1 Introduction

The demand for copper shows a steady increase over the last few years. It is predicted to rise further [1, 2]. For component applications, it is essential to have a joining technology which preserves the desired material properties. This task can be achieved by welding. However, in the case of pure copper, welding is a challenging task due to the high thermal conductivity and expansion as well as high shrinkage [3]. This behavior results in high residual stresses and deformations in conventional welding processes [4,5,6]. The related problems can be avoided by the use of a high-energy source, e.g., the electron beam (EB). Its high power density [7] enables deep penetration welding, also called keyhole welding. Due to the high power density of the incident beam the material of a welded plate can be vaporized locally. This vaporization creates a vapor capillary, also known as keyhole, which penetrates the plate. The keyhole is surrounded by molten metal which is held back by the vapor pressure, also called recoil pressure, of the vaporized material. Due to the relative motion between the plate and the beam new material is vaporized at the front of the keyhole. At the rear end of the keyhole the surrounding melt flows in and solidifies, creating a narrow weld. This process allows welding of thick plates within a single pass without a filler metal. Hence, the productivity in electron beam welding (EBW) is high. Deformations and residual stresses are small as the melt pool is narrow. Due to the lack of a filler metal no impurities are added into the weld [8,9,10]. Additionally, the inert vacuum environment prevents the dissolution of gases in the melt pool and reduces the formation of plasma in the beam path. Also, the absorbed energy of the EB is nearly independent of the surface characteristics for a given acceleration voltage and material and is usually higher than 75%. Therefore, the electron beam is an appropriate tool to weld pure copper [11, 12].

However, keyhole welding produces a wide range of defects that reduce the mechanical properties of the joint considerably. Typical defects are pores, melt ejections, cold-shuts and spiking [13,14,15,16,17,18]. The latter describes periodic deviations of the penetration depth which usually result in the other aforementioned defects [19, 20]. During full and partial penetration EBW of copper spiking was frequently observed. Therefore, it is considered to be critical for the quality of copper welds [21,22,23,24].

Over the last few decades spiking has been investigated intensively. The majority of studies suggested that it is caused by instabilities of the keyhole. Some authors concluded that spiking is inherent to welding with a keyhole due to the increase of closing forces with increasing penetration depth. Hence, the proposed method in order to reduce spiking was to minimize penetration depth [25,26,27,28]. Other studies suggested that spiking is caused by a periodic melt flow in the keyhole. This could be avoided by the addition of volatile alloying elements in the base metal, which increased the recoil pressure [13, 29]. However, there seems to be no generally positive effect of an addition of a volatile alloying element as the effect diminished in dependency of welded material and insertion position [30]. Also, Wei et al. [31] reported a greater spiking amplitude in welds of Al 5083 compared to welds of Al 6061 for different welding speeds and focal positions. In other models, the deviation of the penetration depth was due to the collapse of the vapor capillary. This collapse is brought about by knobs and cavities in the surrounding melt pool, which are caused by the dynamic melt pool flow. The protrusions interact with the beam and therefore reduce the recoil pressure in parts of the keyhole. This, in turn, disturbs the dynamic equilibrium at the gas/liquid-phase boundary and results in the aforementioned collapse. Afterwards, a new keyhole is formed and the process repeats, resulting in an oscillating penetration depth [30, 32,33,34,35,36]. In numerical investigations, these oscillations were simulated [37, 38]. It was shown that the collapsing part of the vapor capillary can remain in the weld seam as gas cavity. This effect regularly occurred, when the thermal conductivity of the base metal was high [20, 39,40,41,42].

Fetzer et al. [43] showed that spiking also occurred in the case of a stable keyhole. Only the occurrence of porosity was a result of the keyhole instability. In their model, spiking was caused by the alternation of the irradiation on the front of the keyhole. It was shown that the frequency of spiking linearly scales with the welding speed. Other studies showed a similar tendency [31, 44].

Linear and circular beam oscillations showed some success in reducing spiking for various metals [21, 45], which was mainly attributed to a more uniform melt flow [46]. Heider et al. [14, 47, 48] and Cho et al. [49] successfully used a sinusoidal power modulation to stabilize the vapor capillary. Moreover, spiking was successfully suppressed by increasing the welding speed [10, 50].

The current work was focused on the suppression of defects due to spiking in full and partial penetration EBW of ETP copper (electrolytic tough pitch copper, UNS C11000). The welds were characterized by macroscopic investigations, microstructure analysis using light optical and scanning electron microscopy (SEM) as well as measurement of the electrical resistance and mechanical properties.

2 Experimental details

2.1 Base metal

For all experiments, Cu-ETP (Hans-Erich Gemmel & Co. GmbH, Germany) was used as base metal (BM). The thickness d of the sheets was 4 mm and 10 mm. The chemical compositions (see Table 1) were determined using gas fusion analysis (Bruker G8 Galileo, O) and optical emission spectroscopy (ARL4460 Metal Analyzer, other elements).

Table 1 Chemical composition of the base metal with a thickness of 4 and 10 mm, mass fraction in ppm
Full size table

Samples were cut in a uniform geometry of \(80\times 50\,\text {mm}^{2}\). The plates for tensile specimens had dimensions of \(125\times 50\,\text {mm}^{2}\). The specimens were brushed with an abrasive fleece and cleaned with acetone. Specimens for experiments with preheating were blasted with alumina powder with 120–150\(\upmu \)m particle size. This process ensured a homogeneously rough surface for pyrometric temperature measurements.

2.2 Welding parameters and setup

The welding experiments were carried out with two electron beam facilities. Specimens with 10 mm thickness were welded in a pro-beam K26-80/15 at a constant acceleration voltage (\(U_{acc}\)) of 80 kV. Specimens with 4 mm thickness were welded in a pro-beam K40-150/30 at a constant acceleration voltage of 150 kV. In both vacuum chambers the pressure was \(2\times 10^{-3}\)mbar. The weld length was 50 mm. Results for 10 mm thickness are marked separately. For welding speeds (v) > 80 mm  \(\text {s}^{-1}\), the welds were made using beam movement. Otherwise, the CNC-controlled axes inside the chambers were used to move the specimens. The welds were made at sharp focus. In order to improve the formation of the upper bead, selected specimens were subjected to a post welding smoothing process (\(U_{acc} = 150\,\text {kV}\), a beam current (\(I_b\)) of \(10\,\text {mA}, v = 10\,\text {mm}\,\text {s}^{-1}\), circular beam scanning with a diameter of 3 mm and a frequency of 3000 Hz).

For preheating, a heating field of \(20\times 50\,\text {mm}^{2}\) with \(100\times 250\) equidistant interaction points was placed symmetrically around the zone that was welded afterwards. Each point had an interaction time of 110 ns. The temperature at the surface of the plate was measured with a two-color pyrometer, which was calibrated using a thermocouple. The pyrometer had a built-in PID-controller to control the beam current based on temperature using the control software SensorControl. The maximum beam current for preheating was set to 55 mA. A thermographic camera was set up above the specimens to measure the surface temperature distributions on the plates. The surface temperature within the pyrometer-spot was set to 900 C and was held for 20 s. Welding was done immediately afterwards. For details on the configuration of the pyrometer and thermographic camera see [51].

The beam current to fully penetrate the plates was determined for different welding speeds for specimens at room temperature (without preheating) and with preheating. As the spiking phenomenon caused individual craters to form, intersected by unaffected surfaces at the bottom of the plate, a weld was considered fully penetrated when the craters resulting from the spiking phenomenon were no longer intersected by unaffected surfaces, i. e. when the craters touched or were connected by solidified melt. A weld was also counted as fully penetrated, when a continuous root was formed. Also, a full factorial design for \(v= 10,~20,~30\) and \(40\,\text {mm}\,\text {s}^{-1}\) and the line energies (calculated according to Eq. 1) \(E = 120, 160, 200\) and \(240\,\text {J}\,\text {mm}^{-1}\) was carried out for \(d= 10\,\text {mm}\). The line energy refers to the energy input per weld length and is commonly used for EBW processes [10, 52,53,54,55].

$$\begin{aligned} E = \frac{I_b U_{acc}}{v} \end{aligned}$$
(1)

2.3 Characterization methods

2.3.1 Macroscopic and microscopic analysis

The upper bead and the root were investigated using a stereo microscope. After welding, cross sections of the welds were prepared. For the determination of the weld geometries, three cross sections were analyzed. Longitudinal sections were cut in the center of the weld, perpendicular to the specimens’ surface. The specimens were ground, polished and etched with a solution of 65 ml ethanol, 25 ml HCl (37%) and 4 ml H\(_2\)O\(_2\). The specimens were etched for up to 15 s and cleaned with ethanol afterwards. For SEM investigations, the specimens were vibration polished (SiO-suspension with a particle size of 0.02\(\upmu \)m) for 24 h.

The etched specimens were examined by light optical microscopy (Olympus GX51). The SEM investigations were carried out with a MIRA 3 XMU TESCAN with an acceleration voltage of 20 kV, a working distance of 20 mm and a tilting of 70. The electron backscatter diffraction (EBSD) analysis was performed with the software OIM ANALYSIS 6.0. The white marked areas in Fig. 1 were scanned with a point distance of 5\(\upmu \)m. The minimal misorientation for the distinction between grains was set to 12.

Fig. 1
figure 1

Positions of the EBSD-scans (white marked sections with numbers I-IV) for a cross sections and b longitudinal sections, displayed values in millimeters

Full size image

The results for all depths (Fig. 1a) were combined into one grain area distribution for different welding speeds. The resulting distributions were compared using the Dunn test with Hochberg correction.

Fig. 2
figure 2

Geometry of the tensile specimens, rectangular cross section (\(10\times 4\,\text {mm}^{2}\)), displayed values in millimeters

Full size image

2.3.2 Ultrasonic testing

For ultrasonic testing, longitudinal sections with a length \(> 10\,\text {mm}\) were cut perpendicular to the plates’ surface and parallel to the welding direction, and ground. The width of the sections was 8 mm, with the weld in the center. For further details regarding ultrasonic testing configuration, please refer to the experimental section of [56]. In the full factorial design concerning variables v and E in Sect. 2.2, the number of spikes was counted and normalized to the scanned weld length.

2.3.3 Tensile tests

Tensile specimens with a rectangular cross section (\(10\times 4\,\text {mm}^{2}\)) were machined according to Fig. 2. The gauge length was 45 mm. The elongation was measured with an extensometer. Tests were done with an electromechanical testing device (inspekt desk 50, Hegewald und Peschke, Germany). The strain rates were chosen to be \(7\times 10^{-5}\,\text {s}^{-1}\) up to a plastic strain of 0.2% and \(2\times 10^{-3}\,\text {s}^{-1}\) afterwards, according to DIN EN  ISO 6892-1.

The fracture strain was calculated based on the gauge length after the test.

2.3.4 Resistance measurements

The electrical resistance was measured using the four-point potentiostatic method (PROMET R300, Kocos) at a current of 100 A on specimens with the dimensions \(80\times 30\times 4\,\text {mm}^{3}\). The measurement length was 8.8 mm. As the specific resistance of the weld was of interest, a connection in series of the weld and the base metal was assumed. Using Ohm’s law and a weld width of 1.6 mm, the specific resistance of the weld was determined [24]. In addition, the Eddy current method was used in the middle of the weld seam (SIGMATEST 2.067, Foerster). The radius of the measurement area was 16 mm.

3 Results and discussion

3.1 Macroscopic phenomena

3.1.1 Definition of speed ranges

The beam currents to fully penetrate the plates were determined for different welding speeds. Figure 3 shows the required beam currents to fully penetrate the plate for different welding speeds, when the specimen was at room temperature prior to welding (without preheating) and after preheating according to Sect. 2.2.

Fig. 3
figure 3

Required beam current \(I_b\) to fully penetrate a 4-mm-thick plate for different welding speeds v with and without preheating (\(U_{acc}\) = 150 kV)

Full size image

Both curves agree well with a linear relationship. As to be expected, the required beam current is lower for the preheated specimens. The line energy E, calculated according to Eq. 1 from the beam currents and the respective welding speed, yields the curves displayed in Fig. 4.

Fig. 4
figure 4

E to penetrate a 4-mm-thick plate for a different welding speeds (v), b for reciprocal welding speeds (1/v)

Full size image

In Fig. 4a, E is plotted against the welding speed. For plates at room temperature, the required line energy to penetrate the plate shows a hyperbolic curve. The line energy remains nearly constant for high welding speeds. Plotting the required line energy against the reciprocal welding speed yields Fig. 4b. Clearly, the required energy to penetrate the plate depended on the time spent per weld length. The curves can be divided into two distinct areas (see areas A and B in Fig. 4). In area A, at higher time unit per weld length, the line energy closely follows a linear relationship on the reciprocal welding speed for both, the specimens with and without preheating. The slopes of the regression lines, \(2115\,\text {J}\,\text {s}^{-1}\) and \(602\,\text {J}\,\text {s}^{-1}\) respectively, can be interpreted as an integral power loss \(P_l\) of the keyhole within the welding process. \(P_l\) was far lower for welds done with preheating, because smaller amount of heat conducted into the base metal. As welding was done in vacuum, only radiation occurred at the surface of the plate. The resulting power losses were negligible. Therefore, the radial power losses of the keyhole and its surrounding melt pool (assuming the axial direction is along the penetration depth) determined \(P_l\). These losses mainly depended on the surface area to the base metal for a set keyhole and melt pool geometry [57, 58].

When the welding speed was increased above a critical value \(v_{c}\), i. e. the time spent per unit weld length was decreased below a critical value \(1/v_c\), \(P_l\) becomes zero (area B in Fig. 4b). The energy to penetrate the plate then was independent of the time spent per weld length. The critical speed was determined by inserting the value of the line energy from area B (Fig. 4) in the linear regression equation from area A. This yielded a critical welding speed of \(v_c \approx 320\)–\(350\,\text {mm}\,\text {s}^{-2}\).

This transition can be explained by the change of radial energy loss. As the welding speed increased, the time for radial heat flow diminished (see Fig. 4b). Furthermore, the heat flux due to the released heat of fusion increased, which decreased the temperature gradient at the rear of the melt pool [59].

As long as the line energy to penetrate the plate depended on the time per weld length (\(v < v_c\)), the radial heat conduction determined the required beam power for full penetration. Hence, this speed regime (area A in Fig. 4) is called the conduction-determined weld speed regime (CD speed regime). If \(v > v_c\) (area B in Fig. 4) the interrelationships were vice versa, hence the speed regime is called non-conduction-determined weld speed regime (NCD speed regime).

Fig. 5
figure 5

Geometrical parameters of the fully penetrated welds without preheating for different welding speeds (v) with MAD

Full size image
Table 2 Results of the statistical analysis of the weld geometry for \(v < v_c\)
Full size table

3.1.2 Weld geometry

For \(v < v_c\), in the CD speed regime without preheating, the integral power loss was shown to be constant. If the constancy of \(P_l\) is mainly attributed to a change in radially conducted heat, there should be no significant changes in the dimensions of the weld pool at a given plate temperature with a varying weld speed. In Fig. 5, the width of the nail head (\(w_{NH}\), width of the weld at the top of the plate) and the width at half penetration depth (\(w_{WS}\)) are plotted with the mean absolute deviation around the mean (MAD) as error bars for different welding speeds.

Statistical analysis in Table 2 shows that \(w_{NH}\) exhibited no significant differences for lower and higher welding speeds.

This indicates that the weld pool formation, which was mainly determined by heat conduction processes at the top of the plate, was nearly identical for different welding speeds in the CD speed regime. \(w_{WS}\) shows significant differences between lower and higher welding speeds (Table 2). However, in contrast to \(w_{NH}\), \(w_{WS}\) is far more sensitive to fluctuations of the keyhole, the absolute beam power and differences in input power as the closing forces are higher [60, 61]. It has also to be mentioned that the determination of a fully penetrated weld was somewhat subjective. This can also result in deviations, especially for sensitive parameters. Hence, these differences are not seen as a measure for the radial power loss due to heat conduction.

If the heat flux due to the released heat of fusion is far smaller than the amount of energy which is transported radially and the amount of heat conduction welding is nearly identical, the length of the trailing weld pool does not depend on v in the CD speed regime. As the thermal conductivity of the base metal is very high (\(394\,\text {W}\,\text {m}^{-1}\,\text {K}^{-1}\) [62]), a small heat flow due to the released heat of fusion, by comparison, is reasonable to be assumed (for more details see [59]) as long as \(v<<v_c\). It has to be mentioned that \(w_{NH}\) also depends on the convective heat transport. For high alloyed steel, Peclet numbers between 1 and 15 were reported [59, 60]. However, this number should be lower for pure copper, due to the far higher thermal conductivity (for details regarding the Peclet number during keyhole welding see [63]). Nevertheless, an increase of convective heat transport with increasing v might contribute to a constant \(P_l\).

In the case of \(v > v_c\), in the NCD speed regime without preheating, \(w_{NH}\) decreases and \(w_{WS}\) increases (Fig. 5). The former indicates that the radial heat conduction was diminished. The latter might be a result of an increased absolute power density.

3.1.3 Upper bead and root formation

The formation of the upper bead and root of the welds in the respective welding speed regime was investigated by stereo microscopy. The spikes were analyzed by light optical microscopy and ultrasonic testing. Figure 6 shows a representative upper bead and a representative longitudinal section with spiking defects of a partially penetrated weld (Fig. 6a and c). In Fig. 6b, the mean number of spikes per millimeter weld, along with the corresponding mean spiking frequency for different welding speeds in the CD speed regime without preheating is displayed. Figure 7 shows the upper bead and the root for a fully penetrated weld at \(v = 40\,\text {mm}\,\text {s}^{-1}\) without preheating. In the previous section it was shown that the effect of welding speed on \(w_{NH}\) was small in the CD speed regime. When welding in the CD speed regime (\(v<v_c\)) without preheating the scaling of the upper bead had a semi-circular shape for all tested welding speeds (Figs. 6a and 7a). Hence, the effect of the welding speed on the shape of the rear end of the weld pool at the surface was small in the CD speed regime.

Fig. 6
figure 6

Representative a upper bead at \(v = 30\,\text {mm}\,\text {s}^{-1}\) for a partially penetrated weld in the CD speed regime, b mean number of spikes per millimeter of weld and the mean resulting spiking frequency in dependency of the welding speed with MAD, c representative spiking for a partial penetration weld at \(v = 30\,\text {mm}\,\text {s}^{-1}\) without preheating

Full size image
Fig. 7
figure 7

a Upper bead and b root at \(v = 40\,\text {mm}\,\text {s}^{-1}\) for a fully penetrated weld in the CD speed regime without preheating

Full size image

The roots for partially and fully penetrated welds without preheating in the CD speed regime exhibited defects (Figs. 6c and 7b). In the former case, this can be seen by significant root porosity caused by spiking (Fig 6c). The number of spikes per unit weld length was nearly constant for all tested welding speeds (Fig. 6b). Hence, the mean spiking frequency scaled linear with the welding speed (Fig. 6b). This means that the time scale for the spiking phenomena to occur was inversely proportional to the welding speed, which is in good agreement with results reported in literature [31, 43]. For fully penetrated welds, there was a visible formation of a root (see arrows in Fig. 7b). However, it was disrupted by periodical, circular melt ejections. Comparing the porosity caused by spiking in Fig. 6c to the craters in Fig. 7b yields the following conclusion: The disruption of the root for fully penetrated welds was due to an increase of recoil pressure, which blows out the melt at the bottom side of the plate. The literature agrees very well on the fact that spiking was caused by an increase in recoil pressure [29, 30, 32,33,34, 40, 43]. Therefore, it is reasonable to assume that the disruption of the root for full penetration welds and the observed spiking for partial penetration welds are caused by the same phenomena. Usually root defects can be avoided by fully penetrating the plate [10, 37, 64, 65]. However, this was not possible for pure copper. In literature, the main reasons reported for this effect are a small solidification temperature range, a low surface tension and viscosity combined with a high thermal conductivity [23, 24].

Figure 8 shows both the upper bead and the root of a weld without preheating at a very high welding speed (\(v = 530\,\text {mm}\,\text {s}^{-1}\)) in the NCD speed regime.

Fig. 8
figure 8

a Upper bead and b root of a representative weld with \(v = 530\,\text {mm}\,\text {s}^{-1}\) without preheating

Full size image
Fig. 9
figure 9

Upper bead and root of welds at \(v = 40\,\text {mm}\,\text {s}^{-1}\) with preheating at a, b \(I_b = 8.5\,\text {mA}\) and c, d \(I_b = 10.0\,\text {mA}\)

Full size image

In contrast to the results in the CD speed regime without preheating, there was a visible root formation for welds made in the NCD speed regime (compare Figs. 7b and 8b). However, the root formation was accompanied by melt ejections at the upper bead and humping at the root (Fig. 8a and b). It can be assumed that the very high power density required to achieve the penetration leads to melt eruptions, when interacting with the closing melt as shown earlier [66, 69–73]. Humping was frequently reported in literature and extensive work has been done to understand the physical phenomena behind it [31, 67, 68]. It was shown that root humps are formed, when the weld pool spreads at the root [65]. This may be also the reason for the root formation in the NCD speed regime. The topography of the root in Fig. 8b shows that the melt was ejected from the bottom of the plate. Also, typical splatters were observed at the bottom of the plate after welding. However, in contrast to welds made in CD speed regime, for \(v > v_c\) the weld pool was elongated due to the high heat flux caused by the released heat of fusion [59]. Therefore, molten metal could flow into the cavities of the previously shown spikes (see Fig. 6c). This also contributed to the volume deficit at the upper bead [65].

Figure 9 shows the upper bead and the root of welds at \(v = 40\,\text {mm}\,\text {s}^{-1}\) in the CD speed regime with preheating at different beam currents (\(I_b = 8.5\,\text {mA}\) and 10.0 mA).

Fig. 10
figure 10

Thermographic measurements of the melt pools for welds a without preheating and b with preheating and the associated temperature profiles with respect to \(\textbf{v}\) at different points in time of the welding process (\(0.1\,s\) and \( 1\,s \)) along the c y-axis and d x-axis at \(v = 40\,\text {mm}\,\text {s}^{-1}\), arrows mark plateaus due to latent heat accumulation. Each shown melt pool resembles a time step (\(t_0 - t_{14}\)) of12.5 ms

Full size image

Both welds show an upper bead without humping (see Fig. 9a and c) as well as a root free of craters caused by spiking (see Fig. 9b and d). The proper root formation for preheated welds can be explained by the changes in radial power loss (shown in Sect. 3.1.1). As mentioned in the literature, pure copper exhibits unfavorable thermophysical properties for EBW, i.e., high thermal conductivity, low viscosity and low surface tension [23, 24]. This means that due to high radial power loss without preheating (Sect. 3.1.1) a very high beam power is required to achieve a desired penetration depth. This results in an overheated weld pool and high evaporation rates, i.e., high recoil pressures. In literature, the former is reported to yield sagging at the root of an Al alloy [65]. In this context, the shown melt ejections are just a case of sagging due to an overheated weld pool, amplified by increases in recoil pressure due to the phenomena causing spiking. The relatively high density of copper, which results in a high hydrostatic pressure, fits in this assumption as well. By decreasing the radial power loss via preheating, the required beam power was significantly reduced (shown in Fig. 3). A lower required beam power reduces the aforementioned overheating of the weld pool and the high evaporation rates. The former increases the viscosity and the surface tension of the melt. The latter reduces the recoil pressure. Thus, the melt is not blown out at the bottom of the plate and a root was formed. A further increase of the beam power (increasing \(I_b\) from 8.5 mA to 10.0 mA) results in melt ejections and splatters (compare Fig. 8b and d, arrows mark melt ejections).

To further investigate the influence of welding speed and preheating on the upper bead and root formation, thermographic imaging was done in situ on welding processes with and without preheating using a thermographic camera. Figure 10 shows the weld pools and their respective temperature profiles at the top of the plate of welds with and without preheating in the CD speed regime (\(v = 40\,\text {mm}\,\text {s}^{-1}\)).

Fig. 11
figure 11

Defect indications from ultrasonic test of the welds generated with \(v = 40\,\text {mm}\,\text {s}^{-1}\) a without preheating and b with preheating. c Defect echo for a weld with \(v = 200\,\text {mm}\,\text {s}^{-1}\) and preheating

Full size image
Fig. 12
figure 12

a Melt pool for a weld with preheating at a welding speed of \(200\,\text {mm}\,\text {s}^{-1}\) and b the associated temperature profiles along the x-axis and y-axis

Full size image

Both welds showed instabilities of the keyhole during the welding process (opening and closing of the keyhole in Fig. 10a and b). As the weld without preheating exhibited defects associated with spiking, while the weld with preheating did not, a stabilization of the keyhole seems not paramount for their prevention. This agrees well with findings of Fetzer et al. [43]. The dimensions of the weld pool for preheating were slightly larger in vertical direction (Fig. 10c). However, the profiles were rather symmetrical. Parallel to the welding direction, the temperature profiles without preheating exhibited nearly identical dimensions over the welding process (compare Fig. 10c and d). This observation proves the nearly circular shape of the weld pool and the earlier statement from Sect. 3.1.1, regarding the insignificant heat flux due to released heat of fusion in the CD speed regime. The temperature profiles changed significantly for the welds done with preheating (Fig. 10d). Firstly, the weld pool was slightly elongated, which can also be seen by comparing Fig. 10a and b. Secondly, the temperature profiles along the x-axis showed a plateau close to the liquid-phase boundary, which increased in size over the welding time (arrows in Fig. 10d) due to the accumulation of latent heat closely behind the weld pool [59]. Hence, the elongation of the weld pool could also contribute to the proper root formation on preheated welds (see Fig. 9b and d). This is sensible as a larger volume of trailing melt with lower temperature, i.e., higher viscosity and surface tension, is available for root formation.

So far the suppression of spiking defects during EBW was only shown for fully penetrated welds. For this case the spreading of the weld pool at the lower end of the plate could contribute to the suppression of defects [10, 37, 64, 65]. Hence, partial penetration welds with and without preheating were investigated using ultrasonic testing in both speed regimes. Figure 11 shows the defect indications of ultrasonic tests done at partial penetration welds. Figure 11a and b show welds at \(v = 40\,\text {mm}\,\text {s}^{-1}\) in the CD speed regime without preheating and with preheating respectively. Figure 11c shows a weld at \(v = 200\,\text {mm}\,\text {s}^{-1}\) with preheating. Regarding the latter, a thermographic image with the respective temperature profiles of the weld pool is shown in Fig. 12.

The weld without preheating exhibited the already shown root porosity due to spiking. For welds with preheating, this porosity was greatly diminished (Fig. 11b, arrows). This is understandable, as the power density was far lower compared to welds without preheating (cf. Sect. 3.1.1, for details see [25,26,27, 31, 44]). For welds generated with preheating at \(v = 200\,\text {mm}\,\text {s}^{-1}\), spiking defects were suppressed (Fig. 11c) as a result of a sufficiently elongated weld pool, cf. Fig. 12.

The length was far larger than for lower welding speeds (compare Figs. 10 and 12). For welds with preheating, the suppression of spiking defects was obtained without melt ejections or serious humping at the upper bead (Fig. 11c), because the required welding speed to achieve the elongation of the melt pool was lower compared to welds without preheating due to the reduced radial power loss. Also the required beam power and therefore the power density were lower. The latter could have been detrimental to avoid the aforementioned melt ejections.

Fig. 13
figure 13

Representative EBSD image quality maps of the microstructure at a the longitudinal section with schematic perpendiculars n\(_1 - \)n\(_3\) to the isotherm of the rear end of the weld pool and b the cross section

Full size image
Fig. 14
figure 14

Box plots of the grain area distributions for different welding speeds with test matrix for pairwise comparison by Dunn test at a the longitudinal section and b the cross section of the welds. Red — significant difference, green — no significant difference

Full size image

Based on the findings presented, it can be concluded that the suppression of spiking defects during EBW is achievable by creating a sufficiently long trailing weld pool while also minimizing the necessary beam power to achieve the desired penetration depth. The former enables the filling of pores or craters due to the keyhole oscillation. The latter reduces the recoil pressure to avoid melt ejections at the upper bead and the root. Both criteria aim at a trailing weld pool with minimal overheating, i.e., maximized viscosity and surface tension. In other studies, these criteria were achieved for a 16Cr-6Mn-9Ni steel by simply increasing the welding speed into the NCD speed regime [10]. In the case of pure copper, the addition of Sn was sufficient [24]. In this study, without a filler, a combination of preheating and a sufficiently high welding speed was necessary to achieve these conditions and suppress the spiking defects. The necessary welding speed was higher for partially penetrated welds.

3.2 Characterization of the microstructure

The microstructure of the welds generated in both speed regimes without preheating was investigated by optical microscopy and SEM. Figure 13 shows representative microstructures of welds in the CD speed regime for different viewing directions relative to the welding direction.

Figure 13a shows that the grains nucleated epitaxially form the non-melted grains (for details see [69]). It is well established that for low welding speeds the solidification speed (\(v_s\)) can be written as a function of v and the angle \(\Theta \) (angle between the perpendicular of the isotherm of the rear end of the weld pool and the welding direction, see n\(_1\) - n\(_3\) in Fig. 13a) (Eq. 2 [69]).

$$\begin{aligned} v_s = v cos \Theta \end{aligned}$$
(2)

In the context of this work, low speeds are in the CD speed regime. The orientation of the grains depicted in Fig. 13a agreed well with the relation given by Eq. 2. As the isotherms at the rear end of the weld pool were semi-circular at temperatures near the melting point of copper in the CD speed regime (cf. Fig. 10a), the solidification speed in the directions n\(_1\) and n\(_3\) in Fig. 13a was comparatively small as \(\Theta \) approached \(\pi /2\) and -\(\pi /2\) respectively. As \(\Theta \) decreases, \(v_s\) increases, reaching its maximum in direction n\(_2\) in Fig. 13a. Hence, the resulting growth direction resembles a curve when the beam passes. For details, refer to [69].

In Fig. 14a box plots of the grain area distributions along the longitudinal sections are plotted for different welding speeds in the CD speed regime.

A pairwise comparison by Dunn test showed that the grain area exhibited no significant differences for a changing welding speed, cf. Fig. 14a. As shown above, the melt solidified mainly epitaxially and Eq. 2 was valid. So, neither the nucleation mechanism nor the solidification morphology depended on \(v_s\). Hence, a change in welding speed (for \(v < v_c\)) only increased the solidification speed according to Eq. 2. The grain area distribution depended on the geometry of the weld pool. As stated above, the melt pool was of circular shape for welds generated in the CD speed regime (see Sect. 3.1.3). The weld pool exhibited significant differences in width (see Sect. 3.1.2) along the penetration depth. However, as the main growth direction was parallel to the welding direction (cf. Fig. 13a and b), these differences were insignificant with regards to the grain area.

Fig. 15
figure 15

Light optical micrograph of an etched cross section of a weld generated with \(v = 530\,\text {mm}\,\text {s}^{-1}\)

Full size image

The grain area distributions at the cross sections perpendicular to the welding direction exhibited significant differences (Fig. 14b) due to the fact that the grain area was measured in the plane perpendicular to the main growth directions (cf. Fig. 13). Therefore, the changes in width, given in Sect. 3.1.2, became significant. This effect can be understood by looking at the isotherm in Fig. 13a. When the semi-circular isotherm is widened or narrowed, the number of grains that are then cut perpendicular to their longest direction of extension, which appear as nearly equiaxed grains in the cross section (see grains in the center of Fig. 13a), changes compared to the grains that are cut parallel to their longest direction of extension (see edges of the weld in Fig. 13a).

Figure 15 shows the optical micrograph of a cross section of a weld produced in the NCD speed regime without preheating (\(v = 530\,\text {mm}\,\text {s}^{-1}\)).

In contrast to the CD speed regime (cf. Fig. 13b), the weld showed two solidification fronts, which grew together in the middle of the weld. This fact was due to the elongation of the weld pool, shown above in Sect. 3.1.3. This circumstance led to a solidification with a certain distance to the interaction point of the beam. Hence, the direction of the solidification was no longer dependent on the welding speed, i.e., Eq. 2 was no longer valid in the NCD speed regime. As heat was mainly dissipated into the base metal (see Sect. 3.1.1), the isotherms of the melt pool at temperatures near the melting point of copper became nearly parallel to the surrounding material (this effect can be seen in Fig. 12a). This type of solidification could be observed on the entire cross section of the weld (Fig. 15). Hence, the weld pool was elongated along the entire cross section of the weld.

3.3 Electrical and mechanical properties

The electrical conductivity of the welds was calculated using the measured electrical resistance of the four-point potentiostatic method for selected welds or measured using the Eddy current method. Table 3 gives the specific electrical conductivity (\(\sigma \)) of the base metal of welds produced at \(v = 40\,\text {mm}\,\text {s}^{-1}\) with and without preheating and welds in the NCD speed regime \(v = 530\,\text {mm}\,\text {s}^{-1}\) including smoothed specimens (SM) without preheating, compared to reported data for different welding processes from the literature. The width of the area, over which the specific conductivity was averaged (\(w_{m}\)), is given for comparison (if available).

Table 3 Specific electrical conductivity (\(\sigma \)) for different welds compared to the literature
Full size table

The specimen welded without preheating in the CD speed regime had a slight drop of the conductivity (\(55.2\text {MS}\,\text {m}^{-1}\)) compared to the base metal (\(57.2\text {MS}\,\text {m}^{-1}\)), which is due to the melt ejections at the root shown in Sect. 3.1.3. The eddy current measurements for the specimen with preheating in the CD speed regime showed an increased conductivity (\(58.6\text {MS}\,\text {m}^{-1}\)) compared to the base metal. However, the changes in microstructure as well as the far larger \(w_{m}\) have to be taken into account, which could result in a compensation of counteracting effects. The specimens welded in the NCD speed regime (\(v = 530\,\text {mm}\,\text {s}^{-1}\)) without preheating exhibited a greater drop due to the severe melt ejections (\(46.3\text {MS}\,\text {m}^{-1}\)). However, the conductivity was increased to \(53.8\text {MS}\,\text {m}^{-1}\) by a subsequent smoothing (see Table 3, \(530\,\text {mm}\,\text {s}^{1-}\) SM). Overall, the specific electrical conductivity was higher compared to processes which used an additional filler (MIG, LMIGHW and EBW + Sn-Layer), as the copper remained pure. Compared to the welding processes without filler, there are only small differences (the differences in \(w_{m}\) have to be taken into account) and the conductivity of the BM was nearly preserved (cf. Table 3 EBW, LBW, DPLBW).

The ultimate tensile strength and the fracture strain of selected welds were measured using tensile tests. Table 4 gives the ultimate tensile strength (UTS) and the fracture strain for welds produced with \(v = 40\,\text {mm}\,\text {s}^{-1}\) in the CD speed regime with and without preheating compared to the base metal (specifications for Cu-ETP according to DIN EN 13599) and values from the literature reported for different welding processes.

Table 4 Ultimate tensile strength and fracture strain (gauge length of 45 mm) for welds generated in the CD speed regime with preheating and without preheating compared to the base metal, values of standards (*according to DIN EN 13599) and values from the literature, MIG — Metal inert gas welding, LMIGHW — Laser MIG hybrid welding, LBW — Laser beam welding, DPLBW — Dual pass LBW, NVEBW — Non vacuum EBW, LCMTHW — Laser-cold metal transfer arc hybrid welding, FSW — Friction stir welding, CCGTAW — Constant current gas tungsten arc welding, PCGTAW — Pulsed current gas tungsten arc welding
Full size table

Both welds exhibit a lower ultimate tensile strength (UTS) compared to the base metal (\(207\,\text {MPa}\), \(216\,\text {MPa}\) and \(280\,\text {MPa}\) for preheating, without preheating and BM respectively). The fracture strain was lower for the welds produced without preheating (\(3.9\%\)), but higher for welds made with preheating (\(25\%\)) compared to the base metal (\(14.1\%\)). Fracture always took place within the weld. Nevertheless, the scatter of the mechanical properties of the welds was less than ±2%, which shows the good reproducibility of the welding process. For the welds made without preheating, the reason for the low ductility is clear, as the lower mechanical properties can be explained by the root defects. In the case of welds produced with preheating the reduction in strength can be attributed to defects as well as to softening caused by a lower dislocation density and coarse grains due to recrystallization and grain growth (preheating was done at 86% of the melting point). When comparing the welds done with preheating to the specifications, the welds exhibited slightly lower strength than the minimum value of R220 (\(207\,\text {MPa}\) compared to \(220\,\text {MPa}\)). The fracture strain was between the values of the standards of R220 and R240 (cf. Table 4). Hence, there was a negative influence of the welding process on the mechanical properties. This is reasonable, as the keyhole was shown to be unstable (Sect. 3.1.3). As the base metals often varied in their reported mechanical properties, no direct comparisons between the different welding processes are possible. It has to be noted that consistently high mechanical properties were reported for friction stir welding. However, there was no data reported concerning the specific electrical conductivity of the joints. The welding processes that utilized a high power density beam (LBW and EBW) without filler (which was shown to be crucial for maintaining a high specific electrical conductivity) without preheating, show a tendency to low fracture strains, due to the mentioned defects. The applied preheating significantly increased the fracture strain and produced a comparable tensile strength.

The mechanical properties may be further increased by optimizing the preheating process (lower temperatures, smaller affected area). This can be done by using a multi-spot technology, which would utilize a very narrow preheating zone into large penetration depths. Also, the welding process can be stabilized by the application of an appropriate beam oscillation [21, 46] or modulation [14, 47, 48] which would reduce defects.

4 Conclusions

Bead on plate EBW was carried for Cu-ETP. Two welding speed regimes were identified for full penetration welds. A critical welding speed of \(v_c \approx ~320\)–\(350\,\text {mm}\,\text {s}^{-1}\) was determined which separated the regimes. For lower speeds the required energy input per weld length was mainly influenced by the radial heat conduction. The root formation was disrupted by melt ejections, which were related to the physical phenomena causing spiking. These defects were eliminated by reducing the radial power loss due to preheating and an increase in welding speed, which elongated the weld pool and reduced the required beam power. Therefore, the overheating of the weld pool was reduced, which, in turn, increased the viscosity and surface tension of the weld pool. This effect led to proper root formation. Also porosity due to spiking was greatly reduced for partial penetration welds. By increasing the welding speed from \(v = 40\,\text {mm}\,\text {s}^{-1}\) to \(v = 200\,\text {mm}\,\text {s}^{-1}\) while preheating the weld area, spiking defects were completely avoided in partial penetration welds. Above \(v_c\), the required energy input per weld length was not determined by the radial heat conduction. The melt pool at the top and the bottom of the plate became elongated. This elongation resulted in the formation of a root. However, root humps and melt ejections at the upper bead were observed.

Electrical conductivity was slightly reduced compared to the base metal for welds made below \(v_c\) without preheating. The conductivity was lower for welds made above \(v_c\) due to the melt ejections at the top. However, conductivity was increased by a subsequent smoothing. For preheated specimens, the changes in microstructure yielded a specific electrical conductivity which was higher than the reference value of the base metal.

UTS were lower for welds made below \(v_c\) without preheating and with preheating compared to the base metal. However, the fracture strain of the preheated specimens was significantly higher than that in BM and the welds generated without preheating.