268
D. Mishra et al.
the welding torch accordingly by means of suitable cameras, monitoring the laser
light. These sensors help in identifying and locating the joint seam before the start
of the welding. One of the major hurdles in this process is to ensure the safety of the
sensors, which includes the laser and the camera, as they are fixed onto the torch. The
sensor may also get affected by the harsh environment, which is another limitation.
The torch embedded with camera and laser becomes inaccessible to all regions. Other
than seam tracking, the optical detectors have been utilized for classification of weld
into defective and defect-free in arc welding technique [34]. In this work, the utilized
optical sensor consisted of two visible detectors, one for detecting the infrared rays,
and the other one for the ultraviolet rays. Various descriptive statistics were observed
from the signals for classifying the welds. For illumination, a diode-laser system was
utilized.
Similarly, the use of vision-based system in FSW technique has spanned from
classifying welds into defective or defect-free, and further classifying the welding
defects. For the purpose of weld classification, a number of descriptive statistics such
as mean, standard deviation, entropy, energy, contrast and homogeneity were derived
from the weld images [35]. It was found that standard deviation was a better indicator
for weld classification. Similarly, features such as energy, variance and entropy were
derived from the wavelet coefficients of the weld images for weld classification [36].
The defect-free weld had high values of energy and entropy, and less variance, as
compared to the defective weld. Another study proposed the feature extraction from
the weld images by using “maximally stable extremal region” algorithm for weld
classification [37]. This algorithm is based on the idea of identifying regions with
almost no variation through a wide range of thresholds. The extracted features were
fed as inputs to a ML model for weld classification. Another study proposed the
use of contour and profile plots for weld classification where the grey level intensity
was found to be distinctly varying for defect-free and defective welds [38]. For the
classification of the welding defects, a study reports usage of image pyramid and
reconstruction techniques [39, 40]. In the purview of ML, “support vector machine”
(SVM) has been utilized for classification of the weld images because of their ability
to generalize problems, even with fewer training samples [36, 37].
The vision-based system is advantageous as it does not directly interact with
the weld pool and carry abundant information. However, because of several practical
limitations, their use is also limited. The major drawback is the precise positioning of
these instruments. The shop floor in an industry may consist of several machineries,
all of which may be operating simultaneously. Thus, the presence of vibration is quite
likely. The optical-based instruments are supposed to be placed in a vibration-free
environment so as to ensure the accuracy of their measurements and repeatability as
well. With a practical industry environment, ensuring both is quite difficult a task.
Other practical limitations of these instruments include the delicate handling and
loss of precision in dusty environment.
D. Mishra et al.
the welding torch accordingly by means of suitable cameras, monitoring the laser
light. These sensors help in identifying and locating the joint seam before the start
of the welding. One of the major hurdles in this process is to ensure the safety of the
sensors, which includes the laser and the camera, as they are fixed onto the torch. The
sensor may also get affected by the harsh environment, which is another limitation.
The torch embedded with camera and laser becomes inaccessible to all regions. Other
than seam tracking, the optical detectors have been utilized for classification of weld
into defective and defect-free in arc welding technique [34]. In this work, the utilized
optical sensor consisted of two visible detectors, one for detecting the infrared rays,
and the other one for the ultraviolet rays. Various descriptive statistics were observed
from the signals for classifying the welds. For illumination, a diode-laser system was
utilized.
Similarly, the use of vision-based system in FSW technique has spanned from
classifying welds into defective or defect-free, and further classifying the welding
defects. For the purpose of weld classification, a number of descriptive statistics such
as mean, standard deviation, entropy, energy, contrast and homogeneity were derived
from the weld images [35]. It was found that standard deviation was a better indicator
for weld classification. Similarly, features such as energy, variance and entropy were
derived from the wavelet coefficients of the weld images for weld classification [36].
The defect-free weld had high values of energy and entropy, and less variance, as
compared to the defective weld. Another study proposed the feature extraction from
the weld images by using “maximally stable extremal region” algorithm for weld
classification [37]. This algorithm is based on the idea of identifying regions with
almost no variation through a wide range of thresholds. The extracted features were
fed as inputs to a ML model for weld classification. Another study proposed the
use of contour and profile plots for weld classification where the grey level intensity
was found to be distinctly varying for defect-free and defective welds [38]. For the
classification of the welding defects, a study reports usage of image pyramid and
reconstruction techniques [39, 40]. In the purview of ML, “support vector machine”
(SVM) has been utilized for classification of the weld images because of their ability
to generalize problems, even with fewer training samples [36, 37].
The vision-based system is advantageous as it does not directly interact with
the weld pool and carry abundant information. However, because of several practical
limitations, their use is also limited. The major drawback is the precise positioning of
these instruments. The shop floor in an industry may consist of several machineries,
all of which may be operating simultaneously. Thus, the presence of vibration is quite
likely. The optical-based instruments are supposed to be placed in a vibration-free
environment so as to ensure the accuracy of their measurements and repeatability as
well. With a practical industry environment, ensuring both is quite difficult a task.
Other practical limitations of these instruments include the delicate handling and
loss of precision in dusty environment.