Accepted Manuscript

Signal processing approach on weld data for evaluation of arc welding electrodes
using probability density distributions

Vikas Kumar, Shaju K. Albert, N. Chandrasekhar

PII: S0263-2241(18)30910-2
DOI: https://doi.org/10.1016/j.measurement.2018.09.076
Reference: MEASUR 5934

To appear in: Measurement

Received Date: 30 May 2018

Revised Date: 25 September 2018
Accepted Date: 29 September 2018

Please cite this article as: V. Kumar, S.K. Albert, N. Chandrasekhar, Signal processing approach on weld data for
evaluation of arc welding electrodes using probability density distributions, Measurement (2018), doi: https://
doi.org/10.1016/j.measurement.2018.09.076

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Signal processing approach on weld data for evaluation of
arc welding electrodes using probability density
distributions
Vikas Kumar1, Shaju K Albert2, N Chandrasekhar3
1School of Electronics Engineering, Kalinga Institute of Industrial Technology, Bhubaneswar, India
2,3Materials Engineering Group, Indira Gandhi Centre for Atomic Research, Kalpakkam, India
*Corresponding author email and contact: 1*[email protected]; +91-9655855934

Abstract
Shielded Metal Arc Welding (SMAW) Process utilizes a constant current type power source with drooping
characteristics. Due to complex nature of welding arc and metal transfer that occurs during welding, there
is a lot of random variations in welding current and voltage which cannot be recorded directly by normal
ammeter or voltmeter. However, acquisition of welding data while welding is in progress and subsequent
analysis of this data can be very useful to evaluate various welding parameters (i.e. welding consumables
etc.). For this purpose, high speed of data acquisition is essential. As noise level in the data will be high
hence, before performing any meaningful analysis filtering of this data is also important. In the present
study, a technique is proposed for the reliable acquisition of welding data to acquire all the possible
variations in arc voltage while welding is in progress using a Digital Storage Oscilloscope (DSO). Various
signal processing methods were used for selecting the appropriate filtering technique. Filtered data thus
obtained were used to evaluate arc welding electrodes with different flux coating using probability density
distributions. The results thus obtained were correlated with the images obtained using high speed camera
setup. This clearly brings out the differences in the arc characteristics for welding consumables. They also
indicate that the proposed technique can be developed as a tool to compare the performance of different
welding electrodes.

Keywords: SMAW; Data processing; Consumables; Process Monitoring; sensors;

1. Introduction
Arc Welding is one of the most common and popular welding technique because of its simplicity in the
operation, versatility and low cost. Due to its inherent merits, this process is extensively used in almost all
type of industries. Random arc behavior and various modes of metal transfer in an arc welding process
leads to dynamic variations in voltage and current while welding is in progress and hence, monitoring of
this process is difficult and challenging. Welding process can be understood in a much better way than it is
done at present, if we can record the variations that occur during actual process itself at the same rate as
they are happening. The challenge in developing such a mechanism is the random behavior of the welding
arc due to various physical processes that happen across the arc in a very short time while welding is in
progress. These include metal transfer, short circuiting, spatter, ionization, gas-metal reactions etc. During
welding these variations are reflected as fluctuations in the current and voltage signals. If such dynamic
variations that occur in voltage and current during welding can be properly acquired and analyzed, then the
same can be used to evaluate various welding parameters. Similar studies on analyzing various welding
parameters like power sources and consumables have already been reported in the literatures. Previous
literature has reported the use of statistical tools like Probability Density Distribution (PDD) [1-8], signal
processing, ANN techniques for this purpose. To show that the arc stability can be improved by using nano
powders of CaCO3 in the flux coating of the welding consumable, PDD technique has been used by Chen
et al. [2]
Recently, voltage and current signals were captured during the actual welding process and the correlation
between various weld defects and the signature patterns of current and voltage PDDs were established [3].
Welding data acquired form Gas Metal Arc Welding (GMAW) process was analyzed using VC++ and
MATLAB programming by plotting Probability Density Distribution (PDD) and Power Spectral Density
(PSD) curve [4]. Very recently, welding data (voltage and current) were acquired using a general purpose
DSO to study various aspects of SMAW process [5-7]. For fault discrimination in a welding process, Li et
al., has derived the range of parameters from the welding data. Subsequently, independent component
analysis was used to separate arcing and short circuiting metal transfer from the sampled data [9].
In another study, weld data (voltage and current) was acquired using cellulosic coated electrodes to
examine the effect of alloy enrichment in the weld metal deposit [10]. Statistical Process Control (SPC)
system is used to compute various statistical parameters like mean and standard deviation. Changes in these
statistical parameters are displayed using control charts and are correlated with weld quality. For this
purpose, the authors have used trending analysis, tolerance analysis, and sequential analysis techniques
[11].
Experimental measurements were made to examine the effect of drop transfer with changing voltage in
pulse gas metal arc welding (GMAW-P). Here the authors have shown that the welding arc voltage is
getting affected by molten droplet detachment, this in turn can be used to understand the characteristics of
molten metal transfer in GMAW-P [12].
Using signature images Simpson S.W. has demonstrated that fault recognition in a GMAW process is
possible on real time basis. Further, it is also shown that the signature image technique behaves reasonably
well to identify unknown faults occurring in a welding environment [13-17].
 Many researchers have used Artificial Neural Network techniques also for analyzing welding data [6-
7]. Pal et al. [18] has used various statistical parameters of arc signals as the input of a neural network model
for predicting the weld quality in pulsed metal inert gas welding. Fuzzy clustering method has been used
by Wu et al. [19] for intelligent monitoring and recognition of short circuiting in GMAW process. In a
separate study [20-21], to identify the process disturbances in a GMAW process fuzzy logic and neural
network techniques has been used. Wavelet packet transform analysis on the welding current has been used
by pal et al. [22] to predict the strength of weld joint in a pulsed metal inert gas welding. Li Di, Song
Yonglun and Ye Feng [23] have used Self Organizing Map (SOM) neural network technique for feature
extraction from arc signal for automatic detection of weld defects in short circuit gas metal arc welding to
find the weld quality and defects in the product. Li. D, song, Y and Ye and Fhanjie Xuebao, have used arc
sensing technique for identifying the weld defects in GMAW process. Here, the authors have extracted the
features from the arc signals by classifying signal histogram using SOM technique [24].
In a separate study MATLAB Simulink package was used to develop different control algorithms to
develop a computer animated Gas Metal Arc Welding emulation module. This was used to train welding
personals so that they can gain sufficient insight about GMAW process and learn this technique [25].
In the present study, choice of appropriate data acquisition rate and filtering technique for welding
signals are proposed. The data were acquired using different welding consumables and a high speed DSO
while welding is in progress. These electrodes have different flux compositions consequently, by statistical
analysis of the data thus obtained it is possible to differentiate welding consumables as soon as the weld
pad has been prepared.
For online storing of entire data an Ethernet cable is interfaced between Digital Storage Oscilloscope
(DSO) and an external computer. Data thus acquired for various electrodes were used for post processing
to study and compare the behavior of various welding consumables and finally results were compared and
correlated with those obtained from the high speed camera.
2. Experimental setup
2.1 Data acquisition setup
The data acquisition setup is similar to the one we have used in our previous experimental work [5-7].
For online acquisition of welding data, a DSO (Model No DSO7054B) of 500 MHz band width with a
maximum sampling rate of 4 GSa/s is used. It has an extended graphic display of 12.1 inch, 4 channels and
A.C to D.C coupling. The in depth memory of the DSO is 8 Mpts with an input impedance of 50 Ω and 1M
Ω. The oscilloscope is used for real time collection and display of welding parameters. For current
measurement, High sensitive 1mA/A or 10mA/A Hall Effect based PAC series current clamp is used. With
this, measurements are possible up to 600A DC and 100A AC. For the voltage measurement differential
probe is used, these probes are optimized for acquiring differential signals.
2.2Welding setup
The arc welding machine setup with data acquisition system has been illustrated using Fig. 1(a) and the
complete schematic of this setup is shown in Fig. 1(c). For comparing different welding consumables and
to show that the procedure developed in this study can be used to differentiate the arc characteristics of
different electrodes data was acquired while welding using basic coated, rutile coated and cellulose quoted
electrode and analyzed. Bead-on-plate welds were made using an inverter power source and E7018 (basic
coated), E6013 (rutile coated) and E6010 (cellulose coated) electrodes on carbon steel plates by the same
welder. Diameter of the electrodes employed for welding was 3.15 for the first two while the cellulose
electrode was of 2.5 mm diameter. Constant current of 100A was set for E 7018 and E 6013 type electrode
whereas 50A current was set for E 6010 type of electrodes. Voltage and current signals were then acquired
for 20 s of duration using DSO setup. To ensure consistency of the data, three bead-on-plate welds were
made using each type of electrode and data was acquired for all the three welds. Data was subsequently
filtered and then subjected to time domain and statistical analysis.
As welding is carried out for all the three types of electrodes using same welding and power source, the
difference in their arc characteristics should be related to differences in the metal transfer behavior across
the arc in these electrodes. Hence, this metal transfer across the arc is viewed using high speed camera with
laser illumination and suitable filters. Fastcam MC2.1 photon focus was the camera used which has a
maximum capturing speed of 10,000 frames per second and the minimum shutter opening time of
1/100,000s. In the current experiment images were acquired at 5000 frames/s. The camera comprises of the
filtering system, macro optical system, capture board and synchronizing module. Images are acquired via
camera to PC. For the 5000 frames/s the high speed camera records the last 4.094 seconds of data from a
total welding duration of 20 seconds in the current experiment. The photograph of this system is shown in
Fig. 1(b) and the schematic of compete experimental setup is shown in Fig. 2.
Though electrode fluxes are classified broadly as cellulose, rutile, basic etc, the exact composition of
the flux can vary within the same type of flux. For example, basic coated electrode may contain some rutile
to improve the arc characteristics or slag detachability. Similarly, depending on other constituents of the
flux, the exact wt. % of the major constituent of the flux can also vary. It is expected that these variations
should result in changes in the characteristics of the welding arc. It is demonstrated that arc data acquired
using DSO is able to differentiate the differences in the arc characteristics caused by subtle changes in the
flux composition. In order to demonstrate this welding data was acquired from the two sets of electrodes;
one is E 7018 type electrode in which rutile content is varied from 0-7 wt.% and another is E 6013 type
electrodes in which rutile content is varied from 45-55 wt.% (Table 1). Data was acquired for each of these
electrodes while making bead-on-plates with 120A current and 25 V by the same welder and same power
source.
(a)Experimental setup of High Speed Data Acquisition System (b) Experimental setup for High Speed Image acquisition
system

(c) Schematic of welding setup with DSO

Fig. 1: Photograph of high speed data and image acquisition Setup.

Fig. 2: Schematic of complete experimental setup.

 Table 1 Rutile content variations in E7018 and E6013 electrodes.

Rutile /Rutile sand

Electrode type
variation
E 7018 nil
E 7018 3%
E 7018 7%
E 6013 45%
E 6013 50%

E 6013 55%

3. Results and Discussions

3.1 Choice of appropriate data acquisition rate
A typical time domain plot using E 7018 electrode has been represented using Fig. 3, which illustrates
the extent of variations taking place in actual welding process. The variations around steady state value
(recorded value of ~ 25 V) will take place due to the random arc gap variations in SMAW process.
Consequently, in these figures (Fig. 3) steady state voltage values are varying from ~ 18 – 35 V. These time
domain plots also show sudden dips in voltage values; the duration of these sudden dips in voltage and the
manner in which they occur can be correlated with the actual welding process.
Although the data was acquired for 20 seconds of duration in all the analysis shown above only 500 ms
data was used for time domain analysis. PDD technique enables us to use the entire data acquired for the
20 seconds to study the arc welding processes. A typical PDD plot using basic type E7018 electrode has
been illustrated using Fig. 4(b) represents. It is observed that voltage PDD for the weld made by E7018
electrode has two distinct peaks, one at the low voltage and the other at the high voltage. First peak in this
PDD corresponds to sudden voltage dips noticed in the time domain analysis of voltage signals (Fig. 3),
whereas the second peak in the voltage PDD corresponds to the steady state voltage value (recorded value
of ~25 V).
 Fig. 3: Time domain analysis of welding voltage acquired at different acquisition rates.

(a) PDD with varying acquisition rates. (b) Three trials of PDD at 100,000 samples/s.
Fig. 4: PDD analysis of voltage data.

Rate of weld data acquisition is extremely important as acquisition of data at a speed lower than that of
the actual events occur during welding would result in modification of the actual signals. Hence, this data
acquisition rate should be sufficiently high compared to variations taking place in actual process. Hence, to
arrive at appropriate data acquisition rate voltage data was acquired at various rates to collect different
quantum of data. This in turn was analyzed using time domain and PDD curve. It should be noted that
current data was not utilized for this purpose because the SMAW process essentially utilizes CC (Constant
Current) type of power source in which only voltage variations will be significant. From the time domain
analysis of voltage data (Fig. 3), acquired at different rates, effect of sampling rate is not clear even though
the sampling rate varied from 20000 to 1,000,000 samples/s. However, from the PDD analysis of voltage
data in Fig. 4 (a) influence of data acquisition rate can be clearly seen. From this figure it is clear that
voltage PDD attains saturation after around 100,000 samples/s. This means that after this acquisition rate
further increase in data sampling rate is not making significant difference in acquired data. Consequently,
100,000 samples/s are chosen as a standard data acquisition rate for weld parameter analysis. To show the
repeatability of acquired data, three trials from each acquisition rate was performed and analyzed. It was
found that data was repeatable for a particular acquisition rate. The fact just presented can be understood
by observing Fig. 4(b) where all the three trials of voltage PDD acquired at 100,000samples/s shows an
excellent repeatability.
PDD analysis of the voltage and current data obtained using DSO (in Fig. 4) were compared with the
PDD of the voltage data acquired using Analysator Hanover [8] (in Fig. 5 and 6). This is done in order to
optimize the time domain and PDD techniques for our applications. For the comparison purpose these data
were acquired using same electrode (E 7018), same welder and an inverter power source. Very good
correlation and resemblance between the PDDs obtained with DSO and the PDD obtained with Analysator
can be seen clearly from these figures.
Hence, with the exercise just presented we optimized time domain and PDD techniques. These techniques
will be used to evaluate different welding consumables in upcoming sections.

(a)Voltage PDD (b) Current PDD

Fig. 5 PDD analysis of the voltage and current data acquired using DSO.
 (a)Voltage PDD (b) Current PDD
Fig. 6: PDD analysis of the voltage and current data acquired using Analysator Hanover [8].

3.2 Performance Evaluation of Filtering Techniques.

The raw data obtained from a data acquisition system in a welding environment is corrupted with
unwanted noise. Hence, electrical signal thus obtained should be filtered with appropriate filtering
techniques so that the noise affects the further processing in the least possible way. There are many
conventional frequency domain filtering like FFT [26] and many recent techniques like the Wavelet
Transform [27].
The two main techniques employed for processing the signal are the pass filters (High pass, low pass,
band pass and band elimination filters) based on FFT filtering and Multi-resolutional Analysis using the
Discrete Wavelet Transform (DWT) Method. The multi-resolutional analysis was carried out using
Biorthogonal (bior) Wavelet and Daubechies (db) Wavelet. To compare the performance of the two filtering
techniques mentioned above, the Signal to Noise Ratio (SNR) was chosen as a criterion to evaluate their
performance. The SNRs of filtered signals has been recorded and tabulated in Table 2. It can be concluded
that the SNR for the Low-Pass FFT method shows an exceptional improvement in SNR as compared to the
improvement by DWT method. Hence, for all the analysis purpose FFT LPF has been used for filtering out
the noise from the weld signal. Here, it should be noted that the cutoff frequency of FFT LPF was optimized
at 500 Hz to obtain the maximum SNR of 48.874 dB in Table 2.
 Table 2: SNR values of various filtering technique with original signal SNR = 26.2267
Filtering Technique Signal to Noise Ratio (SNR in dB)
Low Pass Filter (LPF) 48.874
Low Pass Filter (LPF) with db2 Level 1 Wavelet 28.1033
Low Pass Filter (LPF) with db2 Level 3 Wavelet 29.1325
Low Pass Filter (LPF) with db2 Level 5 Wavelet 32.445
Db2 Level 1 Wavelet 27.711
Db2 Level 3 Wavelet 28.284
Db2 Level 5 Wavelet 28.371
Db2 Level 7 Wavelet 28.7765
Db5 Level 1 Wavelet 27.7188
Db5 Level 3 Wavelet 28.2836
Db5 Level 5 Wavelet 28.4488
Db5 Level 7 Wavelet 28.5085
Db8 Level 1 Wavelet 27.7284
Db8 Level 3 Wavelet 28.288
Db8 Level 5 Wavelet 28.4485
Db8 Level 7 Wavelet 28.6306
Bior 2.2 Level 1 Wavelet 27.5808
Bior 2.2 Level 3 Wavelet 28.159
Bior 2.2 Level 5 Wavelet 28.4415
Bior 2.2 Level 7 Wavelet 28.697

3.3 Variation in the arc characteristics of different electrodes.

The PDD curves thus obtained for the filtered voltage data (using FFT LPF) for welds of E 7018 and for
welds of E 6013 electrode in Fig. 7shows two distinct peaks, the first or lower voltage peak in the PDD
corresponds to short circuit metal transfer in E 7018 and E 6013 electrodes. Whereas the other peak
represents the voltage displayed on the voltmeter of the power source. In Shielded Metal Arc Welding
carried out using E 7018 and E 6013 electrodes, majority of metal transfer occurs in the form of short circuit
[28]. This can be further verified by observing the image of metal transfer obtained using high speed camera
setup in Fig 8. Further from this figure it is observed that in E 7018 and E 6013 electrodes majority of metal
transfer is occurring by the means of short circuit and the droplet size in case of E 6013 is larger than E
7018 electrodes which is in agreement with earlier results reported by Haiyun et. al. [28].
 (a) E 7018 electrode (b) E 6013 electrode
Fig. 7: PDD analysis of voltage data.

(a) Short circuit transfer in E 7018 electrode

(b) Short circuit transfer in E 6013 electrode

Fig. 8: Short circuit image seen with high speed camera.

Fig. 9 shows the voltage PDD obtained using E 6010 electrode, a comparison of this PDD to that of one
obtained in Fig. 7 reveals that, in case of E 6010 electrodes in addition to first two peaks there is an
additional peak at higher voltages. This is due to the explosive transfer that happens in E 6010 electrodes
in addition to short circuit metal transfer [28-29]. The fact just presented can be confirmed by examining
the high speed images of metal transfer across the arc during welding using E 6010 electrode (Fig. 10 (a)).
This figure shows the short circuit metal transfer that occurs in E 6010 electrodes and Fig. 10 (b) is the
explosive/spray transfer that take place in the same electrodes. It should be noted that explosive/spray mode
of transfer was not observed during welding using E 7018 or E 6013 electrodes.

Fig. 9: Voltage and current PDD for E 6010 electrode using inverter power supply [5].

(a) Short circuit mode of metal transfer(b) Spray mode of metal transfer
Fig. 10: Different modes of metal transfer that take place in E 6010 electrode.

3.4 Variation in arc characteristics with flux composition

Fig. 11 shows the time domain analysis of E 6013 electrode. From this figure it can be seen that, when
the rutile content in E 6013 electrode was 45 -50% (Fig. 11 (a, b)) frequent voltage dips on account of short
circuit metal transfer was noticeable. As soon as the rutile content in E 6013 increase to 55% (Fig. 11(c)),
frequency of these voltage dips reduced to a greater extent, this in fact means reduction in short circuit
mode of metal transfer.
 (a) E 6013 with 45% rutile content (b) E 6013 with 50% rutile content

(c) E 6013 with 55% rutile content

Fig. 11: Time domain analysis of E 6013 electrode with different rutile compositions.

Fig. 12: PDD analysis of E 6013 electrode with varying rutile content.

The results just presented using time domain gets reflected in PDD analysis as well. Fig. 12 shows the
PDD analysis of E 6013 electrode with varying rutile content as specified in Table 1. Form this figure we
can see that for E6013 electrode with 45% and 50% rutile content extent of occurrence of first peak, which
corresponds to short circuit metal transfer is more in the voltage PDD. But as the rutile content in this
electrode increases to 55% first peak in voltage PDD gets reduced, this corresponds to reduction of short
circuit metal transfer in this electrode. Hence, it is clear that the addition of rutile content in E 6013
electrodes clearly affects the PDD characteristics but, the influence of the rutile content on E 6013
electrodes needs more discussions. A separate study with more data sets can be carried out to explain this
behavior in detail.
Whereas the time domain analysis of E 7018 electrode in Fig. 13, reveals that irrespective of the rutile
content in this electrode there are frequent occurrence of voltage dips. Here we can see that introduction of
any amount of rutile content does not affects the frequency of voltage dips much. This means that in basic
type of electrodes (i.e E 7018) majority of metal transfer occurs in the form of short circuit.
The facts just presented with time domain analysis of E 7018 are in correlation with that of its PDD
analysis in Fig. 14. From this figure it can be seen that although there is slight difference in their voltage
PDD first peak in these electrodes are predominately high, this means majority of metal transfer in these
electrodes, irrespective of extent of rutile content, are in the form of short circuit. One possible reason of
the results just obtained could be the extent to which rutile content was introduced in these electrodes. E
7018 which is already a basic coated electrode addition of small amount of rutile will not affect the basic
nature of metal transfer in these electrodes (which is predominantly short circuit mode of metal transfer)
and the majority of metal transfer will remain the same. Hence, the first peak of the PDD (which depicts
the short circuit metal transfer) in Fig. 14 remains intact.

(a) E 7018 with nil rutile content (b) E 7018 with 3% rutile content

(c) E 7018 with 7% rutile content

Fig. 13: Time domain analysis of E 7018 electrode with different rutile compositions.
 Fig. 14: PDD analysis of E 7018 electrode with varying rutile content.

Thus it is clear that the procedure developed in this study on the data acquired using a general purpose
DSO is able to bring out the differences in the arc characteristics of various electrodes with different flux
coatings. These differences are in agreement with the differences in the arc atmosphere while using these
electrodes.
4. Conclusion
From the above results, it is clear that for reliable acquisition of welding signal 1,00,000 samples/s will
be sufficient. Fast Fourier Transform Low Pass Filter (FFT LPF) is having highest value of SNR and hence,
the same is appropriate filtering technique for noise processing in the welding data. From the analysis of
data thus obtained, it is possible to differentiate between different modes of metal transfer in SMAW
electrodes. The PDD pattern for one type of electrode is different from other, this means that these patterns
can be used as a signature pattern of a particular electrode to differentiate it from other.Hence, welding
consumables can be differentiated using voltage PDDs alone. Using these signature patterns, it is also
possible to differentiate same type of welding consumables produced from same type of electrodes by
different electrode manufacturers and evaluate them on the basis of their arc characteristics.

5. Acknowledgements
We thank Dr. A. K. Bhaduri, Director, Indira Gandhi Centre for Atomic Research (IGCAR), G
Kempulraj, Head, CWD (Central Workshop Division), Mr. Krishnamurthy, EIC (Engineering in charge),
Fabrication Section, CWD of IGCAR for their support and encouragement during this whole study. We
also acknowledge the support received from our colleagues in materials technology division and CWD in
conducting welding trials. We also gratefully acknowledge the support received from Dr. Santha Kumari,
Senior DGM, Bharat Heavy Electricals Limited (BHEL), Trichy (India).
References

1. Rehfeldt D, Rehfeldt M D (2015) Statistical evaluation of GMAW process disturbances with signature analysis
through Analysator Hannover. Journal of Chemical and Pharmaceutical Sciences (9): 274–279.
2. Chen B, Han F, Hhuang Y, Lu K, Llu Y, Li L (2009) Influence of nanoscale marble (calcium carbonate CaCO3)
on properties of D600R surfacing electrode. Welding Journal (88): 99-103.
3. Sumesh A, Rameshkumar K, Raja A, Mohandas K, Santhakumari A, Shyambabu R (2017)
Establishing Correlation Between Current and Voltage Signatures of the Arc and Weld Defects in GMAW
Process. Arabian Journal for Science and Engineering (42): 4649-4665.
4. Bing Xiao (2011) welding arc signal acquisition and analysis system based on VC++ and MATLAB mixed
programming. IEEE Proceedings 3rd Third International Conference on Measuring Technology and
Mechatronics Automation (ICMTMA) (3): 1150–1153.
5. Kumar Vikas, Chandrasekhar N, Albert S K, Jayapandian J (2016) Analysis of arc welding process using digital
storage oscilloscope. Measurement (81): 1–12.
6. Kumar Vikas, Chandrasekhar N, Albert S K, Jayapandian J (2018) Evaluation of welding skill using probability
density distributions and neural network analysis. Measurement (116): 114–121.
7. Kumar Vikas, Albert S K, Chandrasekhar N, Jayapandian J (2016) Performance analysis of arc welding
parameters using self-organizing maps and probability density distributions. IEEE First International
Conference on Control, Measurement and Instrumentation, Kolkata (CMI): 196–200.
8. Rehfeldt D, Rehfeldt M (2003) Computer-aided quality assurance (CAQ) of Al-GMAW-welding with
Analysator Hannover. Proceedings International Forum on Automobile Welding, Mechanical Engineering
Press, Beijing.
9. Li X, Simpson S W (2009) Parametric approach to positional fault detection in short arc welding. Science and
Technology of Welding and joining (14): 146–151.
10. Ramirez J E, Johnson M (2010) Effect of welding parameters and electrode condition on alloying enrichment
of weld metal deposited with coated cellulosic electrodes. Welding Journal (89): 232-242.
11. Cook G E, Maxwell J E, Barnett R J, Strauss A M (1997) Statistical Process Control Application to Weld
Process. IEEE Transactions and Industry Applications (33): 454-463.
12. Praveen P, Kang M J, Yarlagadda P K D V (2009) Arc voltage behavior in GMAW-P under different drop
transfer modes. Achievements in Materials and Manufacturing Engineering 32(2): 196–202.
13. Simpson S W (2008) Fault identification in gas metal arc welding with signature images. Science and
Technology of Welding and Joining (13): 87–96.
14. Simpson S W (2007) Signature image for arc welding fault detection, Science and Technology of Welding and
Joining (12): 481-486.
15. Simpson S W (2007) Statistics of signature images for arc welding fault detection. Science and Technology of
Welding and Joining (12): 556-563.
16. Simpson S W (2008) Signature image stability and metal transfer in gas metal arc welding. Science and
Technology of Welding and Joining (13): 176–183.
17. Simpson S W (2008) Through arc sensing in gas metal arc welding with signature images. Science and
Technologyof Welding and Joining (13): 80-86.
18. Pal S, Pal S K, Samantray A K (2010) Prediction of the quality of pulsed metal inert gas welding using
statistical parameters of arc signals in artificial neural network. International Journal of Computer Integerated
Manufacturing (23): 453-465.
19. Wu C S, Hu Q X, Sun S, Polte T, Rehfeldt D (2004) Intelligent monitoring and recognition of the short-
circuiting gas–metal arc welding process. Proceedings of the Institutions of Mechanical Engineers., Part B, J.
Eng. Manufacturing (218): 1145–1151.
20. Wu C S, Polte T, Rehfeldt D (2001) A fuzzy logic system for process monitoring and quality evaluation in
GMAW. Weld. Journal 80 (2):16–22.
21. Wu C S, Polte T, Rehfeldt D (2000) Gas metal arc welding process monitoring and quality evaluation using
neural networks. Sci. Technol. Weld. Joining 5(5): 324–328.
22. Pal S, Pal S K, Samantaray A K (2008) Neurowavelet packet analysis based on current signature for weld joint
strength prediction in pulsed metal inert gas welding process. Sci. Technol. Weld. Joining (13): 638–645.
23. Di Li, Yonglun Song, Feng Ye (2000) On Line Monitoring of Weld Defects for Short circuit Gas Metal Arc
Welding Based on the Self-Organize Feature Map Neural Networks. Proceedings of the IEEE-INNS-ENNS
International Joint Conference on Neural Networks (IJCNN'00).
24. Di Li, Yonglun Song, Y and Ye, Xuebao FHanjie (2000) Identification of Weld Defects in MAW Based on
Arc Sensing. Transactions of the China Welding Institute (21): 30-33.
25. Moore K L, Yender R, Tyler J, Naidu D S (1998) Modeling, Calibration and control, Theoretical Analysis of
the GMAW Process. Proceedings of the American Control Conference, Philadelphia, Pennsylvania.
26. Oppenheim Alan V, Willsky Alan S, Hamid S (1996) Signals and Systems. Prentice Hall.
27. Matz V, Kreidl M, md R (2004) Signal-to-Noise Ratio Improvement based on the Discrete Wavelet Transform
in Ultrasonic Defectoscopy. Acta Polytechnica44 (4).
28. Haiyun LIU, Hui LI, Zhuoxin LI, Yaowu SHI (2010) Experimental Study on Metal Transfer and Welding
Spatter Characteristics of Cellulose Electrode. Chinese journal of mechanical engineering (23): 1-4.
29. Welding hand book, American welding society, 9th edition, volume 2.