Influence of welding parameters on grain size, HAZ and degree of dilution of 6063-T5 alloy: optimization through the Taguchi method of the GMAW process

The aim of this work is to design experiments using Taguchi’s method that determine the influence of welding parameters on the grain size, heat-affected zone and degree of dilution of the 6063-T5 alloy. The welding process used is GMAW, and the welding parameters are the power, welding speed and bevel spacing. The study of the influence of these welding parameters on the measurements made during welding (which are the size of the heat-affected zone, the degree of dilution and the grain size) allows one to determine the quality of the joint. In addition, the welding parameter that most influences the minimisation of these three measurements will be determined.


Introduction
Aluminium alloys are characterised by harder or softer mechanical properties depending on the quality and quantity of the alloying elements that are present in the aluminium matrix [1]. These alloying elements can be lost due to the thermal input received during welding [2,3]. On the other hand, the size of the grain in the alloy also determines the mechanical properties of the aluminium alloy, which is more resistant as the size of the grain decreases. Finally, the size of the HAZ is also relevant: the smaller its size, the better the properties of the weld joint, without taking into account the effect of residual stresses. The grain size [4] is determined by microscopic analysis using comparative charts according to the ASTM E112 [5] and NF ISO 643 [6] standards. There are computer programs that, using the microstructure of a weld, can estimate the size of the grains from polarisation micrographs and image orientation [7]. There are other programs that can determine the grain size from statistical analysis, specifically by estimating the grain size from the parameters of the grain [8]. These programs, which are used in specialised metallurgy laboratories and detect the microstructure and grain size of welds, consist of powerful intelligent algorithms that determine the structure to be analysed through values that they compare with a set of standard samples. The main advantage of these programs is that a large number of metallographic replicas can be rapidly analysed. The downside is that the image quality, sample preparation, polish quality, etc., influence the results. This technique can be used when many metallographic replicas are made, and the preparation procedure of the metallographic sample to be analysed is controlled.
The dilution between the filler material and the base material is defined as the ratio of molten base metal to the total amount of molten metal [12]. The dilution is expressed by the following expression: (1) Dilution (%)= Molten base metal Total molten metal The dilution is proportional to the thermal input of the heat source [13,14], that is, the greater the thermal input, the greater the dilution [15]. In the present work, both the relationship between the thermal input and the dilution calculated for the weld bead and the relationship between the molten base material and the total amount of molten material are analysed. The lower the dilution, the lower the presence of base material in the weld bead and the better the quality of the weld bead.
Classical experimental design aims to identify the factors that affect the mean response and their influence on a target value. The advantage of using Taguchi experimental design instead is that it reduces the variability in the deviation from the mean target value [16][17][18][19][20][21].
The robust parameter design of Taguchi methods (orthogonal matrices) allows the influence of the factors to be analysed using just a few experiments. Taguchi's experiment designs are balanced; that is, no factor is analysed independently of the others. Accordingly, it is sufficient to perform nine experiments to evaluate the size of the weld grain, the size of the HAZ and the dilution of the filler material.
There are various studies that have performed the optimisation of welding processes using Taguchi methods. Several studies optimised the weld bead based on the electrical parameters of the source [22][23][24]. There is also an optimisation study that evaluated the welding thermal cycle [25]. All these studies were carried out using the GMAW process [26][27][28][29].
The present work aims to study the effect of welding parameters, that is, the power (P), welding speed (Vs) and edge separation (Sp), on the grain size, HAZ size and dilution of the weld bead.

Experimental procedure
In a previous work [22], the weld that is the focus of the current paper was evaluated by determining the influence that the aforementioned parameters have on the geometry of the weld bead. In [22], which was written by the authors of this paper, the following is indicated.
The experimental procedure is carried out with a robot welding arm (brand -ABB, model -IR 1400), as shown in Fig. 1a. The robotic arm has six degrees of freedom of movement, and a system controls the welding parameters. The welding robot has a high precision, which guarantees the correct finish of the joint. The elements to be joined are nine tubes that each has an outside diameter of 50 mm, a thickness of 2 mm and a length of 150 mm; they have been cut longitudinally and subsequently joined by a welding bead. To be able to longitudinally connect the parts of the tube, a carbon steel tool was designed to support the two parts of the aluminium tube before they are welded. This tool is shown in Fig. 1b.
Aluminium 6063-T5 is usually supplied in the form of a tube and not in the form of sheets [20]. The filler metal used is ER-5356, which is very suitable for welding 6063-T5 [30]. The diameter of the wire used is 1 mm, and the protective gas is 99% argon. The type of transfer used by the welding process is short arc transfer [4,29,31]. The chemical compositions of the base metal and the filler metal are given in Tables 1 and 2, respectively.

Design of experiments
Three variables with three levels each were used to develop nine welding experiments according to the Taguchi method with an L9 orthogonal array. Table 3 shows the three variables used: the power (P), welding speed (Vs) and edge separation (Sp).
Three factorial levels arranged in ascending order were selected. For the power, 1557 W represents the minimum value of the power used in the experiment, 1673 W   [22].
To determine the matrix of nine experiments using Taguchi's method, the statistical software Minitab [32] was used. Table 4 shows the nine experiments determined using the Taguchi methodology. The power is obtained by multiplying the intensity by the voltage. Table 5 shows the intensity and voltage values used to obtain the three power values used in the experimental design. Figure 2 shows the nine welds corresponding to each of the experiments carried out according to Table 4. The welds have been polished with different grain sizes which makes it sound like 'grain sizes' were used to polish the welds, as indicated in [22].

Discussion of results
In reference [33], the effect of kind of gas on the appearance of defects that are showed in these welds is discussed. The main cause for which defects appear is because argon gas is used and not helium gas as is indicated in reference [33]. When argon is used as protection gas, the defects shown in Fig. 2 are more likely to appear. If helium gas is used, the probability of these defects appearing is less.

Macrographs of the nine experiments performed
To measure the grain size, Grani software is used. It allows the 'G' index to be calculated with precision according to the standards ASTM E112 [5] and ISO 643 [6]. A complete   set of measurements is performed for each of the nine experiments using the Grani software. On the macrograph, the grain size is measured in the thermally affected area, that is, in the HAZ. For each experiment, the G index of the grain and the number of grains are measured ( Table 6). The Grani program uses the Hilliard method [23,24] to determine these values. Figure 3 shows the measurements made in each experiment. A red circle marks the area where the grain's G index and the number of grains were measured. Figure 4 represents the relationship between the G index of the ASTM standard and the number of grains identified by the Grani program. Therefore, the grain size is evaluated and not the letter number since the relationship is the same.
To evaluate the effect of each of the variables on the number of grains, the following multivariate linear regression equation is obtained: (2) In the analysis of variance in Table 7, the influence of each of the variables on the grain size is determined. The three P-values shown are large, that is, they indicate that these variables do not have a significant influence on the number of grains in the aluminium weld. Therefore, these three variables do not affect the number of grains in the weld.
In accordance with the above, the effect of the three main variables on the number of grains is not further analysed because they have not influenced the number of grains.

Size of the HAZ in each experiment performed
The size of the HAZ of each weld is measured using a profile meter that shows the test tubes used in each experiment on which the size of the HAZ is measured using the profile meter that is showed in reference [22]. Figure 5 shows the micrographs of the HAZ made for each of the joints.
In each of the experiments, the size of the HAZ is measured; it is shown in red on each of the micrographs. The following HAZ values are obtained.
To evaluate the effect of each of the variables on the size of the HAZ, the following multivariate linear regression equation is obtained: (3) Size of HAZ = −1.30 + 0.001831 ⋅ P, W − 0.0778 ⋅ Vs, mm∕s + 0.1495 ⋅ Sp, mm Equation 3 has an adjusted R2 value of 51.87%, which is not close to 95%. Figure 6 shows the trend of each of the variables versus the size of the HAZ. Table 9 shows the influence of each of the variables on the size of the HAZ as determined by the analysis of variance. The three P-values shown are small, that is, they indicate that these variables have a significant influence on the size of the HAZ of the aluminium weld. Therefore, these three variables significantly affect the size of the HAZ of the weld.
The variable that has the most influence is power (P); it is followed by the edge separation (Sp), which in turn

Number of grains
Index G of grain is followed by the welding speed (Vs). Figure 7 shows the effect of each of the variables on the size of the HAZ. From Fig. 7, the specific contributions of each of the variables to the size of the HAZ are obtained: P = 53%, Sp = 29% and Vs = 18%. Figure 8 shows the area of dilution in the weld bead (area enclosed by a red outline). As shown in Eq. (1), the dilution ratio is determined by dividing the amount of molten base metal by the total amount of molten metal. Figure 8 shows the dilution values (%), weld bead area and molten base metal area. The area of the weld bead was measured in [22] using a profile meter.

Determination of the degree of dilution
The area of the molten base metal is determined from the CAD representation of the weld joint using the profile gauge and the initial geometry of the tubular profile before it is welded. The bead is superimposed on the tube geometry, and the amount of molten base metal found in the weld metal is obtained. Figure 9 represents the weld bead and base metal for each of the experiments. The edge separation, Sp, is considered in determining the degree of dilution.  The fit of Eq. 4 has an adjusted R2 value of 70.09%, which is acceptable because it is close to 100% and because the P-values of Table 11 are low. This is explained in the Minitab guide [32]. Figure 10 represents the trend of each of the welding variables versus the dilution in the weld bead. As the power increases, the dilution in the weld bead increases, and as both Vs and Sp decrease, the dilution in the weld bead decreases.
In the analysis of variance in Table 11, the influence of each of the variables on the degree of dilution is determined. The three P-values shown are small, that is, they indicate that these variables have a significant influence on the degree of dilution in the weld bead. Therefore, these three variables significantly affect the degree of dilution in the weld bead.
In Fig. 11, the influence of each of the variables on the degree of dilution is presented. The variable that has the (4) most influence is the edge separation, Sp, followed by the power, P, which in turn is followed by the welding speed, Vs. From Fig. 11, the specific contribution of each of the variables (P 32%; Sp 42%; Vs 27%) on the dilution of the base metal in the weld bead is obtained.

Optimisation of results
After the experiments are carried out and the grain size, the size of the HAZ and the dilution in the weld are measured, the analysis of variance is used to determine the influence of the welding parameters, P, Vs and Sp, which are then optimised using the Taguchi method and the statistical program Minitab [27]. The criterion to be used in the Taguchi optimisation is 'the smaller the better' for both the degree of dilution and the size of the HAZ.
The grain size will not be analysed because, as indicated above, there is no relationship between the welding variables and the grain size; therefore, an evaluation of the grain size is not performed.

Taguchi optimisation
Methodology of Taguchi is a suitable one in the optimization of welding processes as indicated in reference [34]. According to this, it is considered that the methodology of Taguchi is suitable for the optimization of welding processes.
The noise signal ratio (S/N) for the case 'the smaller the better' is determined as follows.
The S/N ratio is defined as  . 6 Representation of the effect of each of the variables on the size of the HAZ Table 12 shows the S/N responses of the welding parameters for the size of the HAZ and for the percentage of dilution. Table 13 shows a comparison between the influence of the welding variables obtained from the analysis of variance and the influence of these variables obtained from the Taguchi optimisation with the criterion 'the smaller the better'. The table above verifies that the power has the most significant influence; it has similar values for the ANOVA analysis and the Taguchi optimisation, both for the size of the HAZ and the dilution. However, this is not the case for Vs and Sp: the influences of these variables for both the dilution and HAZ size are inversely proportional. From this, it follows that the study of variance is not a valid method to use for the desired optimisation; it only indicates the mean influence of each of the variables on each of the measurements made (Fig. 12).
The figure above shows the optimal parameter values that can be used to obtain the minimum HAZ size. The parameter values P = 1557 W, Vs = 13 mm/s and Sp = 0 mm result in a minimum HAZ size of 0.53 mm (Fig. 13).
The figure above shows the optimal parameter values that can be used to obtain the minimum dilution. The parameter values P = 1557 W, Vs = 13 mm/s and Sp = 1 mm result in a minimum dilution value of 18%.
Methodology of Taguchi is sufficiently robust in welding processes [34]; it is not considered necessary to carry out experiments to obtain the smallest HAZ [P = 1557 W;

Optimised combination of process parameters
The influences of each of the variables on the HAZ size and dilution have been analysed separately. Next, the optimal values of the three welding variables, P, Vs and Sp, needed to obtain minimal dilution and a minimal HAZ size are determined (Fig. 14).
The optimisation graph shows the effect of each input variable (columns) on the weld measurements carried out; the responses are presented in rows. The vertical red lines in the graph represent the configuration of the current value. The red numbers at the tops of the columns show the configuration of the current input variables. The horizontal blue lines and numbers represent the responses for the current factor level.
Desirability in general terms evaluates the effectiveness with which a combination of input variables satisfies the goals that have been defined for the responses. The   Table 10). Similarly, for the HAZ, the minimum value obtained in the figure above is y = 0.71 mm, and the minimum value obtained in the experiments is 0.64 mm (see Table 8); the welding parameters used to obtain the minimum values in both cases are P = 1557 W, Vs = 13 mm/s and Sp = 1.2 mm. The combined optimization of the process parameters corresponds to experiment number 3 of Table 4.

Conclusions
(a) According to the analysis of variance carried out in Table 7, there is no relationship between the welding parameters and the grain size obtained in the experiments. (b) In accordance with the previous conclusion, the variables were not optimised to obtain the smallest possible grain size since there is no relationship between these variables and the grain size. (c) Regarding the size of the HAZ, the variables make the following contributions to the size of the HAZ: P 53%, Vs 18% and Sp 29%. (d) Regarding the degree of dilution, the variables make the following contributions to the degree of dilution: P 32%, Vs 27% and Sp 42%. (e) A Taguchi optimisation is performed to obtain the lowest possible values of both the size of the HAZ and the degree of dilution.
• To obtain the minimum size of the HAZ, the optimal welding parameters are P = 1557 W, Vs = 13 mm/s and Sp = 0 mm. From these values, a HAZ size of 0.53 mm is obtained. • To obtain the minimum dilution value, the optimal welding parameters are P = 1557 W, Vs = 13 mm/s and Sp = 1 mm. From these values, a minimum dilution value of 18% is obtained.
(f) An optimisation of the welding parameters is carried out to obtain minimal values of both the HAZ and the degree of dilution; it produced the following values: P = 1557 W, Vs = 13 mm/s and Sp = 1.2 mm. (g) The adjustment of the optimised variables is acceptable because there are high R 2 values (close to 100) and low P-values, which indicate that the predictive model is good.