REAL-TIME MONITORING of WELD POOL during GTAW using INFRA-REDTHERMOGRAPHY and analysis of Infra-Red thermal images

REAL-TIME MONITORING of WELD POOL during

GTAW using INFRA-RED THERMOGRAPHY
and analysis of Infra-Red thermal images
M. Vasudevan, N. Chandrasekhar, V. Maduraimuthu,
A.K. Bhaduri and B. Raj

Real-time monitoring of the weld pool using infra-red (IR) thermography during gas tungsten arc (GTA) weld-
ABSTRACT

ing is gaining importance due to the requirements for on-line monitoring and control of the welding process.
To facilitate real-time monitoring of the weld pool, a computer-controlled GTA welding machine with sensing
of the weld pool using IR camera has been developed. The IR camera, mounted on the torch assembly, moni-
tors the molten pool and the surface temperature distribution surrounding the weld pool during GTA welding.
Temperature profiles were measured on the plates using thermocouples in combination with IR thermography
to determine the emissivity of the plate surface. GTA welding was carried out on 3 mm-thick 316LN stainless
steel (SS) plates under different welding conditions. IR thermal images were acquired on-line and analysed. A
linear relationship was obtained between the thermal bead width, determined by line-scan analysis technique,
and the actual bead width, measured by cross-sectional optical microscopy. The computed macroscopic
temperature gradient and the actual of weld bead depth of penetration showed an inverse relationship. Full-
frame analysis was carried out to estimate the surface temperature distribution for square-butt weld joints. For
316LN SS weld joints, IR thermal signatures were acquired for various weld defects, such as lack of fusion,
lack of penetration and tungsten inclusions, for use as reference signatures for on-line monitoring during GTA
welding.

IIW-Thesaurus keywords: Defects; Imaging; Infrared; Lack of fusion; Penetration defects; Thermography; Visual
inspection; Welded joints. 83

1 Introduction wire-feeder up and down movement, arc voltage controller

up and down movement, gas purge, homing the machine
(i.e. to keep the machine at zero coordinates on all axes
In recent years there has been considerable advance-
by mechanical reference so that the machine always
ment in automation of welding processes using intelligent
starts from zero coordinates), cycle start and controls the
methodologies to facilitate on-line process monitoring and
torch assembly by setting the parameter into Galil motion
control, which can also be used for remote repair welding
control module and PLC. Visual basic software has been
applications. Intelligent automation of gas tungsten arc
developed and interfaced with the PLC and Galil multi-
(GTA) welding involves development of computer-based
axis motion control, by serial and Ethernet communica-
monitoring and control software with Human Machine
tion. In this way, a computer-operated fully automated GTA
Interface (HMI) that enables powerful, flexible, and user-
welding machine with HMI has been developed and inte-
friendly operator interface by bringing together all displays
grated to facilitate the entire operation and control of the
and functions needed for on-line monitoring and control
machine by a computer.
of the welding machine. The HMI system integrates sys-
tem operation at a single level and it gives the operator
good visualization. Also, a computer-based multi-channel data-acquisition
system, employing a RS485 network, has also been
The HMI monitors and controls the servo and step- developed for validating temperature measurements at
per motors and process parameters of the GTA welding various locations along the entire length on the plate sur-
machine as per standard program, as well as experimen- face close to the weld using C-type thermocouples. For
tal requirements with the help of interfaced Galil motion this purpose, Visual Basic software has been developed to
controller and Programmable Logic Controller (PLC). The acquire and record temperatures measured by tempera-
HMI tracks and records the process parameters of weld- ture indicators as a function of time [1].
ing, such as voltage, current, jog speed, current axis value,
of the welding machine. The HMI sends commands to the In order to monitor and control the welding operations for
Galil and PLC for controlling the jogging, point-to-point producing quality welds in non-accessible areas, sensors
positioning, vector positioning, multiple move sequence, capable of monitoring and controlling the welding process

Doc. IIW-2044, recommended for publication by Commission V “Quality Control and Quality Assurance of Welded Products.”

08 2011 Vol. 55 WELDING IN THE WORLD

N° 07 Peer-reviewed Section
 REAL-TIME MONITORING of WELD POOL during GTAW using INFRA-REDTHERMOGRAPHY and analysis of Infra-Red thermal images

is required to be incorporated into the robot. The signifi- 4. Estimate the surface temperature distribution for
cant advances in cameras in recent years have made square-butt weld joints from IR thermal images.
possible the development of vision sensors for real-time 5. Acquire and record IR thermal signature images for
monitoring and control of arc welding processes. Vision weld joints with and without weld defects.
sensors have been used to monitor the seam position,

2 Experimental
weld pool width and depth of penetration in arc welding
processes [2-4]. Among the vision sensors, infrared (IR)
sensing is a natural choice for weld process monitoring,
as welding is inherently a thermal processing method.
Infrared sensing for weld process monitoring has been The experimental set-up used for real-time monitoring of
used by many researchers [3, 5-9]. IR sensing can also be the GTA welding process using an infrared (IR) camera
used for detecting the occurrence of various weld defects is shown in Figure 1. The IR camera is mounted on the
during welding as they are known to cause perturbations torch assembly and moves behind the torch at the weld-
in the surface temperature distributions of the plate being ing speed to capture instantaneous images of the weld
welded. The main advantage of IR sensing is its ability to pool and the surrounding area of the plate being welded.
monitor several weld parameters simultaneously. However, The IR thermography system used is an ALTAIR work-
the use of IR thermography for quantitative temperature station that comprises a JADE MWIR camera, FG9800
measurements in GTA welding is limited by difficulties digital PCI frame grabber and ALTAIR software for image
in handling the surface emissivity variations, and the acquisition and processing. The JADE MWIR broad-band
reflected radiation from the arc light and the hot tungsten camera uses a high quantum efficiency focal plane array
electrode during welding. Therefore, a number of issues LnSb detector that is cooled by a high reliability Stirling
must be addressed when using IR thermography for real- cooler and provides images of high sensitivity in the wave-
time imaging of the weld pool during GTA welding. length range of 1.5-5.1 μm. The IR camera uses a high-
pass band filter that permits only a portion of the emitted
energy in the wavelength range of 4.99-5.10 μm, thereby
In GTA welding, arc radiation occurs in the spectral range
minimizing the interference from arc light and hot tungsten
of 0.34 and 1.8 μm, while for wavelengths greater than
electrode on the image quality. This IR camera with high-
2 μm the weld radiation is greater in magnitude than the
pass band filter is capable of measuring temperatures in
arc energy [10]. Thus, it is possible to improve the image
the range 200-1 500 °C with an accuracy of ± 2 % over
84 quality acquired by an IR camera by filtering method and
the entire range. Each scan of the camera is transferred
this now being commonly followed to minimize the inter-
as a frame consisting of 320 × 240 discrete intensity
ference from radiant reflection of the arc and the elec-
measurements at 50 frames per second. The thermal
trode. Filters selectively pass the desired wavelengths
images appear as high-resolution colour images in real-
of IR radiation to the detector while attenuating the
time on a computer monitor and are recorded. The calibra-
unwanted wavelengths that degrade the image quality.
tion of the surface temperature measurements is carried
Bicknell [11] improved the IR image quality by using high-
out using C-type thermocouples, which are spot-welded
pass and band-pass filters. Farson [12] used a band-pass
close to the weld region, and the data are acquired in a
filter and shield to weaken the arc and electrode interfer-
computer through temperature indicators over a RS485
ence from the IR thermal radiation in GTA welding. The
network. The emissivity of 316LN SS plate is determined
accuracy of the measurement of the surface temperature
by measuring temperatures close to the weld/base metal
distributions of the plate during welding using the IR cam-
interface, using both the thermocouple and the IR camera
era depends on the emissivity of the plate, which in turn
simultaneously at three different locations, viz. at the start,
is highly dependent on the surface condition of the plate.
middle and end of the weld length, from which the heating
The surface emissivity increases during cooling and oxi-
and cooling cycles are generated.
dation and hence, it is difficult to measure the actual emis-
sivity during the cooling phase. As emissivity is a function
Autogenous GTA welding trials were carried out on 3 mm-
of temperature, variations in the surface emissivity of the
thick 316LN SS plates of size 125 × 50 mm2 to prepare
plate needs to be experimentally quantified.
square-butt weld joints. The welding parameters were
varied as given in Table 1 for achieving different weld
The objectives of the present investigations are as
bead widths and depths of penetration. A thin coating of
follows.
activated flux was applied on the weld joint area prior to
1. Apply IR sensing in real-time to monitor the molten
weld pool and the surface temperature distribution sur-
Table 1 – GTA welding parameters used
rounding the weld pool during GTA welding of 316LN
for IR thermography trials
stainless steel (SS) plate.
2. Relate the weld bead width and depth of penetration to Parameters Values
the measured surface temperature distributions. Current 50–95 A
3. Compare the actual weld bead width and depth of pen- Arc voltage 12 V
etration with the thermal bead width and the macro- Torch speed 100 mm/min
scopic temperature gradients computed from line-scan
Argon gas flow rate 10 l/min
analysis of the IR thermal image.
 REAL-TIME MONITORING of WELD POOL during GTAW using INFRA-REDTHERMOGRAPHY and analysis of Infra-Red thermal images

Figure 1 – Automatic GTA welding set-up

Figure 2 – Typical recomposed IR thermal image
of a GTA weld pool
welding to mitigate variable weld bead penetration caused
by low sulphur content (< 50 ppm). Samples cut from the
welded plates at the three thermocouple locations were un-oxidised condition and 0.8-0.9 in oxidised condition.
polished and etched to reveal the macrostructure from The temperature lag of the IR and thermocouple mea-
which the weld bead width and depth of penetration were surements was due to non-synchronization of triggering
measured using a machinists’ microscope. between the IR temperature recording and data logger for
thermocouple measurement, as also the low pixel resolu-
During these welding trials IR thermal images are tion (320 × 240) of the IR camera that affects the deter-
acquired. To improve the accuracy of IR temperature mea- mination of exact location of thermocouple junction in the
surements, two integration times of 230 μs and 48 μs IR image.
85
were used for the temperature ranges of 200-600 °C and
545-1 500 °C, respectively. The recorded IR images were Line-scan analysis was carried out at the same pixel value
then recomposed to produce composite images repre- in all three frames corresponding to the three locations
senting the entire temperature range. Further analyses, for weld joint, during which the respective emissivity val-
such as spot analysis, line-scan analysis, isothermal con- ues were used to obtain accurate temperature distribu-
touring, and full-frame temperature profile analysis, of the tions. A typical line-scan profile and its first derivative plot
recomposed IR images were carried out using the ALTAIR are shown in Figure 4. By analysing the first derivative of
image processing software. Figure 2 shows a typical IR the thermal profile, it is possible to identify the location
thermal image of a GTA weld pool. of solid-liquid interface, where a change in temperature
gradient occurs, due to the difference in emissivity of the
To determine the emissivity values at the three thermo- solid and liquid metals. Vertical lines are drawn between
couple locations, the heating and cooling cycle data from the line-scan profile and the first derivative plot to identify
both the sources were made to match well (Figure 3). the inflection points which corresponds to the solid-liquid
The emissivity values estimated were 0.6, 0.44 and 0.4, interface. Thus, for any selected line-scan, the weld bead
respectively, at the start, middle and end of the weld. width can be computed as the distance between the two
These values are in agreement with the emissivity val- vertical lines or the distance between the left-hand-side
ues reported for SS, which vary in the range 0.3-0.6 in valley and the right-hand-side peak in the first derivative

a) Start b) Middle c) End

Figure 3 – Heating and cooling cycles during GTA welding measured using C-type thermocouples
and IR thermography on plate surface at three weld length locations

08 2011 Vol. 55 WELDING IN THE WORLD

N° 07 Peer-reviewed Section
 REAL-TIME MONITORING of WELD POOL during GTAW using INFRA-REDTHERMOGRAPHY and analysis of Infra-Red thermal images

3 Results and discussion

The comparison of the actual weld bead width and the
weld bead width computed by line-scan analysis of IR
thermal profiles is shown in Figure 5 for the three loca-
tions of the welded joint. There is a good linear relation-
ship between the computed (thermal) and the actual weld
bead width values, with a minimum correlation coefficient
of 0.8. This is in agreement with results reported in the
literature [3, 5, 6].

The macroscopic temperature gradient exhibited a good

inverse linear relationship with the depth of penetration
Figure 4 – Line-scan analysis for estimation of the weld bead, with a minimum correlation coefficient
of weld bead width
of 0.88 (Figure 6). The macroscopic temperature gradi-
ent increased with a decreasing depth of penetration. This
is in agreement with results reported for welds of steel
plot. From this the macroscopic temperature gradients plates [4]. The scatter in the present data may still be due
can be computed as: to interference from the electrode radiation on the image
quality, and variation of emissivity with temperature and of
Macroscopic temperature gradient the plate surface conditions. However, the present results
Peak temperature - Interface temperature give confidence that real-time monitoring and control of
weld bead width and depth of penetration during GTA
Distance between peak and interface
welding is feasible using IR sensing.

Full-frame analysis using the ALTAIR software was car- The temperature distribution surrounding the weld pool
ried out on the recorded IR images by the timing graph can be estimated from the acquired IR images of the weld
86 technique to obtain plots of IR intensity as a function of pool. The temperature distribution on the plate surface for
frame number, which were then used as IR thermal signa- 3 mm thick square-butt weld joint of 316LN SS is shown
ture images for good and defective welds. in Figure 7.

a) Start b) Middle c) End

Figure 5 – Comparison between thermal (computed) and actual weld bead width at three weld length locations

a) Start b) Middle c) End

Figure 6 – Correlation between macroscopic temperature gradient and depth of penetration of weld bead
at three weld length locations
 REAL-TIME MONITORING of WELD POOL during GTAW using INFRA-REDTHERMOGRAPHY and analysis of Infra-Red thermal images

generating IR thermal signature images for welded joints

of 316LN SS with and without defects.

The photograph of a welded joint without any defect

and the corresponding IR thermal signature image plot
is shown in Figure 8. The thermal signature image plot
consists of variation in the surface temperature of the
plate along the weld centre-line as a function of frame
number or weld length. The temperature almost remained
constant for the entire length of the weld implying that
welding conditions used produced a welded joint without
any defect. This thermal signature is taken as reference
thermal image for a good weld.

Figure 7 – Temperature distribution surrounding Figure 9 shows the photograph of a welded joint with lack
the weld pool on the plate surface for 3 mm-thick of fusion (LOF) weld defect with the defect area marked
square-butt weld joint of 316LN SS with a circle and the corresponding IR thermal signature
image plot. Here again the defect area is marked with
Each welding process has unique thermal signatures and a circle in the plot. In the region corresponding to the
any deviations would lead to perturbations in the surface LOF defect, there is a sudden drop in temperature. This
temperature distribution, and this forms the basis for is attributed to the differences in the emissivity between

87

a) Photograph of welded joint without defect b) Corresponding IR thermal signature image

Figure 8

a) Photograph of welded joint with lack of fusion defect b) Corresponding IR thermal signature image

Figure 9

08 2011 Vol. 55 WELDING IN THE WORLD

N° 07 Peer-reviewed Section
 REAL-TIME MONITORING of WELD POOL during GTAW using INFRA-REDTHERMOGRAPHY and analysis of Infra-Red thermal images

the liquid and the solid, as there was no melting in this inclusions were verified using X-ray radiography. The weld
region. This image is taken as reference signature image in the present case ends with a tungsten inclusion and
for LOF defect. This defect was generated by shutting off hence there is a corresponding rise in the temperature
the welding current for a few seconds during welding. at the end region of the welded joint. This image plot is
taken as reference for tungsten inclusions. Small tung-
The photograph of a welded joint with lack of penetration sten pieces were introduced into the weld pool during
(LOP) defect is shown in Figure 10 along with the cor- welding to produce this defect.
responding IR thermal signature image plot. The defect
areas are marked in the photograph as well as in the

4 Conclusions
image plot. There is drop in the temperature plot for the
entire length of the LOP defect. The LOP caused per-
turbations in the surface temperature distribution. This is
taken as reference image for LOP defect. This defect was The conclusions from the present investigations are as
generated by varying the current throughout the welding follows.
by PLC programming.
1. A linear relationship exists between the weld bead
The photograph of a welded joint with tungsten inclu- widths computed from IR thermal profile and the actual
sion defects and the corresponding IR thermal signature measured values.
image plot is shown in Figure 11. Here again the defect 2. An inverse linear relationship exists between the mac-
areas are marked in both the photograph and the image roscopic temperature gradient computed from the IR
plot. At locations of tungsten inclusions, there is a definite thermal profile and the measured depth of penetration
peak in the temperature plot. These locations of tungsten of the weld bead.

88

a) Photograph of welded joint with lack of penetration defect b) Corresponding IR thermal signature image

Figure 10

a) Photograph of welded joint with tungsten inclusion defect b) Corresponding IR thermal signature image

Figure 11
 REAL-TIME MONITORING of WELD POOL during GTAW using INFRA-REDTHERMOGRAPHY and analysis of Infra-Red thermal images

3. Temperature distribution surrounding the weld on the [6] Menaka M., Vasudevan M., Venkataraman B. and Raj
plate surface can be estimated as isothermal contours B.: Estimating bead width and depth of penetration dur-
of IR thermal images. ing welding by infra-red thermal imaging, Insight, 2005,
4. The IR thermal signature images generated for weld vol. 47, no. 9, pp. 564-568.
defects, such as lack of fusion, lack of penetration and
tungsten inclusions can be used as reference images [7] Ghanty P., Vasudevan M., Mukherjee D.P., Pal N.R.,
for on-line monitoring of weld quality using IR thermog- Chandrasekhar N., Maduraimuthu V., Bhaduri A.K., Barat P.
raphy during GTA welding of 316LN SS. and Raj B.: Artificial neural network approach for estimat-
5. IR thermography is a potential tool for on-line moni- ing weld bead width and depth of penetration from infrared
toring and control of weld bead geometry during GTA thermal image of weld pool, Science and Technology of
welding. Welding and Joining, 2008, vol. 13, no. 4, pp. 395-401.
Based on these investigations, a computational intelli- [8] Balfour C., Smith J.S. and Al-Shammma A.I.: Novel
gence (CI) based control system is under development edge feature correlation algorithm for real-time computer
and testing for closed-loop feed-back control of depth vision-based molten weld pool measurements, Welding
of penetration from IR thermal images in real-time. This Journal, 2006, vol. 86, no. 1, pp. 1s-8s.
CI-based control system would carry out on-line correc-
tion of any deviation in the depth of penetration by adjust- [9] Al-Habaibeh A. and Parking R.: An autonomous low-
ing the welding current in real time by sending appropriate cost infrared system for the on-line monitoring of manu-
signals to the power source through the PLC. facturing processes using novelty detection, International
Journal of Advanced Manufacturing Technology, 2003,
vol. 22, no. 3-4, pp. 249-258.

References [10] Huang R.S., Liu L.M. and Song G.: Infrared tempera-
ture measurement and interference analysis of magne-
[1] Chandrasekhar N. and Vasudevan M.: Computer sium alloy in hybrid laser-GTA welding process, Materials
based operation, monitor and control of GTA welding Science and Engineering A, 2007, vol. 447, no. 1-2,
machine and temperature measurements during weld- pp. 239-243.
ing, International Conference on Advance Manufacturing
Technology, Indian National Academy of Engineering, 89
[11] Bicknell A., Smith J.S. and Lucas J.: Infrared sen-
Chennai (6-8 February 2008). sor for top face monitoring of weld pools, Measurement
Science and Technology, 1994, vol. 5, no. 4, pp. 371-378.
[2] Nagarajan S., Chen W.H. and Chin B.A.: Infrared sens-
ing for adaptive arc welding, Welding Journal, 1989, [12] Farson D., Richardson R. and Li X.: Infrared measure-
vol. 68, no. 11, pp. 462s-466s. ment of base metal temperature in gas tungsten arc weld-
ing, Welding Journal, 1998, vol.77, no. 9, pp. 396-401.
[3] Chen W.H. and Chin B.A.: Monitoring joint penetra-
tion using infrared sensing techniques, Welding Journal,
1990, vol. 69, no. 4, pp. 181s-185s.

[4] Smith J.S. and Lucas W.: Vision based systems for
controlling the arc welding operation and inspecting the
weld bead profile, Welding in the World, 1999, vol. 43,
Supplementary Issue, pp. 103-115.
About the authors
[5] Banerjee P., Govardhan S., Wikle H.C., Liu J.Y. and Chin Dr. Muthukumaran VASUDEVAN ([email protected]), Mr. Neelamegam
B.A.: Infrared sensing for on-line weld geometry monitor- CHANDRASEKHAR ([email protected]), Mr. Velu MADURAIMUTHU
([email protected]), Dr. Arun K. BHADURI ([email protected]) and Dr. Baldev
ing and control, Journal of Engineering for Industry, 1995, RAJ ([email protected]) are all with Indira Gandhi Centre for Atomic Research,
vol. 117, no. 3, pp. 323-330. Kalpakkam (India).

08 2011 Vol. 55 WELDING IN THE WORLD

N° 07 Peer-reviewed Section