Journal of Materials Processing Technology 172 (2006) 4251

Estimation of heat source and thermal efciency in GTAW process by using inverse techniques
C.V. Goncalves, L.O. Vilarinho, A. Scotti, G. Guimar es a
Federal University of Uberl ndia, School of Mechanical Engineering, Campus Santa M nica, a o Bloco M, Av. Jo o Naves de Avila 2121, 38400-902 Uberl ndia, MG, Brazil a a Received 6 January 2005; received in revised form 6 January 2005; accepted 31 August 2005

Abstract This work presents a comparison between two techniques to study thermal phenomenon that occurs in the base metal during welding operations. The techniques combine optimization methods, such as the simulated annealing (SA) and gold section, in two different physical models. The rst thermal model considers a quasi-stationary heat conduction in the interior plate, whereas the second one uses the general transient equation of heat diffusion with phase change. In both cases, the heat ux generated by the welding process is estimated, and, from this point, global thermal efciency and fusion efciency are also estimated. Moreover, the use of the SA method with the numerical model allows the weld pool geometry identication. In this study, thermocouples are placed on the plate (AISI304) at the opposite side concerning a weld carried out by GTAW (tungsten inert gas). It is concluded that the phase change model demonstrate to be able of identifying the global thermal efciency and fusion efciency at each instant of the process. 2005 Elsevier B.V. All rights reserved.
Keywords: GTAW process; Phase change; Welding; temperature; Optimization; Identication

1. Introduction Welding processes are employed in a large number of structures such as buses, airplanes, power stations or oil pipes. Since these structures require high level of safety, the manufacturing processes, must be carefully considered and are effective control over their performance must be reached. Gas tungsten arc welding (GTAW) is one of the most widely employed welding processes that is applied with success for welding of stainless steels and non-ferrous materials. In this process, a tungsten electrode is shielded by a ow of inert and/or reducing gases such as argon (normally employed), helium, nitrogen, hydrogen or blends. The union of two or more workpieces is obtained through the heat from a voltaic arc, which is a moving intense heat source. The analysis of the thermal behavior of the physical phenomenon that takes place in the process is crucial to understand, for example, weld geometry formation, microstructure changes in the thermally affected base metal and residual stress distribution in the joint. Knowledge of the heat input intensity and the

Corresponding author. Tel.: +55 34 32394415; fax: +55 34 32394206. E-mail address: gguima@mecanica.ufu.br (G. Guimar es). a

temperature gradients in the workpiece are extremely important for that welding process studies. Also, the ratio between the effective heat delivered to the workpiece and electric power consumed is a good indicative of the process performance. It is well known that only part of the total electrical power (voltage versus current) converted into heat is absorbed by the workpiece. The difference is due to the heat losses to the environment by convection or radiation and conducting from the electrode to the gun. Thus, the task is how to precisely estimate the heat input to the workpiece and, consequently, the thermal efciency. From literature, it is possible to delimitate two lines of research: one which the welding is carried out on a water cooled anode (without phase change) and the other which weldments are made in practice conditions, i.e., with phase change. Results published so far indicate that these two approaches of heat input calculation lead to different results. The cooled anode approach provides values of thermal efciency of the order of 80%, whereas thermal efciency from actual weldments is of the order of 60%. This difference (20%) may be due to the latent heat, i.e., the fraction of the heat input that is not converted in temperature increase, but only in phase change. Whatever the calculation method employed by the several authors or the approach that take or not into account the phase

0924-0136/$ see front matter 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.jmatprotec.2005.08.010

C.V. Gon alves et al. / Journal of Materials Processing Technology 172 (2006) 4251 c

43

change, the validation of the results faced the problem on how to measure temperature. It is very difcult and subjected to error to measure temperature into the pool (arpon technique) or close to it. Alternatively, the inverse heat conduction problem represents, in this case, a way of obtaining the heat ux that goes to the workpiece. This methodology is justied by the difculties to obtain temperature measurements in the weld area by other means. The great advantage of the inverse technique is the chance of obtaining the heat ux imposed in the weld face of the workpiece using just temperatures measured far away from the weld bead. One requisite to determine the temperature eld under welding processes is to know either the actual heat ux or the geometry and interface position of weld pool. Once these parameters are known, a direct problem is established and the temperature eld can, then, be calculated. Several pieces of work, analytical [16] or numerical [710], that deals with this direct problem can be found in the literature. Besides the hypothesis of knowing the heat ux, important phenomena like metallic phase changes, thermal properties dependence on the temperature or heat losses by convection or radiation are neglected or considered known [1115]. However, the identications of heat ux or the weld pool geometry and position are not trivial in a real welding. The main difculty, in this case, is that these parameters are not reliable available or they are not so easily measured directly. As mentioned, a mean of overcoming this difculty is to use inverse techniques. However, the majority of work that deal with inverse problem techniques in welding problems [1618] just use simulated studies. Thus, the main goal of this work is to contribute to the welding related studies on heat ux and temperature eld by presenting a solution based on inverse technique applied to the GTAW process. 2. Physical models for GTA welding process In order to develop the inverse algorithm, two theoretical models are used to solve the direct problem: a simplied twodimensional physical model based on quasi-stationary Rosenthals equation [1] and a transient two-dimensional physical model based on the work of Al-Khalidy [15]. In contrast to the model 1, the transient model takes into account metallic phase changes, thermal properties dependence with temperature and heat losses by convection and radiation. 2.1. Rosenthals model (model 1) Rosenthal proposed solutions to the heat ux equation of a moving punctual heat-source in 1941 [1]. In his model, it is assumed that the energy of the heat source moves with a constant speed u along the x-axis of a xed rectangular co-ordinate system, as shown in Fig. 1. Considering Fig. 1, the general energy equation for an isotropic and homogenous material, with constant properties and negligible heat losses, is reduced to 2 T 2 T 2 T 1 T + 2 + 2 = 2 x y z t (1)

Fig. 1. A moving point heat source: (a) xed coordinates x, y, z and (b) moving coordinates , y, z.

where is the thermal diffusivity and x, y, z are the xed coordinate systems. A moving punctual heat source (q) traveling with a constant velocity (u) in the x direction (Fig. 1) is assigned to the origin of a moving coordinate systems , y, z. Thus, the transformation of a point P(x,y,z,t) in the xed system by = x ut becomes P(,y,z) in the moving system. Then Eq. (1) becomes: 2 T 2 T 2 T + 2 + 2 = 2 y z 1 T t u T x (2)

Assuming that the workpiece is long enough, in such a way that the quasi-stationary condition can be established, Eq. (2) reduces to 2 T (, y, z) 2 T (, y, z) 2 T (, y, z) u T (, y, z) + + = 2 2 2 y z (3) Eq. (3) can be solved using the separation of variables method, as follows: T (, y, z) = T0 + u q exp 2k 2 K0 ur 2 (4)

where K0 is the modied Bessel function of the rst kind of order zero and r = ( 2 + y2 + z2 )1/2 . The assumptions involved with Eq. (4) include a quasi-steady condition in the workpiece, the voltaic arc being represented by a punctual source, the thermal properties being considered constant with temperature and the heat losses from the plate due to convection or radiation and the metallic phase changes being neglected. If the value of the heat ux, q, is known, Eq. (4) represent the direct problem solution related to the inverse problem studied. 2.2. Physical model considering phase changes (model 2) Transient heat transfer problems involving melting or solidication, as the problem due to a GTA welding process, are usually referred as phase-change or moving-boundary problems. The solution of such problems is inherently difcult because the interface between the solid and liquid phases is moving as the latent heat is absorbed or released at the interface. As result, the location of the solidliquid interface is not known a priory and it

44

C.V. Gon alves et al. / Journal of Materials Processing Technology 172 (2006) 4251 c

must follow as a part of the solution [24]. In the welding process, the thermal problem is, then, established when the molten pool is formed as the result of the heat source moving with a constant speed. The physical model to solve the direct problem related to phase change presented here is based on a previous work of Al-Khalidy [15] that considers a two-dimensional system and a nonlinear thermal properties temperature dependence in the solid region. The theoretical model developed in Ref. [15] can, then, be given by the numerical solution of the governing equations: C() (, y, t) (, y, t) + C()u t h() (, y, t) + Q()m in R (5a)

3.1. Heat ux, q, estimation by using simulated annealing techniques (model 1) One way of estimating the heat ux q, present in Eq. (4), is to require that the traditional least square function, Fq , dened as the error between the computed temperature Ti and the measured temperature Yi be minimized with respect to the unknown q. Fq is, then, dened as
M

Fq =
i=1

(Yi (, y) Ti (, y, q))2

(6)

= k()(, y, t)

where (,y) Rs Rl and R is the region of pure material split in a liquid region Rl and a solid region Rs . The two regions are separated by an interface R. Eq. (5a) is subjected to the boundary conditions: l = 0, y s (, y, t) = 0 s = 0, y u = 0, y at y = 0 (5b) (5c)

at y

and to an initial condition: (, y, 0) = 0 at , y R (5d)

In Eq. (5a), the phase change is isolated in a latent heat source term as Q()m = H() t (5e)

where the h() is the overall convective heat transfer, which is calculated as h() = hc + hr , where hr and hc are the radiative and the convective heat transfer coefcient, respectively. Additionally, the interface condition includes the isothermal condition: m (x, y, t) = Tm T at R (5f)

where i represents the index for the thermocouple position. There are several inverse techniques besides of the SA that could solve this welding problem optimization. It can be cited, for example, the conjugate gradient with adjoint equation method [22], parameter estimation approach [23] and sequential time domain [24]. These methods uses temperature histories experimentally determined in the sample (workpiece) to calculate the corresponding input heat ux for a given set of system parameters (welding). The simulated annealing was chosen here due, mainly, to its robust characteristic [20]. It can be performed in optimization by randomly perturbing the decision variable and keeping track of the best objective function value for each randomized set of variables. After many trials, the set that produced the best objective function value is assigned to be the center, over which perturbation will take place for the next temperature. The temperature, that in this technique is the standard deviation of the random number generator, is then reduced, and new trials performed. In related to Eq. (6), temperature Ti are calculated from Eq. (4), and Yi represents the experimental temperature measured, in this present case, on the opposite face to the weld bead. The iterative computational procedure to estimate the heat ux q can be summarized as follow: Step 1: To start the iterations, an initial estimate is made for the heat ux q, which can be chosen as constant, say close to zero. Step 2: Solve the direct problem given by Eq. (4) and compute T(,y). Step 3: Assuming that the search region to the optimal value of heat ux, q, is between a lower bound zero and a upper bound given by the power P supplied by the GTA gun (voltage versus current), the optimization process of Fq goes until the optimized value of q is found (or a stopping criterion is satised). Here, the criterion is Fq < 1 K2 . 3.2. Molten pool radius and interface position estimation by using gold section method (model 2) The use of model 2 requires a new objective function. In this case, since the model is transient and the location of the phasechange is unknown, the new objective function is dened as
N

If the parameters Q()m and the weld pool geometry are known, the numerical solution of Eq. (5) represent the direct problem related to the inverse problem studied. For obtaining the numerical solution of Eq. (5), the nite volume method [19] is used with a grid of 360 90 volumes. 3. Inverse procedure In addition to the experimental data and the physical models, two techniques of optimization are used to solve the inverse algorithm. The techniques are the simulated annealing, SA [20], and the golden section method [21], respectively. For models 1 and 2 both, physical models and optimization techniques are used over the same experimental temperature data, what allows a fair comparison of results.

Fm =
i=1

(Yi (, y, t) Ti (, y, t, r))2

(7)

C.V. Gon alves et al. / Journal of Materials Processing Technology 172 (2006) 4251 c

45

where i represents the time index, r the molten pool radius, and T is the temperature that now is calculated from the numerical solution of Eq. (5). The inverse procedure presented by Al-Khalidy [16] for a quasi-static model is adapted here to the thermal transient model. On the contrary of that work, here the molten pool shape is supposed to be circular and the objective of the inverse procedure is to estimate the radius of the weld pool as well as the temperature eld within the solid region. The inverse procedure, here, is stated as an optimization problem to minimize the function Fm , Eq. (7). The iterative computational procedure to estimate the radius, r, can be, then, summarized as follow: Step 1: Assume an arbitrary value to the radius r of the weld pool. Step 2: Assume initial values to H(,y,t) = H0 (,y,t) = 0. Step 3: The nonlinear simultaneous set of equations, Eq. (5) is solved to determine T(,y,t). Step 4: Optimize the results with respect to Eq. (7). At this step, the nite volumes in where the phase change occurs are identied using the optimal r value. Step 5: Steps 3 and 4 are repeated up to convergence is reached. Step 6: The known values of the temperatures T(,y,t) and H(,y,t) at any time is assumed to be the previous value in the next time step. Step 7: The calculation procedure for the new time step is repeated. Since the optimization procedure needs to obtain, many times, the numerical solution of Eq. (5), the use of simulated annealing technique here could dispended so much time of computing. Instead, as we have only one variable to estimate, the radius r, the golden section method is chosen to minimize Eq. (5). The golden section method for estimating the minimum of a one-variable function is a popular technique since it does not need to have continuous derivatives, it is easily programmed and it is found to be reliable for poorly conditioned problems. See Ref. [21] for more details.
4. Experimental procedure 4.1. Experimental rig and welding setup
Fig. 2 represents the welding rig used in the experimental procedure. The welding gun, which represents the point heat source, moves at a specic speed along a straight path by using a totally automated XY coordinate table. To avoid dimension interference, the test-plate is held in air by four-end pointed cylindrical screws, so that, only a small contact area exists. Details of these screws can be seen in Figs. 2 and 3. A set of experiments was carried out on AISI304 stainless steel test plates with dimensions of 200 mm 50 mm 4 mm, using the following welding setting conditions: direct polarity (CC), current at 78 A, arc length (distance from the electrode top to the plate surface) adjusted to 4 mm, shielding gas (Ar and Ar + 25% He). The electrode used was a 1.6 mm of diameter, AWS EWTh-2 (tungsten doped with 2% of thory) with an included angle of 60 . The shielding gas ow was set at 8 l/min (0.133 103 m3 /s) and the welding speed at 0.00833 m/s. The resultant arc voltage was monitored around 15 V. The values of the thermal properties of the AISI304 as a function of temperature were obtained from literature [15]:

Fig. 2. Experimental rig.

k(W/mK) = 14.42 + 0.0169T 2.44 106 T 2 , cp (W/kg K) = 486.6 + 0.159T + 18.07 106 T 2 and h = 1.32 T
0.25

and

Tm (K) = 1700

+ 0.61

0.2

2 + (5.67 108 )0.95(T 2 + T )(T T )

(8a) if Gr Pr = 109 , or h = 1.32 T

0.25

+ 1.43

0.333

2 + (5.67 108 )0.95(T 2 + T )(T T )

(8b) if Gr Pr > 109 , where Gr and Pr are the Grashof and Prandtl number, respectively. Capacitor discharge is used to attach the thermocouples to the plate surface. Four thermocouples, type K, are attached to the bottom face of the test-plate (AISI304) (Fig. 3). A well-detailed description of this procedure can be found in Vilarinho [25]. Fig. 4 presents in detail the thermocouple location in the sample of these four thermocouples. The welding is carried out on a at position, using the bead on plate technique. Two independent data acquisition systems are used, in the rst one, the voltage and current signals are logged at 12 bits and 10 kHz per channel. The second one, composed by a microcomputer-based data acquisition system (HP 75000 B E1326B), DAS for short, is used to acquire and store the thermocouple data. The DAS, under a software control, sampled (multiplexed) each thermocouple signal at intervals of 0.38 s (totaling 1028 points for each thermocouple).

Fig. 3. View of the thermocouple attachment under the test-plate.

46

C.V. Gon alves et al. / Journal of Materials Processing Technology 172 (2006) 4251 c

Fig. 4. Thermocouple locations at opposite face of the test-plate.

Fig. 7. Measured temperatures on bottom surface of the test-plate from the four thermocouples shown in Fig. 4. enough, at fusion level. Since the model is three-dimensional, any variation along the thickness can be felt. Fig. 6 shows the behavior of the temperature at the opposite face in the region towards the symmetry line and away 0.008 from this line. It can be seen that 8 mm is sufcient to obtain the same temperature at any plane along the thickness.

Fig. 5. Isotherms from a 3D Rosenthals model with q = 1.2 kW, applied on a sample of AISI304 steel with large dimensions.

4.2. Hypotheses of innite plate and two-dimensional temperature eld

The hypotheses of an innite plate, used in both physical models, and the two-dimensional heat transfer (model 2) must be assured. In this sense, the experiment must be designed in such way that the test-plate geometry has the minimum size to attend these conditions. For verifying the innite assumption, an analysis of the 3D Rosenthal model is performed. Fig. 5 shows the results. It can be observed that a sample width of 0.05 m is wide enough for obtaining values of temperature close to the environment temperature 300 K (innite condition). Another physical effect that can occur is the inuence of the molten pool on the temperature along the thickness. It means, since the two-dimensional numerical model is used, no change in temperature in the thickness direction is expected. This difculty can be overcome by placing the thermocouples away from the weld pool thermal resistance inuence. These locations, in this work, are chosen to be 0.008 m away from the test-plate symmetry line, in which the gun is applied, as shown in Fig. 4. This above-mentioned inuence is estimated through a solution of a 3D numerical heat transfer diffusion with a heat moving. The heat-moving source is used here to simulate the phase change presence in the symmetry line. This source must supply heat enough to allow the sample reaching a temperature high

5. Results and discussion 5.1. Input data Fig. 7 shows the measured temperatures on the bottom surface of the test-plate for the four thermocouples shown in Fig. 4. From these temperatures and the choice of the model, either the heat ux, q, or the weld pool radius, r, can be obtained. It can be noted that the temperature from thermocouple number four presents unexpected behavior in the cooling period. This fact can be assigned to intrinsic experimental difculties like, for example, interference due to electrical noises. Despite of this, these signal data are used in the estimation procedure. 5.2. Temperature eld and heat ux estimation, q (model 1) Table 1 presents the results of the q estimation using model 1. Since the heat ux to the workpiece is estimated, the temperature

Fig. 6. Behavior of the temperature at the opposite face: (a) positioned away 0.008 from the symmetry line and (b) in the symmetry line.

C.V. Gon alves et al. / Journal of Materials Processing Technology 172 (2006) 4251 c Table 1 Heat ux and estimation results using SA and model 1, P = 1178 W Thermocouple 1 2 3 4 Qu (W/m2 ) 656 816 901 798 (%) = Qu /P 55.6 69.2 76.5 67.7

47

distribution can then be obtained solving the direct problem given by Eq. (4). Fig. 8 presents examples of temperature evolution estimated at four different elapsed times t = 2, 10, 20 and 25 s. One must observe that the Rosenthals physical model gives rise to good results just for temperature away from the moving heat source. This is due to the fact that the model does not take into account the phase change presence. Interface position of phase change is partially identied by using a low pass lter (threshold), where any temperature higher than the fusion value is assumed equal to T = Tm . The main disadvantage of this approach is that the position of the molten pool is dependent on the choice of the type and place of the lter. 5.3. Temperature eld and radius of weld pool estimation, r (model 2) As mentioned before, on the contrary of the previous q estimation, the unknown variable in model 2 is the fusion radius, r, and the location of the interface of phase change. In this case, an advantage of this procedure is the possibility of validation of the radius estimation, since the experimental width, Lc , of weld pool would correspond to the diameter of the pool. It means, if the weld pool is assumed circular, the estimated value of the radius, r, can be compared to the half of the value of Lc (r = Lc /2) (Fig. 9).

Fig. 9. A weld bead from a typical experimental condition, obtained by using images from an optical microscopic, used for the width measurements. Table 2 Radius estimation (Lcinv ) using golden section and model 2, P = 1178 W, compared with the calculate radius (Lc ) from the actual radius (bead width, r) Thermocouple 1 2 3 4 r (103 m) 1.96 2.13 2.03 1.95 Lcinv (103 m) 3.92 4.26 4.06 3.90 Lc (103 m) 4.20 4.18 3.94 3.75 Relative error (%) 6.8 2.0 3.2 4.0

The width of weld bead for the experimental conditions was obtained using images from an optical microscope as illustrated in Fig. 9. The resulting weld bead width compared with r estimation is shown in Table 2. The search interval used was

Fig. 8. Resulting temperature eld at the welding face at different elapsed welding times (2, 10, 20 and 25 s). Estimation using model 1 and thermocouple 1.

48

C.V. Gon alves et al. / Journal of Materials Processing Technology 172 (2006) 4251 c

5.4. Comparison of results using models 1 and 2 Fig. 11 presents a comparison between the estimated and experimental temperature using the two models proposed. It can be observed that model 2 presents results closer to the experimental evolution. This behavior can be attributed to effects such as phase change and the values of thermal properties varying with temperature, which are considered in this model. The effect of radiation and convection losses could be more evident at the cooling period. In order to obtain better comparison among the results, a quadratic error evolution for each thermocouple are presented in Fig. 12. Again, it can be observed that the model 2 presents better results, fact that can be veried by the minor values of ST . It can also be observed that in model 1 the maximum discrepancy is in the cooling period (after 24 s), fact that can be justied by the absence of the heat convection and radiation effect in the Rosenthals model. 5.5. Thermal efciency estimation
Fig. 10. Estimated temperature evolution at three different elapsed times t = 2, 15 and 25 s. Estimation using model 2 and thermocouple 1.

0.04 103 m and the discrepancy between the values was less than 6.8%. Fig. 10 shows the temperature eld at the welding face at different times for thermocouple 1.

As mentioned before, the heat ux that reaches the base metal is extremely important for welding analyses. The ratio dened by the effective heat delivered to the workpiece and the total electrical power, that is, the thermal efciency, is a very good parameter to indicate the process performance. The heat power expressed by voltage versus current is not totally used to develop the welding process. Relevant heat transfer losses, as heat loss

Fig. 11. Comparison between the estimated and experimental temperature using the two models proposed. Thermocouples: (a) 1, (b) 2, (c) 3 and (d) 4.

C.V. Gon alves et al. / Journal of Materials Processing Technology 172 (2006) 4251 c

49

Fig. 12. Objective function, ST and Sq , using the two models proposed. Results obtained by using the thermocouples: (a) 1, (b) 2, (c) 3 and (d) 4.

due to the phase changes and heat losses due to the convection and radiation from the plate and arc column to the environment are present in the process. It should be observed that in model 2 the heat delivered to the workpiece, named here Qu , just can be obtained through the energy balance around the plate. In this case, the Qu can be obtained by adding the effect of heat quantities due to the fusion, Qfus , to the rate of energy storage in the plate, Qint , and to the heat losses due to the convection/radiation from the plate to the environment, Qcv/rad . Thus, the value of Qu can be evaluated as Qu = Qfus + Qint + Qcv/rad where Qfus can be calculated as (10)

ny n

Qfus (t) =
j=1 i=1

(i) y(j)( H(i, j) t

H0 (i, j)) (11)

where ny and n are, respectively, the total node number in the mash in x- and y direction, and x and y are the height and width of the control volume, respectively. Qcv/rad can be calculated through the balance at the upper and the lower plate surface as
ny nx

Qcv/rad =
j=1 i=1

h(i, j)(T (i, j) T ) x(i) y(j)

(12)

Fig. 13. Heat loss by convection/radiation, Q(t)cv/rad , in: (a) absolute values, [W] using the whole range of time and (b) relative values, Qcv/rad /P, referring only to the heating time.

50

C.V. Gon alves et al. / Journal of Materials Processing Technology 172 (2006) 4251 c

and the rate of energy storage in the plate can be evaluated as

ny nx

Qint =
j=1 i=1

cp

T (i, j) Ti x(i) y(j) t

(13)

With these considerations a transient thermal efciency, , can dened as Qu = (14) P where P, as mentioned before, is the total electric power. It can be observed that this procedure avoid the necessity of identication of the heat losses due to the convection/radiation from the arc column to the environment, which involves a solution of very complicate physical phenomena as uid ow and heat transfer in plasma. Figs. 1316 present the heat quantities and efciency involved in the process under consideration here. Fig. 13 presents the heat loss, Qcv/rad in absolute and relative values. It can be observed that the value of Q(t)cv+rad rises up to reaching the maximum value, near the duration of the welding, 24.1 s. This behavior can be justied due to the higher temperature gradients between the plate and the environment. After this time, just workpiece cooling is present and the heat loss decay with the plate temperature. The ratio between the heat used to melt the plate, Qfus , and the total electric power, P, is known by thermal efciency of fusion, fus . Fig. 15 presents the behavior of fus for the weld conditions presented here with a welding speed of 8.3 mm/s. It can be observed that the values are situated in a range of 12.520%. Giedt et al. [26] have found values in a range of 1545% for weld speeds of 20 mm/s. Fig. 14 presents the resulting global thermal efciency, . The values are situated in the range of 6080%, with a larger variation at the beginning of the weld. The decaying of the ther-

Fig. 15. Global thermal efciency as a function of the time.

Fig. 16. Global thermal efciency, thermal efciency of fusion and convection and radiation efciencies.

mal efciency can be attributed to the presence of phase change in early times. The global thermal efciency, thermal efciency of fusion and convection and radiation heat losses are shown together in Fig. 16. They were obtained using experimental temperature from thermocouple 1. It must be pointed out that, during the literature survey, values of transient thermal efciency were not found. 6. Conclusion The physical model that considers the phase change, thermal properties varying with temperature and heat loss to the environment for convection/radiation demonstrated to be able of identifying the global thermal and fusion efciency at each instant of the process. Moreover, the transient model (model 2) presented more accurate results when compared to the simplied one (model 1). However, if just average values are required, the

Fig. 14. Thermal efciency of fusion as a function of the time.

C.V. Gon alves et al. / Journal of Materials Processing Technology 172 (2006) 4251 c

51

use of Rosenthals model represents a very good option, since the simplied model is easier for implementation and it is very cheep in the view of computational time consuming. It also important to mention that the circular shape assumed to the weld pool shown to be a very good approximation condition for GTAW with heat source speed at u = 0.008333 m/s and welding current at order of 78 A. A low difference (less than 7%) between measured and estimated values conrms this conclusion. Acknowledgements The authors would like to thank the Brazilian Agencies for Research Development, CAPES, CNPq and FAPEMIG for the nancial support. Thanks are also due to Dr. Sandro M.M. Lima e Silva and to MSc Solid nio Rodrigues Carvalho, for the helpful o discussions. References
[1] D. Rosenthal, Mathematical theory of heat distribution during welding and cuting, Weld. J. 20 (1941) 220234. [2] C.L. Tsai, C.A. Hou, Theoretical analysis of weld pool behavior in the pulsed current GTAW process, J. Heat Transfer 110 (1988) 160165. [3] K.S. Boo, H.S. Cho, Transient temperature distribution in arc welding of nite thickness plates, Proc. Inst. Mech. Eng. 204 (1990) 175183. [4] S.K. Jeong, H.S. Cho, An analytical solution to predict the transient temperature distribution in lled arc welds, Weld. J. 76 (1997) 223232. [5] T. Nguyen, A. Ohta, K. Matsuoka, N. Suzuki, Y. Maeda, Analytical solutions for transiente temperature of semi-innite body subjected to 3D moving heat sources, Weld. J. (1999) 265274. [6] M.D. Shian, P.S. Wei, Three-dimensional analytical temperature eld around the welding cavity produced by a moving distributed highintensity beam, J. Heat Transfer 115 (1993) 848855. [7] J. Goldak, M. Bibby, J. Moore, B. Patel, Computer modeling of heat ow in welds, Metall. Trans. B 17B (1986) 587600. [8] G.H. Kraus, Thermal nite element formulation and solution versus experimental results for thin-plate GTA welding, J. Heat Transfer 108 (1986) 591596.

[9] T. Zacharia, A.H. Eraslan, D.K. Aidun, Modeling of non-autogenous welding, Weld. Res. Suppl. (1998) 18s27s. [10] E.A. Bonifaz, Finite element analysis of heat ow in single-pass arc welds, Weld. Res. Suppl. (2000) 121125. [11] J. Crank, Free and Moving Boundary Problems, Claredon Press/Oxford University Press, New York, USA, 1984. [12] V.R. Voller, C. Prakash, A xed grid numerical modeling methodology for convection/diffusion mushy region phase change problems, Int. J. Heat Mass Transfer 30 (1987) 17091719. [13] Y. Cao, A. Faghri, W. Chang, A numerical analysis of stefan problems for generalized multi-dimensional phase-change structures using the enthalpy transforming model, Int. J. Heat Mass Transfer 32 (1989) 12891298. [14] C.R. Swaminathan, V.R. Voller, On the enthalpy-method, Int. J. Numer. Meth. Heat Fluid Flow 3 (1993) 233244. [15] N. Al-Khalidy, Enthalpy technique for solution of Stefan problems: application of the keyhole plasma arc welding process involving moving heat source, Int. Commun. Heat Mass Transfer 22 (1995) 779790. [16] N. Al-Khalidy, Application of optimization methods for solving inverse phase-change problems, Num Heat Transfer, Part B (1997) 477497. [17] B. Rubinsky, A. Shitzer, Analytic solutions to the heat equation involving a moving boundary with applications to the change of phase problem (the inverse Stefan problem), J. Heat Transfer 105 (1978) 550554. [18] Y.F. Hsu, B. Rubinsky, K. Mahin, An inverse nite element method for the analysis of stationary arc welding processes, J. Heat Transfer 108 (1986) 734741. [19] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere Publishing Corporation, USA, 1980. [20] S.F.P. Saramago, E.G. Assis, V. Steffen, Proceedings of the 20th Iberian LatinAmerican Congress on Comparative Methods in Engineering, S o a Paulo, Brazil, 1999. [21] G.N. Vanderplaats, Numerical Optimization Techniques for Engineering Design, McGraw-Hill, USA, 1999. [22] O.M. Alifanov, Solution of an inverse problem of heat conduction by iterations methods, J. Eng. Phys. 26 (1974) 4. [23] J.V. Beck, B. Blackwell, Handbook of Numerical Heat Transfer, John Wiley & Sons Inc., New York, 1988. [24] J.V. Beck, B. Blackwell, C.R. St. Clair, Inverse Heat Conduction, IIIposed Problems, WileyInterscience Publication, New York, 1985. [25] L.O. Vilarinho, Development of experimental and numerical techniques for TIG arc characterization, Doctorate Thesis, Federal University of Uberidndia, Brazil, 2003. [26] W.H. Giedt, L.N. Tallerico, P.W. Fuerschbach, Weld. J. (1989) 2832.