border. Transmittances in the that camera spectral c = 100% at the image center to c = 0% at the image channel vary, correspondingly, from τ = 0.52 to τ = 1. As border. Transmittances in the camera spectral channel previously, ambient temperature is fixed vary, correspondingly, from τ = 0.52 at to Tτa =280 = 1.K. As

Radiance

0.01

Concentration (%) 36.78 100 36.78

1.83

1.83

0.01

5.1. Inspec

0 0.15

Tb = 250 K

The comp as follow

Tb = 270 K

0.1

Tb = 280 K

Imaging have expen hand, sensin detect detect impro consu roboti advan both featur design simpl the g Vanad SCD ( chose follow

L (W/m2·sr)

Minimum Detectable Concentration of CH4 (%)

2.5

Tb = 290 K

0.05

Tb = 310 K

10

20

30 # Pixel

40

50

60

Concentration (%) 0 0.16

0.01

1.83

36.78

100

36.78

1.83

0.01

0

0.14

previously, ambient temperature is fixed at Ta =280 K. Concentration (%) 36.78 100 36.78

100 0

0.01

1.83

1.83

0.01

0

Tb = 250 K

0.08

eak

) using for several assume that 0 mK @300K ectral

of the IR d using the We assume regions of K, ordered previously, the camera, 0 K with a l dimension

0.07 Tb = 270 K

Tb = 280 K 0.04

2

0.03

Tb = 290 K

0.02 0.01

Tb = 310 K

0 10

20

30 # Pixel

40

50

60

Concentration (%) 0 0.09

0.01

1.83

36.78

100

36.78

1.83

0.01

0

0.08 0.07 0.06 T = 250 K b

0.05

Tb = 280 K 0.04

Tb = 310 K

0.03 0.02 0.01 0

10

20

30

40

50

0.08

Tb = 250 K L (W/m ·sr)

0.05

0.1

0.06

0.06

L (W/m2·sr)

|) vs. CH4 eratures; the se equivalent hout external rast between e. (Bottom) much smaller

L (W/m2·sr)

0.12 90

60

Figure 6. (Top) Synthetic image of the radiance ℒoff (passive mode). Five Figure 6. (Top) Synthetic image of the radiance Lo f f (passive horizontal with background T b = 250, 270, 280, 290 mode). regions, Five horizontal regions,temperatures with background temperatures Tb = 270, top 280, and can 310beK appreciated. (from top toAbottom) be and 310250, K (from to 290 bottom) C H 4 gascan cloud, appreciated. A CH4 gas cloud, uniform in the vertical direction, , and with l = 10cm uniform in the vertical direction, with a thickness with a thickness l = 10cm, and with a gaussian concentrationa Gaussian concentration profile in the horizontal direction (minimum profile in the horizontal direction (minimum transmitance τ = 0.52) extends between background andbackground the camera. (Bottom) transmittance extends between the and the camera. τ = 0.52)the Horizontal radiance profiles for for three different (Bottom) Horizontal radiance profiles three differentbackground background temperatures. The continuous lines are the values without noise. temperatures. The continuous lines are the values without noise.

With these data, synthetic images have been constructed for ℒoff (Figure 6) and ℒon (Figure 7). Referring first to

www.intechopen.com as a dark cloud Figure 6, the gas leak (at T g = 280K ) is seen

against the hot background (T b = 310K ), and as a clear region

against the coldest background (T b = 250K ), although in this

case, the positive contrast is much smaller than the negative contrast in the hot background. When T b = T g , as in Profile

Tb = 280 K

0.04

Tb = 310 K 0.02

10

20

30

40

50

Te

60

Figure 7. (Top) Synthetic image of the radiance ℒon (active mode), for the Figure 7. (Top) Synthetic image of the radiance Lon (active same conditions of same Figure 6. (Bottom) Horizontal radiance profiles for three mode), for the confitions of Figure 6. (Bottom) Horizontal different background Thebackground continuous temperatures. lines are the values radiance profiles fortemperatures. three different The continuous without noise. lines are the values without noise.

S

Co

With these data, synthetic images have been constructed 2,for theLleak is invisible even in the ideal7). case without first noise. 6) and Lon (Figure Referring to o f f (Figure Figure 6, thehand, gas leak (at T seen (Figure as a dark On the other profiles for the280K) activeismode 7) g = cloudthat against background (Tb = for 310K), andback‐ as a show therethe is ahot much better contrast all the clear region against the coldest background (T = 250K), b grounds and it is negative in all cases. although in this case, the positive contrast is much smaller than the negative contrast in the hot background. When 5. System implementation Tb = Tg , as in Profile 2, the leak is invisible even in the ideal case without noise. On the other hand, an profiles for In the previous section an approach to design imaging the active mode (Figure 7) show that there is a much better sensor able to detect pollutant gases remotely with a good contrast for all the backgrounds and it is negative in all SNR ratio, even when the gas has no thermal contrast with cases. the background, has been proposed. The approach is based on the use of an IR camera equipped with an interference 5. System implementation filter and an IR source to illuminate the scene (i.e., operating the previous section it has been proposed an approach inInactive mode). In this section, a specific implementation to design an imaging sensor able to detect pollutant gases of this design is worked out, in order to demonstrate its remotely with a good SNR ratio even when the gas has viability as a robotic sensor system. We will describe it in no thermal contrast with the background. The approach two parts: first, the inspection system (i.e., all the required is based on the use of an IR camera equipped with an elements to perform thean remote sensing tasks) andthe then the interference filter and IR source to illuminate scene robotic platform (including the mobile platform, the robot (i.e., operate in active mode). In this section, a specific software and the robot system).out, in order to implementation of thisexploration design is worked demonstrate its viability as a robotic sensor system. We will describesystem it in two parts: first, the inspection system 5.1 Inspection (i.e., all the required elements to perform the remote The components of the inspection system are summarized sensing tasks) and then the robotic platform (including mobile platform, the robot software and the robot asthe follows: exploration system). Although quantum detectors have the best NETD figures, they are relatively expensive and need cryogenic cooling.

Ramon Barber , Miguel A. Rodriguez-Conejo, Juan Melendez and Santiago Garrido: Design of an Infrared Imaging System for Robotic Inspection of Gas Leaks in Industrial Environments www.intechopen.com

Table 2. S SCD core. F=1.

Optical s and t the fo Radio alarm accou spectr protot 35 mm use in

Illumina menti to enh is fix determ covere consu applic mana infrar chose 3. In t 7

Design of an infrared imaging system for ro

On the other hand, due to recent developments in micro‐ bolometer sensing technology, now there are uncooled detectors with good enough performance for gas detection. Microbolometer technology represents an improvement in terms of cost, weight and power consumption, which are critical parameters for a robotic system implementation. They have also the advantage of covering a wide spectral range, including both medium and thermal infrared regions. This feature provides high versatility to the proposed design, since several compounds can be detected simply by selecting interference filters matched to the gas spectral signature. For these reasons, a Vanadium Oxide (VOx) microbolometer core from SCD (SemiConductor Devices) manufacturer has been chosen. Specifications of this detector are shown in the following table 1.

the angle dependence of the measurements, the addition of a pan/tilt system will be considered. forward horizontally. Taking into account the angle be considered to add a pan/tilt system. Voltage 12.0 Volts

Technical specifications formeasurements, IR-18 infrared source dependence of the it will

Temperature Technical

specifications1150 for Celsius IR-18 infrared source

Current

Voltage Temperature Power Current Emissivity Power Emissivity Radiating area Radiating area Collimation Collimation Lifetime Lifetime

12.0 Volts 1.5 Amperes 1150 Celsius 1.5 Amperes 0.75 18 Watts 0.75 1.5 mm dia. x 4.1 mm length 1.5 mm dia. x 4.1 mm length Parabolic reflector Parabolic reflector 5000 + hours 5000 + hours 18 Watts

Table Summary of the moremore relevant parameters for IR-18for source Table3.3. Summary of the relevant parameters IR-18 source.

VOx

Spectral Wide

3 - 5 μm

Array

640x480 pxeles

Pixel Pitch

25 μm

NETD (F=1)

50 mK @ 300K

Communication

Camera Link

Digital res.

14 bits

Frame rate

60 f ps

Power

lower than 5 W

Figure 9. inspection s

5.2.2. Robo

Table 2. Summary of the main parameters for the selected SCD core. NETD value corresponds to optics with an f-number F=1

Depending on both the working distances and the size of the gas leakages to be detected, the focal length of the system will be established. Radiometric constraints such as detection limit, false alarm probability, and so on might be taken into account to define other key parameters as F number, spectral transmittance or coatings. In this initial prototyping stage, a simple optical system based on a 35mm focal distance and 1.1 F number is chosen for its use in short distance environments.

Figure8. 8. Two views of the embedded inspection Figure Two views of the embedded inspection systemsystem

5.2.Mobile Mobile platform platform 5.2 The mobile mobile platform platform involves The involves the the hardware hardwaredesign designand andthe the software that controls it. Both items are described software that controls it. Both items are described in in the the followingssubsections. subsections. following 5.2.1. Robot description 5.2.1 Robot description The mobile platform, where the gas sensing system is The mobile whereisthe gas sensing is going going to beplatform, implemented, a robot calledsystem TurtleBot [18] to be implemented, is a robot [18] (Figure (Figure 9). It is a round shapecalled robotTurtleBot of 36cm of diameter 9). is a round shape robot 36cmup of of diameter and 10 cm andIt 10cm height, which is of made a Roomba base, height, which is made up of a Roomba base, motion a Kinect motion sensor attached to it anda aKinect laptop that takes care of alltothe communications the of devices sensor attached it and a laptop that among takes care all the and where the navigation software and the gas analysis communications among the devices and where the navi‐ algorithms are installed. gation software and the gas analysis algorithms are installed. TurtleBot is a low-cost, personal robot kit with open-source software for indoor navigation and it is used as a robotic TurtleBot is a low-cost, personal robot kit with open-source experimental platform to test robotics prototypes in our software for indoor navigation and is used as a robotic labs. Its weight is about 5kg and it allows a load capacity experimental platform to test prototypes our over 5kg, enough to carry therobotics inspections devices.inThe robot can move around and detect the environment while it is moving using a Kinect [19] used as a range finder

As has been previously mentioned, an active infrared source will be required to enhance the detection limit. Once this parameter is fixed, radiometric specifications will be mainly determined by the working distance and the area covered by the optical field of view. Since power consump‐ tion is an important constraint at this application, radiative efficiency must be properly managed through the use of reflectors. The IR-18 infrared source from HawkEye Technologies has been chosen, whose main parameters are described in table 1. In the present prototype the source is fixed, pointing forward horizontally. Taking into account 8

The batte inside th hours of of the ro to locate high-perfo and the ins laptop wo (Robotic O

It is important to previously define which application will Communication protocols It is important to previously hostdefine the processing of the acquired data. For tasks that which application will host the processing involve only detection, this For can tasks be carried on board, of the acquired data. that out involve only from an embedded system, but to implement the proposed detection, this can be carried out on-board, from an quantization in an robot might be embeddedalgorithms system; but to operative implement the itproposed quantization in data an operative robot it might more reasonablealgorithms to transmit to another point with be more reasonable to transmit datapresent to another point higher processing capability. In the prototype with however, higher processing capability. In the present version, the simplest solution (i.e., an onboard prototype version, however, the simplest solution (i.e., laptop computer) has been used. an onboard laptop computer) has been used. Components described are shown in Figure 8. Components described are shown in Figure 8.

Technical specifications for BIRD-640 SCD core Detector

telemeter create a m the detect 0.65m/s, b algorithms

Int J Adv Robot Syst, 2015, 12:0 | doi: 10.5772/60058

8

Int J Adv Robotic Sy, 2013, Vol. No, No:2013

The contr device is an open-s to develop a single f libraries a of creating way. ROS of an oper provides such as ha implemen exchange ROS is bas takes plac multiplex planners, a

ROS has tw

• The pa above by the called simulta percep

• The lo diagno are tra

telemeter for the navigation system of the robot. ItROS can labs. Its weight is about 5 kg and it allows a load capacity has two basic components: create a over map ofenough the area tothebe inspected and it can locate 5 kg, to carry inspection devices. The robot • The part of the OS, ROS, as has been described above, can move around and detect the environment while it is speed is the detected leaks over the map. Its maximum and ROS-pkg, a suite of packages provided by the [19] used as a range finder telemeter 0.65m/s,moving but itusing cana Kinect be reduced according to the inspection contribution of the users (settled in groups called stacks) for the navigation system of the robot. It can create a map that implement functionalities such as simultaneous algorithms. of the area to be inspected and can locate the detected leaks over the map. Its maximum speed is 0.65 m/s, but it can be batteries of this autonomous robot are reduced according to the inspection algorithms.

mapping and localization, planning, perception, simu‐ lation, located etc.

The • The2 low-level tasks components for sensor access, inside the base, featuring an autonomy up to The batteries of this autonomous robot are located inside diagnosis reporting, power management, etc., that are hours of continuous work. Finally, the hardware the base, featuring an autonomy up to two hours of transparently handled by ROS. This allows the develop‐ of the continuous robot iswork. completed a ofmodular Finally, the with hardware the robot isstructure ment of controllers and higher-level tasks for a variety completed with a modulardevices structure to and locate the to locate the inspection theinspection ultraportable, of platforms. devices and the ultraportable, high-performance laptop in high-performance laptop in which the mobile platform TurtleBot can be managed by ROS through its robotic which the mobile platform and the inspection device are software development environment. It includestasks an SDK and the connected inspection device are connected for its control. The development of controllers and higher-level for forproblem. This app for its control. The laptop works under the Linux a variety of platforms. particle carries a the TurtleBot, a development environment for the desktop operative under system using ROSoperative (Robotic Operative System laptop works Linux system using ROS using adaptive and libraries for visualization, planning, perception, as a framework. TurtleBot can be managed by ROS through its robotic particles in a Rao (Robotic[20]) Operative System [20]) as a framework. control and error handling. The software tools are com‐ software development environment.

It includes an

grid maps.

T

SDK the other TurtleBot, a development environment for asdistribution takin pleted by for some developments using C++, OpenCV desktop image and libraries for visualization, planning, the the available processing library and softwareof the robot but perception, control and error handling. The software tools drastically decrea implementation the developments inspection devices and forin the prediction are completedto bycontrol some other using C++, the inspection OpenCV as algorithms. the available image processing library and the to selectively ca software implementation to control the inspection devices

reduces the probl

Figure the software components involved in the and10forshows the inspection algorithms. robotic control system under the ROS operating system. In order to sens Figure 10 shows the software components involved in the robotic control system under the ROS operating system.

scanner is neede as long-range la alternative to tra horizontal slice o nearest distance (

Using this techn unknown environ task, navigating, placing itself in been completed information abou the map.

The developed a software module implementes as R Main components of the software architecture are: for the with an inspectio Main components of the software architecture are: Planner and Nav • ROS as operating system and and as of high levellevelKinect or gas dete • ROS as operating system asmanager manager of high task and of the control of the robot. thought a roslunch task and of the control of the robot. • A robot task module, that coordinates the path planner, The navigation m the SLAM andcoordinates completes the the map path • A robot task navigation module, tasks which a path using Fast information with the location of the leaks. planner, the SLAM navigation tasks and completes thegiven previously. • Robot controlwith module, that manages sensorial been defined as map information the location of the leaks. information for navigation, actuators, and the I/O build a map of the with the inspection device.sensorial infor‐the odometry of th • Robotcommunication control module, which manages mation for navigation, actuators, and the I/O communi‐stored map to loc All components are encapsulate in ROS nodes and they kinect and from t cation with the as inspection device. exchange. use ROS topics way of information through ROS topi Figure 10. Software architecture of the robotic system

Figure 10. Software architecture of the robotic system

Figure 9. Turtlebot robot used as mobile platform for the inspection system Figure 9. Turtlebot robot used as mobile platform inspection system.

5.2.2 Robot software architecture

The control of the mobile base and of the inspection device [20]. ROS is an opensource software widespread use which allows the devel‐ The control the applications mobile base and of theininspection opmentof robotic and their integration a single framework. It includes a collection of tools, libraries device is integrated in ROS architecture [20]. ROS is and conventions that seek widespread to simplify the tasks of creating an open-source software use which allows software for Advanced Robotics in a robust way. ROS, to develop applications their integration in ratherrobotic than a framework, has theand functionality of an For the development of this work,inROS Groovy All components are encapsulated ROS nodesunder and useOnce the map is g 12.04 LTS hasof been used. C++ language has been map, taking into a a singleoperating framework. It includescluster. a collection system on a heterogeneous ROS providesof tools, ROS Ubuntu topics as a means information exchange. used as programming language. in the SLAM proc the standard services of an operating system such as libraries and conventions that seek to simplify the For tasks the development of this work, ROS Groovy underto the robot. hardware abstraction, low level device control, implement‐ 5.3. Robot of creating software for Advanced Robotics in a robust Ubuntu 12.04exploration LTS wassystem used. C++ language was used as theThe proposed e ing functionality of common use, messages exchange operation modes: The exploration system is based on the technique called programming language. way. ROS, rather than apackage framework, hasROS theis functionality between processes and maintenance. based SLAM (Simultaneous Localization And Mapping ) [21]. • Inspection tas on graph architecture, processing takes place in of an operating system where on a the heterogeneous cluster. ROS SLAM allows to locate and build the map of environment 5.3 Robot exploration system the case of mo simultaneously when the robot is moving. the nodes that can receive, send and multiplex messages provides the standard services of an operating system of establishing from the sensors, control, states and planners, among has severalsystem packagesisto based carry outon theaSLAM technique The ROS exploration technique called to locate the such as others. hardware abstraction, low level devices control, using vision. For the development of this work the SLAM (Simultaneous Localization And Mapping) [21]. considered. In

5.2.2. Robot software is integrated in architecture ROS architecture

gmapping package has been used, which provides the implementing functionality of common use, messages basic functionality for the exploration algorithm. Several Barber , Miguel A. Rodriguez-Conejo, Melendez modifications have beenJuan carried out onand thisSantiago package Garrido: in exchange between processes and packageRamon maintenance. Design of an Infrared Imaging System order for Robotic Inspection of Gas Leaks in Industrial Environments to integrate the information related to the detected • ROS is based on graphs architecture, where the processing leaks. takes place in the nodes that can receive, send and The implementation of gmapping is based on the Rao-Blackwellized particle filters to solve the SLAM multiplex messages from the sensors, control, states,

robot follows the robot on th 9 information to

Exploration ta where the robo an unknown s the robot is ab

SLAM enables the location and building of the map of environment simultaneously when the robot is moving.

background was a Raymax 1330 Wat area at 350 ◦ C and this relatively high source has been r described.

ROS has several packages to carry out the SLAM technique using vision. For the development of this work the gmapping package was used, which provides the basic functionality for the exploration algorithm. Several modi‐ fications have been carried out on this package in order to integrate the information related to the detected leaks. The implementation of gmapping is based on the RaoBlackwellized particle filters to solve the SLAM problem. This approach uses a particle filter in which each particle carries an individual map of the environment, using adaptive techniques to reduce the number of particles in a Rao-Blackwellized particle filter for learning grid maps. The approach computes an accurate distribution, taking into account not only the movement of the robot but also the most recent observation. This drastically decreases uncertainty about the robot’s pose in the prediction step of the filter, applying an approach to selectively carry out resampling operations which reduces the problem of particle depletion. In order to sense the environment, a long-range laser scanner is needed. In this work the kinect is used as a longrange laser sensor, providing an inexpensive alternative to traditional laser scanners, by cutting a horizontal slice out of the Kinect image and using the nearest distance (closest depth) in each column. Using this technique, the robot can locate itself in an unknown environment. The robot starts its exploration task, navigating, building the environment map and placing itself in it simultaneously. The algorithm has been completed so that, once the leak is detected, the information about its position and parameters is added to the map. The developed algorithm is shown in Figure 11. All software modules involved in the robotic systems are implemented as ROS nodes. The navigation task starts with an inspection order as an ROS topic, which starts the Planner and Navigation modules. Sensorial systems such as Kinect or gas detector modules are launched by the system thought a roslunch command at the beginning. The navigation module starts to move the robot following a path using the Fast Marching Square technique [22] if it is given previously. If not, a contour following policy has been defined as the default movement. The robot starts to build a map of the environment with the Kinect sensor and the odometry, using the information of the stored map to locate its position on it. Sensorial information from the kinect and from the odometry are also shared among modules through ROS topics. Once the map is generated detected leaks are added to the map, taking into account the position of the robot obtained in the SLAM process and the estimated position of the leak to the robot. 10

Int J Adv Robot Syst, 2015, 12:0 | doi: 10.5772/60058

11. Implemented Inspection algorithm the SLAM Inspection algorithm basedbased on theon SLAM technique FigureFigure 11. Implemented technique

skills such exploration as contour following in order to build the two The proposed system contemplates map, locate on it and add the leak position and the operation modes: information in the map in a simultaneous way.

• Inspection task: If the map is a priori known, as is the 6. Overall platform test 13. Two sm case for most of the industrial plants, the possibility ofFigure measurement config The test environment for the full prototype consists of establishing the previous inspection routes, limited to a room in which several gas pipes run parallel to the locate the robot and the leak within the map, is consid‐The IR camera, w walls. Due to the extreme flammability of CH4 , and ered. In this case the navigation of the robotat 2 m from the the difficulties to handle and store itsystem in the laboratory, 4300 nm with a 2 CO was used in these tests. The measurement principle 2 the path, avoiding obstacles, relocating follows the robotused. Gas pressu is exactly the same as previously described for CH4 , on the map and adding the leak position and informationatmospheric, so t although additional difficulties could be expected in this leaks and no therm to case the map. because of the presence of carbon dioxide in the atmosphere. However, no problem related to atmospheric

leaks were undete

• Exploration task: map is not known, as is the casethe figure, provid CO2 was found in If thethe tests. where robotthe works emergency situation In thethe scenario, robot in hasan a previous map of the room.or in7. Conclusions anThe unknown scenario, without information, robot moves its location on theprevious map following the wall the in which the to pipes areto placed. Whenusing a leak is detected, theskillsIn this work a d robot is able start navigate predefined has been propose robot places in front of the leak and takes the measurement such as contour following in order to build the map,implementing act of the gas concentration. Figure 12 shows the robot in the locate on it and add the using leak position the to information platform, provide operating environment the slam and module locate for its use in inspe gas leak in the way. map. in the therobot mapand in athe simultaneous

The main advant of an inexpensiv remote sensing im

6. Overall platform test

The test environment for the full prototype consists of aRegarding the ga is mainly based room in which several gas pipes run parallel to the walls. array, which supp Due to the extreme flammability of C H 4, and the difficultiessize, weight and

in handling and storing it in the laboratory, CO2 was usedwith standard IR

including both ex

in these tests. The measurement principle is exactly thebeen properly ch same as previously described for C H 4, although additionalbeen evaluated th

leakage simulatio

difficulties could be expected in this case because of theshows the conve presence of carbon dioxide in the atmosphere. However,where minimum d Figure 12. Robot moving in the environment detecting a gas CO2forwas no problem related to atmospheric found in thestrongly reduced. leak

tests.

Experimental resu the feasibility of t Figure 13 shows a gas duct where two small leaks have In thebeen scenario, theatrobot has anut previous map the room.common gas sens found: one the faucet’s and a other at aofpinhole inspection along at themoves bottom.its The imageon is the the map resultfollowing of subtracting The robot location the wall or metallurgical p a background image (Lnoleak , obtained previously to gas on in which the pipes aresignal placed. When detected, circulation) to the image (Lleaka). leak Theiswall at the thethe huge versatil on

robot places itself in front of the leak and takes the meas‐ urement of the gas concentration. Figure 12 shows the robot 10 Int J Adv Robotic Sy, 2013, Vol. No, No:2013 in the operating environment using the slam module to locate the robot and the gas leak in the map. Figure 13 shows a gas duct where two small leaks have been found: one at the faucet’s nut and another at a pinhole at the bottom. The image is the result of subtracting a back‐ ground image ( L onnoleak , obtained previously to gas circula‐

In the scenario, the robot has a previous map of the room. The robot moves its location on the map following the wall in which the pipes are placed. When a leak is detected, the robot places in front of the leak and takes the measurement of the gas concentration. Figure 12 shows the robot in the operating environment using the slam module to locate the robot and the gas leak in the map.

Figure 12. 12. Robot moving in thein environment for detecting a gas leaka gas Figure Robot moving the environment for detecting leak leak tion) to the L on ). The wall the background Figure 13 signal showsimage a gas (duct where two at small leaks have been one at the faucet’s nut and awas other at a pinhole was atfound: room temperature. The IR source a Raymax 1330 2 subtracting at the bottom. The image is25.5 the× 15.5 result of cm 350 Watlow radiant heater, with a area at background was at roomnoleak temperature. The IR source was a C a background image (Lon , obtained to gas and 110 watt Duepreviously this relatively Raymax 1330power Watlowconsumption. radiant heater,leak with ato 25.5 × 15.5 cm2 circulation) to the signal image (L ). The wall at to the ◦ on area at 350 inCthe andfinal 110 watt power consumption. Due high power, prototype design the source was this relatively high power, in the final prototype design the replaced by the IR-18 source previously described.

source has been replaced by the IR-18 source previously described.

10 Int J Adv Robotic Sy, 2013, Vol. No, No:2013

7. Conclusions In this work a design for a gas detection mobile robot has been proposed, including an infrared imaging sensor, implementing active detection techniques, and a robotic platform, provided with control and navigation features for its use in inspection tasks at industrial environments. The main advantage of the proposed system is the use Regarding the gas sensing strategy, the described design is of an inexpensive robotic platform that makes possible mainly based on a spectrally tuned microbolometer array, remote sensing imaging of gas leaks. which supposes an improvement in terms of cost, size, Regarding the gas sensing strategy, the design weight and power consumption indescribed comparison with is mainly based on a spectrally tuned microbolometer standard IR cameras. The full detection system, including array, which supposes an improvement in terms of cost, both weight externaland IR source imaging in sensor, has been size, power and consumption comparison properly characterized and The its performances been with standard IR cameras. full detection have system, evaluatedboth through radiometric A set of leakage including external IR sourcemodelling. and imaging sensor, has been properlybased characterized and(Cits havethe simulations on methane H 4performances ) detection shows been evaluatedofthrough radiometric modeling.where A set mini‐ of convenience the proposed methodology, leakage simulations based on methane (CH4 ) detection mum detectable concentration levels have been strongly shows the convenience of the proposed methodology, reduced. where minimum detectable concentration levels have been strongly reduced. Experimental results obtained for CO leaks underscore the 2

Experimental obtainedstrategy, for CO2 to leaks underscore feasibility of results the proposed be used in most the feasibility the proposed strategy,natural to be used most common gasofsensing applications: gasinpipelines common gasalong sensing natural gas inspection wideapplications: networks, monitoring of pipelines chemical or inspection along wide networks, monitoring of chemical metallurgical plants and so on. Remarkable also is the huge or metallurgical plants and so on. It is also remarkable versatility of this design fordesign the detection a large variety the huge versatility of this for the of detection of a of compounds (as long as they have spectral signature in the IR range) through simple modifications, as a conse‐ quence of the wide spectral bandwidth of the microbolom‐ www.intechopen.com eter technology.

Finally, an inspection algorithm for the mobile platform has been proposed, based on mapping and localization, that includes several possibilities for path planning and mapping the detected gas leaks. As future work, the system will be provided for a pan/tilt platform in order to enlarge the range of inspection. This requires the recalculation of detection parameters, taking into account the inspection angle.

sed on the SLAM

er to build the sition and the s way.

8. Acknowledgements Figure 13.small Two small detected at aduct CO2with ductan with an active Figure 13. Two leaks leaks detected at a CO active measure‐ 2

pe consists of parallel to the of CH4 , and he laboratory, ment principle bed for CH4 , xpected in this dioxide in the o atmospheric

measurement configuration.

ment configuration

The IR camera, with a focal distance of 50 mm, was placed

The IR camera, with a focal distance of 50 mm, was placed at 2 m from the duct; an interference filter centered at at 24300 m from the duct; an interference centred at 4300 nm with a 200 nm full width at filter half maximum was nmused. with aGas 200pressure nm fullinwidth at half was used. the duct was maximum only slightly above that no was adiabatic the Gasatmospheric, pressure insothe duct only cooling slightlyoccurs aboveatatmos‐ leaks and no thermal contrast is present. In this conditions, pheric, so that no adiabatic cooling occurs at the leaks and leaks were undetectable in passive mode, but, as shown in no thermal contrast is present. In this conditions, leaks were the figure, provided a very strong contrast in active mode. undetectable in passive mode, but, as shown in the figure, p of the room. provided a very strong contrast in active mode. 7. Conclusions

owing the wall In this work a design for a gas detection mobile robot s detected, the has been proposed, including an infrared imaging sensor, measurement 7. Conclusions implementing active detection techniques, and a robotic he robot in the In this workprovided a designwith for acontrol gas detection mobile features robot has platform, and navigation dule to locate been including infrared environments. imaging sensor, for proposed, its use in inspection tasksan at industrial

implementing active detection techniques and a robotic The main advantage of the proposed system is the use platform, provided with control andthat navigation features of an inexpensive robotic platform makes possible for remote its use sensing in inspection inleaks. industrial environments. imagingtasks of gas

the gas sensing strategy, thesystem described design TheRegarding main advantage of the proposed is the use of mainly based on a spectrally tuned microbolometer an is inexpensive robotic platform that makes possible the array, which supposes an improvement in terms of cost, remote of gas leaks. size, sensing weight imaging and power consumption in comparison

detecting a gas

The authors would like to thank the RoboCity2030-II project (S2009/DPI-1559), funded by Programas de Activi‐ dades I+D en la Comunidad de Madrid and co-funded by Structural Funds of the EU. 9. References [1] Jose Maria Martnez-Otzeta, Aitor Ibarguren, Ander Ansuategi, and Loreto Susperregi. Laser based people following behaviour in an emergency environment. In Intelligent Robotics and Applications, pp. 33–42. Springer, 2009. [2] Zheng Li and Yi Ruan. Autonomous inspection robot for power transmission lines maintenance while operating on the overhead ground wires. Intl. J. of Advanced Robotic Systems, 7(4): 111–116, 2010. [3] Klaus-Peter Möllmann and Michael Vollmer. Industrial Application: Detection of Gases. Wiley-VCH Verlag GmbH & Co. KGaA, New York, 2010. [4] Gas Detection Handbook: Key concepts and reference material for permanently installed gas monitoring systems. MSA, 4 edition, 2006.

with standard IR cameras. The full detection system, including both external IR source and imaging sensor, has been properly characterized and its performancesRamon have Barber , Miguel A. Rodriguez-Conejo, Juan Melendez and Santiago Garrido: of an Infrared Imaging System for Robotic Inspection of Gas Leaks in Industrial Environments been evaluated through radiometricDesign modeling. A set of leakage simulations based on methane (CH4 ) detection shows the convenience of the proposed methodology, where minimum detectable concentration levels have been strongly reduced.

11

[5] Marco Trincavelli. Gas discrimination for mobile robots. KI-Künstliche Intelligenz, 25(4): 351–354, 2011. [6] Syeda Erfana Zohora, A.M. Khan, and Nisar Hundewale. Chemical sensors employed in elec‐ tronic noses: A review. In Advances in Computing and Information Technology, pp. 177–184. Springer, 2013. [7] Amy Loutfi, Silvia Coradeschi, Lars Karlsson, and Mathias Broxvall. Putting olfaction into action: Using an electronic nose on a multi-sensing mobile robot. In Intelligent Robots and Systems, 2004.(IROS 2004). Proceedings. 2004 IEEE/RSJ International Conference on, 1: 337–342, 2004. [8] Mickey B Frish, Richard T Wainner, Joy StaffordEvans, Byron D Green, Mark G Allen, S Chancey, J Rutherford, G Midgley, and P Wehnert. Standoff sensing of natural gas leaks: evolution of the remote methane leak detector (rmld). In Photonic Applica‐ tions Systems Technologies Conference, p. JThF3. Optical Society of America, 2005. [9] Remote methane leak detector. http:// www.gmiuk.com/product/remote-methane-leakdetector-rmld/, july 2014. [10] http://www.flir.com/cs/emea/es/, july 2014. [11] Thomas Barz, Gero Bonow, Jens Hegenberg, Karim Habib, Liubov Cramar, Jochen Welle, Dirk Schulz, Andreas Kroll, and Ludger Schmidt. Unmanned inspection of large industrial environments. In Future Security, pp. 216–219. Springer, 2012. [12] S Soldan, G Bonow, and A Kroll. Robogasinspectora mobile robotic system for remote leak sensing and localization in large industrial environments: overview and first results. In IFAC WorkshopAutomatic Control in Offshore Oil and Gas Production, pp. 33–38, 2012. [13] Colin N Banwell and Elaine M McCash. Fundamen‐ tals of molecular spectroscopy. McGraw-Hill London, 4 edition, 1994.

12

Int J Adv Robot Syst, 2015, 12:0 | doi: 10.5772/60058

[14] John Houghton. The physics of atmospheres. Cam‐ bridge University Press, 2002. [15] Laurence S Rothman, Iouli E Gordon, Alain Barbe, D Chris Benner, Peter F Bernath, Manfred Birk, Vincent Boudon, Linda R Brown, Alain Campar‐ gue, J-P Champion, et al. The< i> hitran 2008 molecular spectroscopic database. Journal of Quan‐ titative Spectroscopy and Radiative Transfer, 110(9): 533–572, 2009. [16] Hitran on the web. http://hitran.iao.ru/, july 2014. [17] Infrared source. http://www.hawkeyetechnolo‐ gies.com/IR18.pdf, july 2014. [18] [18] Brian Gerkey and Ken Conley. Robot developer kits [ros topics]. Robotics & Automation Magazine, IEEE, 18(3): 16–16, 2011. [19] Oscar Martinez Mozos, Hitoshi Mizutani, Ryo Kurazume, and Tsutomu Hasegawa. Categoriza‐ tion of indoor places using the kinect sensor. Sensors, 12(5): 6695–6711, 2012. [20] Morgan Quigley, Ken Conley, Brian Gerkey, Josh Faust, Tully Foote, Jeremy Leibs, Rob Wheeler, and Andrew Y Ng. Ros: an open-source robot operating system. In ICRA workshop on open source software, 3(5), 2009. [21] MWM Gamini Dissanayake, Paul Newman, Steve Clark, Hugh F Durrant-Whyte, and Michael Csorba. A solution to the simultaneous localization and map building (slam) problem. Robotics and Automation, IEEE Transactions on, 17(3): 229–241, 2001. [22] Santiago Garrido, Luis Moreno, Javier V. Gomez and Pedro U. Lima. General Path Planning Meth‐ odology for Leader-Followers based Robot Forma‐ tions. Intl. J. of Advanced Robotic Systems, 10(64) 1– 10, 2013.