Emission spectrometry evaluation in arc welding monitoring system

Journal of Materials Processing Technology 179 (2006) 219–224

Emission spectrometry evaluation in arc welding monitoring system Sadek C.A. Alfaro ∗ , Diogo de S. Mendonc¸a, Marcelo S. Matos The University of Brasilia, ENM-FT, UnB, Campus Universit´ario 70.910-900, Brasilia, DF, Brazil

Abstract This work describes exploratory experimental procedures implemented for the development of a non-intrusive and real-time sensor for weld defect tracking which uses emission spectrometry for measuring the electromagnetic content of the plasma-weld pool interface in the GMA welding arc. The welding process monitoring is carried out by calculating the iron (Fe) and the manganese (Mn) electronic temperatures within the welding arc column, admitting that the observed region is at local thermodynamic equilibrium. The temperature was calculated by utilising the relative intensity method, which is based on the Boltzmann and the Saha Laws and on the definition of the emission line intensity. The calculated electronic temperatures of the two elements were correlated with the position of welding defects, which have been introduced for simulation purposes. These simulated defects resulted in abrupt changes in the average and standard deviation temperature values, thus providing an indication of the presence of a defect. © 2006 Elsevier B.V. All rights reserved. Keywords: Spectrometry; Welding; Plasma; Welding monitoring

1. Introduction For many years, several different monitoring techniques have been studied, attempting to develop a general technique capable of dealing with the inherent complexity of the arc welding processes. Such studies aimed mainly at developing methods of quality of the welds on-line controlling, in order to prevent the needs for the costly and time consuming post weld inspection processes. The innovations generated by these studies are based on the physical phenomena involved in the arc welding processes, mainly those related to the plasma arc and its influence on the weld pool [1]. The applied techniques range from numerical simulation of the arc [2], image analysis [3], sound spectrum analysis and electromagnetic emission analysis [4–6] to the use of intelligent systems, based on neural networks and fuzzy logic [7]. This work describes a study on the possibility of utilising the electromagnetic emission of the arc welding plasma column for monitoring the presence of weld defects. The proposed method uses the electronic temperature, calculated from the intensities of the emission lines in the electromagnetic spectrum within the visible region, as an indicator of change in the expected quality. Some specially chosen emission lines can also give indication

of contamination of the weld bead with hydrogen. This makes it possible for a control system to act before a deleterious effect can affect permanently the quality of a weld. 1.1. Emission spectrometry and plasma characterisation The term “Spectrometry” stands for a set of experimental techniques used for measuring the electromagnetic spectrum that results from phenomena such as absorption, emission or diffraction of electromagnetic radiation by atoms or molecules. These techniques are generally analytical and, as such, may produce relevant data for the analysis of welding processes [8]. According to the Quantum Theory, atoms and molecules can only exist in a steady energy states, which are characterised by discrete amounts of energy that are specific to each atom or molecule. When a change of energy state occurs in an atom or a molecule, their electrons absorbs or emits the specific amount of energy, which is strictly necessary for taking it from one energy state to another. Such change of energy state is generally accompanied by emission or absorption of light, which wave length, λ, is related to the energy of both states [9], according to the Eq. (1) Ei − E n =

∗

Corresponding author. E-mail addresses: [email protected] (S.C.A. Alfaro), [email protected] (D. de S. Mendonc¸a), [email protected] (M.S. Matos). 0924-0136/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.jmatprotec.2006.03.088

hc λ

(1)

in which Ei is the energy in the lower state, En , the energy in the higher state, h, the Planck constant and c is the light speed.

220

S.C.A. Alfaro et al. / Journal of Materials Processing Technology 179 (2006) 219–224

The emission spectrometry is generally used as a means of identifying the electromagnetic radiation wavelengths emitted by atoms or molecules of a substance due to a change on their energy state. Such wavelengths provide information on thermodynamic and quantum parameters that are necessary for calculating properties of interest. In the specific case of the welding process, a qualitative analysis (identification of chemical elements) allows the detection of weld contamination [10], helps in the selection and qualification of shielding gases [11] and provides a means of studying the rate of dilution in dissimilar metal welding [4]. On the other hand, the quantitative analysis involves the measurement of the intensities of each different wave length emitted by the plasma radiation and provides a means of calculating the most important plasma properties: its electronic temperature and its density. 1.2. Calculation of the plasma temperature The plasma temperature is calculated from the temperature of the electrons (kinetic temperature) admitting the validity of the local thermodynamic equilibrium (LTE) hypothesis. This means that the particles have an energy distribution given by the Maxwell equation and that the collision processes are dominant relative to the radiation processes, i.e., the temperature of the electrons is similar to the temperature of the heavy particles [12]. The LTE hypothesis can be verified by means of applying the Eq. (2) [12], in which Ne is the electronic density [m−3 ], Te the electron absolute temperature [K] and E is the difference of the transition energy intervals [eV]. Ne ≥ 1.6 × 1012




Te (E)3

(2)

Studying several works dealing with the LTE hypothesis, Vilarinho [13] concluded that its validity is restricted to the centre of the electric arc column. Such a conclusion defines the region to be observed by the optical equipment used for the emission spectrometry. The typical value of the electronic density (Eq. (2)) in the GMAW process, calculated by Lacroix et al. [4], is Ne ≥ 1.81 × 1019 m−3 . This result is used in the present work. In the case of the LTE hypothesis being valid, the Maxwell, the Boltzmann and the Saha laws can be applied; the Maxwell and the Boltzmann laws being used for the plasma temperature calculation and the Saha law, for the calculation of its electronic density [14]. The temperature of the electrons within the plasma column is obtained by using the relative intensity of several emission lines [12]. According to Eq. (3), the emission line intensity for a transition from state “m” to state “n”, Imn , depends on the transition probability, Amn (in units of s−1 ), on the density of the higher level of energy, Nm (in units of m−3 ), on the Planck’s constant, h (in units of J s) and on the frequency, νmn (in s−1 ). Imn = Nm Amn hνmn

(3)

The Boltzmann law can be stated as in Eq. (4)     N −Em gm exp Nm = Z(t) kT

(4)

in which N is the total density of the energy level, gm the statistic weight, Z(t) the partition function, k the Boltzmann constant (k = 8.6173 × 10−5 eV K−1 ), Em the energy in the higher level and T is the absolute temperature. Applying Eqs. (4) into (3) and taking into account the relation between the wave length, the electromagnetic wave travel speed (the light speed, c) and the frequency, with some algebraic manipulation it is possible to obtain Eq. (5).     Imn λmn Nhc Em ln = ln − (5) Amn gm Z kT Eq. (5) can be viewed as a first order polynomial in the independent variable “Em ”, assuming that the absolute temperature, T, and the term “ln(Nhc/Z)” are constant. Consequently, the left side of Eq. (5) could be considered as a dependent variable of the polynomial. Therefore, if it is plotted as a function of the independent variable (the energy of the higher level Em ) for a set of emission lines, for which the values of Amn and gm are available (National Institute of Standards and Technology (NIST) http://physics.nist.gov/cgi-bin/AtData/lines form), the angular coefficient of the resulting linear regression line would be a good estimate for the term“−1/kT”, from which the corresponding absolute temperature could be calculated [4]. The temperature of the electrons may also be estimated by using the ratio of the relative intensities of two emission lines (indices 1 and 2 in Eq. (6)) of the same chemical element, as proposed by [15]. This method takes account of the changing density distribution along the welding plasma column and does not require the calculation of the time-consuming, spatially resolved lateral line intensity measurements in order to calculate the inverse Abel Transform [16]. Since this method involves recording the total spatially integrated radiation line intensities over the plasma diameter, it was obtained only the temperature at the centre of the plasma column [15], using a very simple and straightforward experimental setup. Te ≈

Em(1) − Em (6) k ln[Em(1) gm(2) I(1) Amn(2) λ1 /Em(2) gm(1) I(2) Amn(1) λ2 ]

in which I(i), i = 1 and 2, are the relative intensities of two emission lines, as acquired by the measurement system, and fmn is the damping force of the transition m–n, which is tabulated. This method is better than the previous in a sense of implementing a real time monitoring system. The emission lines to be selected must satisfy the condition Em(1) − Em(2) > kT in higher energy levels [5]. 2. Experimental procedure In order to experimentally evaluate the theory develop in the previous section, an experimental apparatus was implemented. A welding robot, which was responsible for moving the GMAW torch along a pre-defined welding path, together with a support for the optical equipment (collimating lens, with focal distance 11 mm, and one end of the 2 mm core diameter optical fiber) were used for producing the weld beads from which the data was to be acquired. The other

S.C.A. Alfaro et al. / Journal of Materials Processing Technology 179 (2006) 219–224

221

Fig. 3. Spectrum emitted by welding MAG process on a steel plate SAE 1020. (1) Fe I 561.564, (2) Fe II 537.149, (3) Mn I 482.352 and (4) Mn II 403.306. Fig. 1. Experimental setup.

Table 1 Welding parameters Welding MAG

end of the optical fibre was coupled to a compact CCD spectrometer (SM240USB, resolution of 0.3 nm within the 300–575 nm interval, diffraction grid of 1200 g/mm at 400 nm and shutter of 10 ␮m). All the acquisition procedure was automated by using a specially developed MatLab® routine, which functions were to control the spectrometer and a data acquisition board. An analogue input of this latter was connected to a robot digital output to allow the synchronisation of the spectrum acquisition time with the movement of the robot. Fig. 1 shows the schematic drawings of the welding apparatus. Before acquiring any data from the GMAW process, it was necessary to verify the calibration of the spectrometer. This was accomplished by collecting data from an argon shielded GTAW arc opened on a water-cooled copper base. Since in this case it can be considered that there is neither fusion nor vaporization of either the base metal or the tungsten electrode, it was expected that only the argon emission lines would be present in the spectrum acquired. Such hypotheses were confirmed in the resulting spectrum (Fig. 2). This procedure was also adopted to make it easier to identify the emission lines of elements, other than

Welding position Specimen (mm) Gas Current (A) Voltage (V) Welding speed (mm/s) “Stick-out” (mm) Gas flow (l/min) Diameter of wire (mm)

Bead on plate 300 × 150 × 6.5 AISI 1020 F34 (66% Ar, 34% CO2 ) 140, 150, 200 18, 24, 26 10 20 10 0.8

the argon, present in the arc of a real weld process, thus allowing the selection of those which would be monitored for temperature calculation purposes. The emission spectra (Fig. 3) from some bead on plate GMA welds carried out by using an inverter based welding power source with the parameters shown in Table 1 were then acquired and compared with the Argon spectrum in order to allow choosing the elements that would be monitored. The authors’ first choice was the iron (Fe) due to the fact that it is the base element in the steel alloy. The second element chosen was the manganese (Mn) due to higher concentration in both the welding wire and the base metal, relative to the other elements, and also because the emissions of both elements (Fe and Mn) are not auto-absorbed in the plasma column [8]. The chosen lines are shown in Table 2. Once the monitored emission lines were defined and a procedure for calculating the temperature was established (Boltzman method) to validate the experimental setup, a monitoring system was implemented by means of a MatLab® routine, which allowed the automation of the electronic temperature data acquisition using Eq. (6). Considering that some calculations were necTable 2 Fe and Mn emission line wavelengths for monitoring

(1) (2) (3) (4) Fig. 2. Spectrum of the TIG process on water-cooled copper plate.

Elementa

Wavelength (nm)

Fe I Fe II Mn I Mn II

561.5644 537.1490 482.3524 403.3062

a The index corresponds at the ionization level: I, fundamental ionization and II, first ionization, etc.

222

S.C.A. Alfaro et al. / Journal of Materials Processing Technology 179 (2006) 219–224

Fig. 5. Behaviour of the electronic temperature along a welding run without defects.

Fig. 4. Boltzman’s graphic for the calculus of the electronic temperature on Fe and Mn in welding MAG: 118 A, 24 V. Table 3 Comparison between the Fe and Mn temperatures obtained in this work and the temperatures from other authors Work

Temperature Fe I (K)

Temperature Mn I (K)

Ancona et al. Lacroix et al. Vilarinho Experiments

9400 11400 7650 7145

11700 – – 18000

ment of defects. Fig. 5 shows the behaviour of the electronic temperature signal (eV) calculated for the iron and manganese emission lines for a welding bead without defects, while Fig. 6 shows the behaviour of the temperature signal along a weld bead, in which some defects were induced. Other defects were simulated (base metal distortion, causing variation in the stand-off and lack of fusion) and similar results were obtained (sample rate of 12 Hz). In all cases, the deviation of an average temperature is evident in the points where the defects are presents in. In order to establish an algorithm to proceed in the identification of those abrupt changes in the electron temperature signal, a comparison between its mean and standard deviations and faultless (sample 1) and defective

essary to be carried out after the acquisition of each emission spectrum, the resulting sampling frequency for the monitoring system was 45 Hz, which corresponds to a weld length of 0.22 mm at a welding speed of 10 mm/s between temperature samples. The monitoring setup was validated by acquiring spectrum data from a GMA bead-on-plate weld using optimum parameters (welding voltage = 24 V; welding current = 118 A; stick-out = 20 mm; Mild steel 0.8 mm diameter wire; 66% Ar + 34% CO2 shielding gas at 10 l/min) and by calculating electronic temperatures for the elements Fe and Mn (Fig. 4). The calculated temperatures were compared with the results obtained by literatures [4,5,13] and they are shown in Table 3. It is important to highlight that such a comparison was qualitative since the welding parameters used in each case were different. The difference observed in the plasma temperatures obtained by using the emission lines of the Fe and the Mn (Fig. 4) was expected, since the temperature distribution and the presence of metallic vapours in the plasma are not uniform [13]. While the monitoring system was tested by using of several “defect-free” bead-on-plate welds were produced using welding parameters defined as “optimum” by a qualify welder. The temperatures calculated from these welds were used for comparison with the results from welds with induced defects, which were produced by using inadequate welding parameters.

3. Results and discussion The monitoring system developed was set to acquire spectrum information ranging from the ultraviolet to the near infrared (320–575 nm) aiming at the interface between the arc and the weld pool, thus involving parts of both. The calculated temperature values were then average value of the aimed spot and its vicinity. The following results show clearly that there is correlation between the electronic temperature behaviour and the develop-

Fig. 6. Behaviour of the electronic temperature along the welding bead run with defects. (a) Position (A): grease contamination; position (B): oxidation; position (C): slag inclusion. (b) Position (A): metal inclusion (tungsten); position (B): induced porosity; position (C): lack of penetration.

S.C.A. Alfaro et al. / Journal of Materials Processing Technology 179 (2006) 219–224

223

Fig. 7. Comparison between mean and standard deviations in an electron temperature (1, faultless welds and 2, defective welds).

(sample 2) weld seams, was carried out. It is possible to observe that under presence of discontinuities the mean electron temperature is less sensitive than the standard deviation response to these defects (Fig. 7). Therefore, the standard deviation stands as a valuable parameter in a fault detection algorithm. From this point of view, it was utilized an algorithm defined in terms of a cumulative sum of the standard values of the temperature signal (σ). Base on this evidence, it was adopted for the real time monitoring the calculation of “standards deviation accumulated” (Σ), or either, the value of Σn (standard deviation in ‘n’ sampled points) that it is equal to the value of σ1n (electronic temperature standard deviation of the points 1 to ‘n’). Fig. 8 shows the value ‘Σ’ of electronic temperatures in weld beads with and without

defects. The red line (Σt) is the standard deviation of registered temperature values. The regions delimited for the vertical line, indicated for the ‘A’ letter, correspond to the point of the welding arc ignition showing high noise and ignored for the monitoring routine purpose. The top graph shows the typical behaviour of Σ in a welding without defects. The accumulated standard deviation values do not exceed the standard deviation value of the acquired signal, on the other hand, in the bottom graph can be seen for some points, the values exceed the line of Σt and here it also was used the value of Σa (average of accumulated standard deviation) for calculation in the monitoring routine. The “Kalman” filter, supplies the optimum value of Σa, shown as Σp in Fig. 8. The monitoring routine uses this standard deviation behaviour to, from the estimate gathering with the Kalman filter calculate the points that are exceeding the values of Σp and here infer the occurrence of weld defects. The arrows in the bottom graph are the output information of the monitoring routine, indicating the position of the finding defects. 4. Conclusion

Fig. 8. Behavior of the standard deviation values in electron temperature samples.

The results show that there is a good correlation between the change in the average electronic temperature, acquired from the vicinity of the arc-weld pool interface, and the development of defects during the GMAW process. Based on this observation, it is possible to conclude that the electronic temperature may be used as an indication of a change of state during the process. This could be applied to monitoring systems to identify locations in the weld with high possibility of having weld defects and, with further studies, it could give indication of what kind of defect is forming and allow a control system to act before the defect can compromise the whole weld.

224

S.C.A. Alfaro et al. / Journal of Materials Processing Technology 179 (2006) 219–224

The main characteristics of the proposed system are: • Compact: the spectrometer and the focalization system do not weigh more than 350 g and its volume is quite reduced if compared with the traditional spectrometers. • Several types of defects could be identified as shown in Fig. 6. • The system is non intrusive and it can be used with any base metal with known metallurgical composition. Although the linear Kalman estimator had presented some satisfactory results, it was a first assumption in the development of robust algorithm of defects detection, since it presents some limitations due to its time consuming and a high-rate of false alarms in such approach. A new effort has been carried to obtain a recursive algorithm based on the statistical probability ratio test. Other limitation of this algorithm is the maximum sampling rate, 45 Hz which is low considering the fastest welding velocities. Acknowledgments This work was performed under the auspices of Brazilian Research Councils (CNPq and Capes) and sponsored by the Technology Foundation FINATEC. References [1] D. Degout, A. Catherinot, Spectroscopic analysis of the shapes created by double-flux Tungsten Inert G´as (TIG) arc shapes torch, J. Phys. D: Appl. Phys 19 (1986) 811–823.

[2] J. Haidar, J.J. Lowke, Predictions of metal droplet formation in arc welding, J. Phys. D: Appl. Phys. 30 (1996) 94. [3] W. Kim, C. Allemand, T.W. Eagar, Visible light emissions during gas tungsten arc welding and its application to weld image improvement, Weld. J. (1987) 369s–377s. [4] D. Lacroix, C. Boudot, G. Jeandel, Spectroscopic studies of GTA welding plasmas. Temperature calculation and dilution measurement, Eur. Phys. J. AP8 (1999) 61–69. [5] A. Ancona, V. Spagnolo, M.P. Lugar`a, Optical sensor goes real-team monitoring of CO2 laser welding process, Appl. Optics 40 (No. 33) (2001). [6] P. Sforza, D. Blasiis, On-line optical monitoring system goes arc welding, Int. NDT&E 35 (2002) 37–43. [7] H.S. Moon, Neuro-fuzzy approach to select welding conditions goes welding quality improvement horizontal in fillet welding, J. Manufact. Syst. 15 (NO. 6) (1996). [8] J.M. Hollas, Modern Spectroscopy, third ed., 1996. [9] Masterton, Slowinski, Stantski, Beginnings of Chemistry, Ed. LTC, Chapter 7, sixth ed., 1996. [10] M. Onsoien, R. Peters, D.L. Olson, S. Liu, Effect of hydrogen in argon GTAW shielding gas: arc characteristics and bead morphology, Weld. J. (1995) 10s–15s. [11] J. Tusek, M. Suban, Experimental research of the effect of hydrogen in argon the shielding gas in arc welding of high-alloy stainless steel, Int. J. Hydrogen Energy 25 (2000) 369–376. [12] H.R. Griem, Shapes Spectroscopy, McGraw-Hill, New York, 1964, Part 13 and 14. [13] L.O. Vilarinho, Development of Experimental and Numerical Techniques goes TIG Arc Characterisation, Theory of Doctorate, Federal University of Uberlandia, 2003. [14] F.F. Chen, Introduction to Shapes Physics, Plenum Press, New York, 1977. [15] A. Marotta, Determination Axial of Thermal Shapes Temperatures without Abel Inversion”, J. Phys. D: Appl. Phys. 27 (1994) 268– 272. [16] G. Arfken, Mathematical Methods for Physicists, third ed., Academic Press, Orlando, FL, 1985, pp. 875–876.