Chai Supplement February 2016_Layout 1 1/14/16 1:48 PM Page 57
WELDING RESEARCH
Liquation and Liquation Cracking in Partially
Melted Zones of Magnesium Welds
Magnesium alloys are susceptible to liquation cracking, but the crack susceptibility
can be predicted and eliminated based on a simple criterion
BY X. CHAI, T. YUAN, AND S. KOU
ABSTRACT
Liquation, that is, liquid formation, can occur in the partially melted zone of a Mg alloy dur­
ing welding and the liquid formed along grain boundaries can weaken them to cause intergran­
ular cracking, that is, liquation cracking. The present study investigated liquation and liquation
cracking in arc welding of Mg alloys. The Mg alloys investigated were the most widely used Mg
wrought alloy, AZ31 Mg, and the most widely used Mg casting alloy, AZ91 Mg. The filler metals
were AZ31 Mg, AZ61 Mg, and AZ92 Mg. To evaluate the susceptibility to liquation cracking,
bead­on­plate circular welds were made by gas metal arc welding. The microstructure of the
partially melted zone was examined to check for evidence of liquation and liquation cracking.
AZ91 Mg was found to be much more susceptible to liquation than AZ31 Mg, and the
difference was explained based on their distinctly different microstructures. The curves of tem­
perature T vs. fraction solid fS during solidification were plotted for the workpiece and the weld
based on their compositions. The prediction and elimination of the crack susceptibility based
on comparing T­fS curves were demonstrated.
KEYWORDS
• Liquation Cracking • Partially Melted Zone • Magnesium Alloys • AZ31 Mg
• AZ91 Mg • Gas Metal Arc Welding (GMAW)
Introduction
Magnesium (Mg) alloys have been
increasingly used for vehicle weight reduction (Refs. 1–3), the density of Mg
(1.7 g/cm3) being about one-third less
than that of Al (2.7 g/cm3), and the
weldability of Mg alloys has become an
important issue in welding (Refs. 4–7).
Gas metal arc welding (GMAW) has
been widely used for welding sheet
metals, such as Al alloys, steels, and
stainless steels. But its use for welding
Mg sheets has long been delayed by severe spatter. Wagner et al. (Ref. 8)
showed recently that, due to the low
Mg density, the Mg filler metal melts
easily to form a globule and the glob-
ule is too light to detach by gravity.
When the globule finally touches the
weld pool and causes short circuiting,
the current rises sharply, causing the
reinitiating arc to heat up suddenly,
expand (explode), and expel the globule as spatter. In conventional GMAW,
it is not unusual for short circuiting to
cause spatter. However, it is the huge
Mg globule providing extra liquid Mg
for the arc to expel that makes the
spatter so severe. Although the surface
tension of Mg is relatively low (lower
than that of Al), the very large but
light globule can be pushed by the arc
to become nearly horizontal and continue to grow bigger before short circuiting. With a process controller,
called controlled short circuiting (CSC)
(Ref. 9), Wagner et al. (Ref. 8) kept the
current from rising sharply and effectively eliminated spatter.
Wagner et al. (Ref. 8) and Chai et al.
(Ref. 10) discovered that more defects
can occur persistently in GMAW of Mg
alloys (even with CSC), including hydrogen porosity, oxide-film entrapment inside butt-joint welds, high
crowns on butt-joint welds. and fingers from lap-joint welds. They established mechanisms to explain how the
unusual physical and chemical properties of Mg cause these defects and how
to eliminate them effectively. These
Mg properties include the high reactivity with moisture in air to form
Mg(OH)2 (hydrogen porosity), low
formability (sheets chipping along
edges during shearing before welding),
and high reactivity with oxygen in air
to form MgO (oxide-film entrapment
inside butt-joint welds), low density
and hence low fluidity (high crowns),
and low density (fingers).
With these five defects eliminated
effectively, the potential of GMAW as
a versatile process for welding Mg alloys has greatly increased. Gas metal
arc welding is simple, inexpensive, and
capable of mass producing highquality welds. Unlike friction stir welding, it requires no rigid clamping or
backing and it makes fillet welds easily. It can help accelerate the use of Mg
alloys as the lightest structural metal.
The present study focuses on a sixth
defect in welding Mg alloys — liquation cracking.
Since Mg alloys are similar to Al alloys in several physical and metallurgi-
X. CHAI is graduate student and S. KOU is professor, Department of Materials Science and Engineering, the University of Wisconsin, Madison, Wis. T. YUAN is
graduate student, School of Materials Science and Engineering, Tianjin University, Tianjin, China.
FEBRUARY 2016 / WELDING JOURNAL 57-s
Chai Supplement February 2016_Layout 1 1/14/16 1:48 PM Page 58
WELDING RESEARCH
high thermal diffusivity, wide
melting temperature range with
respect to their
relatively low liquidus temperature, and low eutectic temperature (Refs. 15,
16). Thus, like Al
alloys, Mg alloys
are likely to be
susceptible to liquation and liquation cracking.
Liquation
cracking in Mg alloys has been reFig. 1 — Top view of workpiece before welding.
ported (Refs.
17–20) even in
friction stir weldcal properties, the studies of Kou and
ing (Refs. 21, 22). Recently, Yuan et al.
coworkers on liquation and liquation
(Ref. 23) studied liquation cracking in
cracking of Al alloys (Refs. 11–14) can
butt-joint welds between dissimilar
help explain liquation and liquation
Mg alloys by circular-patch welding. A
cracking in Mg alloys. Like Al alloys,
circular patch of one Mg alloy was
Mg alloys tend to have low temperaplaced inside a round hole in the workture gradients normal to the fusion
piece of a different Mg alloy and weldboundary and hence a wide partially
ed in a butt-joint configuration. The
melted zone because of their relatively
present study involved no dissimilar
Mg welding. Instead, a workpiece
without a hole was welded by bead-onplate welding to study both liquation
and the effect of the filler metal on liquation cracking.
Experimental Procedure
Both wrought and cast Mg alloys
were investigated, including the most
widely used Mg wrought alloy AZ31
and the most widely used Mg casting
alloy AZ91. Magnesium alloys are
known to have low formability because
of their hexagonal closed packed (hcp)
structure (Ref. 24). Thus, the availability of Mg wrought alloys is limited,
and Mg casting alloys often have to be
used. This is why it is desirable, in
studying the weldability of Mg alloys,
to include Mg casting alloys. The compositions of the materials used for
welding are listed in Table 1 (Refs. 25,
26). Plates of AZ31 Mg were cut from
a 3.2-mm-thick AZ31 Mg sheet while
plates of AZ91 Mg were cut from an
as-cast ingot. They were machined to
desired dimensions.
A Jetline Engineering weld process
controller called CSC-MIG (Ref. 9) was
connected to a Miller Electric Invision
456P, a power source for conventional
GMAW. The parameters used in CSCGMAW are shown in Table 2. The waveforms of the welding current and arc
voltage were recorded using a computerbased data acquisition system and the
software LabView. The data-sampling
rate for each signal was 15,000 Hz.
It was found that a workpiece of
102  102  1.6 mm tended to distort
Table 1 — Nominal Compositions of Materials Used for Welding (Refs. 25, 26)
Al (wt­%)
Zn (wt­%)
Mn (wt­%)
Mg (wt­%)
Workpiece
AZ31
AZ91
3.0
9.0
1.0
0.7
0.6
0.2
balance
balance
Welding wires
AZ31
AZ61
AZ92
3.0
6.5
9.0
1.0
1.0
2.0
0.6
0.3
0.3
balance
balance
balance
Table 2 — Parameters Selected for CSC Welding
CSC Welding Parameters
Arc Phase
Current
(A)
Short Circuit Phase
Wire Feeding
Current
(A)
Delay
(ms)
Feed
Rate
(m/min)
Delay
(ms)
Current (A)
0
58
CT Opening
(mm)
Argon
Flow
(L/min)
16
16.5
Delay
(ms)
Preheating
start
44
4
start
96
2.5
down
8
0
mid
end
54
46
6
mid
end
106
85
3.0
up 1
up 2
8
8
6
P Delay: penetration delay
CT Opening: distance between contact tube and workpiece
ms: millisecond
m/min: meter per minute
L/min: liter per minute
58-s WELDING JOURNAL / FEBRUARY 2016, VOL. 95
Delay (ms)
90
P Delay
(ms)
Arc
Length
(mm)
0.2
Chai Supplement February 2016_Layout 1 1/14/16 1:48 PM Page 59
WELDING RESEARCH
A
B
Fig. 2 — Bead­on­plate circular welding experiment: A — Workpiece; B — vertical cross
section of the apparatus.
welding gun was stationary with its tip
positioned 25.4 mm away from the center of the workpiece, so, while the
turntable rotated, a circular weld was
made with the radius of 25.4 mm.
The diameters of the wires are also
shown in Table 3. The diameter of the
AZ31 welding wire was 1.2 mm. However, welding wires AZ61 and AZ92
were older and had to be cleaned with
sandpaper and washed with acetone
before welding in order to avoid gas
porosity. This reduced the wire diameters slightly. The AZ61 wire diameter
was 1.16 mm, and the AZ92 wire 1.19
mm.
The linear travel speed in mm/min
was 2  25.4 mm  (rotation speed in
rpm). Since the rotation speed was 1.6
rpm, the linear travel speed was 255
mm/min in all experiments. The values of the average power P in Table 3
were calculated using the following
equation:
P=
t
( I E )dt / t
( 1)
0
Fig. 3 — Volume of the workpiece Vworkpiece and volume of the filler metal Vfiller in the
weld. The broken lines indicate the boundaries of the workpiece before welding.
Table 3 — Conditions for Circular Welding Experiments
Workpiece
Filler Metal
Wire Size
(mm)
Travel Speed
(mm/min)
Power
(W)
AZ91
AZ91
AZ31
AZ31
AZ31
AZ92
AZ61
AZ92
1.20
1.19
1.16
1.19
255
255
255
255
686
710
935
819
after welding. So, beyond the thinner
inner square of 70  70  1.6 mm, the
workpiece was 3.2 mm thick to prevent distortion, as shown in Fig. 1.
Figure 2 shows schematically the
circular welding test. To keep the
workpiece from contracting during
welding, it was bolted down tightly
with a torque wrench to a thick copper
plate and the thick stainless steel base
plate below it. This allowed severe tension to be induced along the outer
weld edge to cause cracking. No
cracking occurred along the inner
edge of the weld because it was in
compression.
Table 3 shows the conditions used
for circular welding experiments. The
apparatus was laid flat on a positioner
to rotate at the predetermined speed
during welding. The welding position
was flat and the welding gun was vertical with a distance of 16 mm between
its contact tube and the workpiece. The
where I is the current, E is the voltage,
and t is the welding time. The differences in the wire physical properties
and diameter contributed to the differences in the power.
Figure 3 shows the volume of the
workpiece in the weld Vworkpiece and that
of the filler metal in the weld Vfiller. The
total volume of the weld Vweld = Vworkpiece
+ Vfiller. Thus, the ratios Vworkpiece/Vweld
and Vfiller/Vweld represent the volume
fractions of the workpiece and the
filler metal in the weld, respectively.
The ratio Vworkpiece/Vweld is called the dilution of the filler metal by the workpiece (Ref. 15). The higher it is, the
more the filler metal is diluted by the
base metal. Based on these volume
fractions and the compositions of the
workpiece and filler metal (shown in
Table 1), the composition of the weld
can be calculated as follows:
(wt-% E)weld = (Vworkpiece/Vweld)
(wt-% E)workpiece + (Vfiller/Vweld)
(wt-% E)filler
(2)
where (wt-% E)weld, (wt-% E)workpiece, and
(wt-% E)filler are the wt-% of element E
in the weld, workpiece, and filler metal, respectively.
It should be pointed out that Equation 2 is based on the assumption of
FEBRUARY 2016 / WELDING JOURNAL 59-s
Chai Supplement February 2016_Layout 1 1/14/16 1:48 PM Page 60
WELDING RESEARCH
satisfactorily.
The welds
were cut in the
A
transverse direction, polished,
and etched in order to determine
volumes Vworkpiece,
Vweld, and Vfiller.
For AZ91 Mg,
they were then
first etched for 3
s in a solution of
3 mL HCl (concentrated) + 100
mL ethanol and
then etched for
another 3 s in a
solution of 5 mL
acetic acid + 5 g
B
picric acid + 10
mL distilled water + 100 mL
ethanol. As for
AZ31B, they
were etched for
15 s in a solution of 5 mL
acetic acid + 6 g
picric acid + 10
mL distilled water + 100 mL
ethanol. After
etching, transverse macrographs of the
Fig. 4 — Top views of example welds: A — Weld showing liquation
welds were takcracking along the outer weld edge; B — weld showing no cracking.
en using a digital
camera with a
complete mixing in the weld pool,
close-up lens and bellows. With the
which has been verified by Houldcroft
help of computer software, the macro(Ref. 27) in single-pass welds made by
graph was used to determine Vworkpiece
GMAW, and which is used routinely in
and Vweld. The thickness of the workthe calculation of the weld composipiece Hworkpiece was measured and used
tion. The very fast Lorentz forcerespectively in the calculations of
driven flow (~ 10 cm/s) and MarangoVworkpiece, that is, Vworkpiece = Hworkpiece
ni flow (~ 1 m/s) (Refs. 28–31) suggest
((Rb)2 – (R a)2). In view of the fact the
mixing across the weld pool tens or
vertical thickness of the weld varied in
hundreds of times before solidificathe radial direction R, the weld was dition. In the previous studies of Kou
vided into about a dozen rings of equal
and coworkers (Refs. 16, 32–34) on
DR. The volume of each ring was calculiquation cracking of Al alloys, this
lated. The summation of the volumes
assumption was found to work
of these rings was used to calculate the
volume of the weld Vweld. The volume
of the filler metal in the weld can be
calculated by Vfiller = Vweld – Vworkpiece. The
cross sections were also examined by
optical microscopy and scanning electron microscopy (SEM) to show liquation inside the partially melted zone.
Results and Discussion
Liquation
Two examples of the circular welds
made in the present study are shown
in Fig. 4. Figure 4A shows the weld
made by welding AZ91 Mg alloy with
an AZ31 Mg filler metal. Cracking is
clear along the outer edge of the weld.
However, no cracking is visible along
the inner edge. This is because, as the
weld shrank during welding, the outer
edge was in tension and the inner edge
in compression. Figure 4B, on the other hand, shows the weld made by welding AZ31 Mg alloy with an AZ92 filler
metal. No cracking is visible either
along the outer or inner edge. The
weld is narrower as compared to that
in Fig. 4A mainly because the AZ31
Mg workpiece has a higher liquidus
temperature than the AZ91 Mg
workpiece.
The microstructure of the partially
melted zone in AZ91 Mg alloy is
shown in Fig. 5. The weld was made by
welding AZ91 Mg with an AZ31 Mg
filler metal. In Fig. 5A, the boundary
between the fusion zone and the partially melted zone is indicated by the
broken line. Near the fusion boundary,
cracking along grain boundaries is visible. A SEM image of the base metal
taken far away from the fusion boundary, beyond the right-hand side of the
area covered by Fig. 5A, is shown in
Fig. 5B. Figure 5C shows a SEM image
of the partially melted zone. An optical micrograph at the borderline between the partially melted zone and
the base metal is shown in Fig. 5D
(where the peak temperature during
welding was the eutectic temperature).
Table 4 — Fractions of Phases in As­Cast AZ31 Mg Alloy
Phase
Al12Mg17
AlMgZn_phi
Al8Mn5
AlMgZn_tau
MgZn
Al11Mn4
Al4Mn
Fraction
2.10  10–2
5.5  10–3
3.87  10–3
3.20  10–4
9.66  10–4
1.19  10–5
5.65  10–6
60-s WELDING JOURNAL / FEBRUARY 2016, VOL. 95
Chai Supplement February 2016_Layout 1 1/14/16 1:48 PM Page 61
WELDING RESEARCH
A
B
C
D
Fig. 5 — Transverse micrographs of as­cast AZ91 Mg weld made with AZ31 Mg welding
wire: A — Optical micrograph (broken line indicating fusion boundary); B, C — SEM im­
ages and results of EDS analysis; D — optical micrograph at border line between partially
melted zone and base metal, which is beyond the right­hand side of A.
Figure 5D was taken from an area beyond the right-hand side of Fig. 5A.
In order to explain the microstructure in Fig. 5, the Mg-rich side of the
binary Mg-Al phase diagram (Ref. 35)
in Fig. 6 will be used as an approximation. According to the phase diagram,
the eutectic temperature is 437°C.
Based on the phase diagram, the microstructure of an as-cast binary Mg9Al alloy should be a matrix of the Mgrich phase  embedded with an eutectic consisting of  and  (Mg17Al12).
The SEM image of the base metal in
Fig. 5B shows a divorced eutectic in
the form of coarse  (Mg17Al12) particles instead of an ordinary compositelike structure of the / (Mg17Al12) eutectic. In other words, the  phase of
the eutectic is connected to and thus
indistinguishable from the surrounding  matrix. The fine needle-like precipitates are the  (Mg17Al12) phase
that forms from the  phase, not from
the liquid, along some specific orientations as temperature falls below the
solvus line. The composition analysis
by EDS indicates that point 1 is close
to the Mg-rich  phase in composition
and point 2 close to  (Mg17Al12). According to the Mg-Al phase diagram,
the  phase consists of about 58 wt-%
Mg and 42 wt-% Al.
The SEM image of the partially
melted zone in Fig. 5C shows the matrix of the Mg-rich phase  is embedded with an ordinary composite-like
structure of the / (Mg17Al12) eutectic. The composition analysis by EDS
indicates that point 3 is close to the
Mg-rich  phase in composition and
point 4 close to eutectic / (Mg17Al12)
in composition, which consists of
about 67 wt-% Mg and 33 wt-% Al.
The eutectic formed here by the constitutional liquation mechanism was
originally proposed by Pepe and Savage (Refs. 36, 37). Essentially, due to
rapid heating during arc welding, the
coarse  (Mg17Al12) particles did not
have time to dissolve completely when
the local temperature rose above the
solvus. The fine  (Mg17Al12) precipitates most likely dissolved completely
before reaching the eutectic temperature. Upon reaching the eutectic temperature TE, the remaining coarse 
(Mg17Al12) particles reacted with the
surrounding  matrix to form liquid,
that is, to cause liquation. This is indicated by the eutectic reaction in the
phase diagram as  +   LE, where LE
is the eutectic liquid. Thus, the coarse
 (Mg17Al12) particles can be considered as the low melting point segregates along grain boundaries (and
within grains). Upon further heating
to above TE, more  dissolves into the
eutectic liquid to make it hypoeutectic
FEBRUARY 2016 / WELDING JOURNAL 61-s
Chai Supplement February 2016_Layout 1 1/14/16 1:48 PM Page 62
WELDING RESEARCH
Fig. 6 — Binary Mg­Al phase diagram (Ref. 35).
A
C
B
Fig. 7 — Transverse micrographs of a circular weld of AZ31 made with AZ92 welding wire:
A — Transverse cross section (the inset enlarging the boxed area and broken lines indicating
fusion boundary); B — SEM image of base metal; C — SEM image near fusion boundary.
62-s WELDING JOURNAL / FEBRUARY 2016, VOL. 95
(less solute than the eutectic composition). Upon subsequent cooling,
the hypoeutectic liquid solidified
first as the  phase and last as the
/ (Mg17Al12) eutectic.
The microstructure of the partially
melted zone in AZ31 Mg alloy is
shown in Fig. 7. The weld was made
with an AZ92 Mg filler metal. Figure
7A is a transverse macrograph of part
of the weld, again with a broken line
indicating the fusion boundary. No
cracking is visible. The fractions of intermetallic phases that form during
the solidification of AZ31 Mg, calculated using Pandat (Ref. 38) and PanMagnesium (Ref. 39), are summarized
in Table 4. As shown, about 2% of
 (Mg17Al12) can be present in an ascast AZ31 Mg. However, no 
(Mg17Al12) could be found in the base
metal by optical or scanning electron
microscopy. This is because the AZ31
Mg is a wrought alloy, and it is normally heat treated after casting to
dissolve  (Mg17Al12). The phase diagram in Fig. 6 shows that AZ31 Mg,
which is essentially Mg-3Al alloy, can
be solutionized in the Mg-rich phase 
to dissolve  (Mg17Al12). Some particles
are visible in the partially melted zone
in Fig. 7A and the inset enlarges the
boxed area in the same figure to show
the particles more clearly. In the inset
the region to the right of the broken
line that contains scattered particles
represents the partially melted zone
and the particles were liquid during
welding. Figure 7B shows the SEM image of the AZ31 Mg base metal. No
 (Mg17Al12) is visible. The composition
analysis by EDS indicates that point 1
is close to the Mg-rich  phase in composition and point 2 an intermetallic
compound consisting of essentially Al
and Mn alone. According to Table 4,
the Al-Mn intermetallic compound at
point 2 is most likely Al8Mn5. Al11Mn4
and Al4Mn are two and three orders of
magnitude lower in quantity, respectively. Based on the atomic weights of
Al (26.98 g/mole) and Mn (54.94
g/mole), Al8Mn5 consists of 44 wt-% Al
and 56 wt-% Mn, which are close to
the values of 44.52 wt-% Al and 52.9
wt-% Mn measured by EDS.
Figure 7C is a SEM image taken
close to the fusion boundary. Both
particles are composite-like, suggesting liquation by eutectic reactions between the Mg-rich  matrix and some
Chai Supplement February 2016_Layout 1 1/14/16 1:48 PM Page 63
WELDING RESEARCH
A
C
B
D
Fig. 8 — Criterion for susceptibility to cracking along weld edge: A — Phase diagram
showing alloy Co (workpiece); B — microstructure around weld pool in welding of alloy
Co; C — cracking likely in welding of an alloy; D — cracking unlikely in welding another
alloy with another filler metal. Rectangular box in B is enlarged in C, D.
very small intermetallic particles. The
composition analysis by EDS indicates
that point 3 is rich in Mg, Zn, and Al.
It is likely to be the eutectic reaction
product between the  matrix and an
AlMgZn-phi or AlMgZn-tau particle
that still existed in the workpiece before welding. It is more likely to be the
former because it is about ten times
more abundant according to Table 4.
The particle at point 4 consists of
much Al and Mn and some Mg. It is
perhaps the reaction product between
the  matrix and an Al8Mn5 particle.
In any case, there is much less liquation in the partially melted zone of
AZ31 Mg than that of AZ91 Mg. There
is no evidence of constitutional liquation caused by  (Mg17Al12) particles as
in the case of AZ91 Mg.
As explained above, the differences
between Figs. 5C and 7C are mainly
caused by coarse  (Mg17Al12) particles
reacting with the Mg-rich  matrix to
form much liquid, which is present in
the as-cast AZ91 Mg alloy but not the
wrought AZ31 Mg alloy. The liquid is
enough to form continuous films
along grain boundaries to separate
grains and weaken the partially melted
zone of AZ91 Mg. In the case of AZ31
Mg, if any liquid forms in the partially
melted zone, it is very small in quantity and discontinuous, thus still allowing firm bonding between grains.
Thus, the as-cast AZ91 Mg alloy
can be expected to be more susceptible
to liquation cracking.
Liquation Cracking
The extents of cracking in the circular welds are shown in Table 5, along
with the levels of dilution and the com-
positions of the welds. In the case of
welding AZ91 Mg with filler metal AZ31
Mg, the outer edge cracked along 66.7%
of its circumference. In order to explain
the results, Fig. 8 illustrates the interaction between weld (S+L), i.e., the mushy
zone, and workpiece (S+L), i.e., the partially melted zone. Kou and Le (Refs. 16,
40) established the semisolid microstructure around the weld pool illustrated in Fig. 8B by quenching the weld
pool and its surrounding area with water during welding and by using the
phase diagram.
The mechanism of liquation cracking in the partially melted zone can be
explained based on the semisolid
microstructure around the weld pool.
With the bonding between grains
weakened by significant liquid formation along grain boundaries, such as
that illustrated in Fig. 8C, cracking can
occur in the partially melted zone if
significant tension exists. This cracking is called liquation cracking. It occurs along the edge of the weld and
hence can also be called weld-edge
cracking. The tension in the partially
melted zone is induced by the solidifying and contracting weld pool that is
much larger than and immediately
next to the partially melted zone. Contraction of the weld pool, in turn, is induced by solidification shrinkage and
thermal contraction. When the weld
pool solidifies, it shrinks because the
solid has a higher density than the liquid, about 6.6% for Al and 4.2% for
Mg (Ref. 41). The solidified weld metal
contracts during cooling because of
contraction due to its thermal expansion coefficient.
Based on the liquation cracking
mechanism, Kou and coworkers (Refs.
32–34) proposed the following criterion for the susceptibility of Al alloys to
liquation cracking:
Susceptible if weld fS > workpiece fS (3)
This criterion is illustrated in Fig.
Table 5 — Cracking Extent, Dilution Levels, and Compositions of Welds
Workpiece
Filler
Metal
Cracking
along Outer Edge
Dilution by
Workpiece (%)
AZ91
AZ91
AZ31
AZ31
AZ31
AZ92
AZ92
AZ61
66.7% cracked
no cracking
no cracking
no cracking
36.9
44.2
36.5
45.5
Al in
Weld (wt­%)
5.21
9.0
6.81
4.91
Zn in Weld
(wt­%)
Mn in Weld
(wt­%)
Mg in Weld
(wt­%)
0.89
1.43
1.64
1.0
0.45
0.26
0.41
0.44
balance
balance
balance
balance
FEBRUARY 2016 / WELDING JOURNAL 63-s
Chai Supplement February 2016_Layout 1 1/14/16 1:48 PM Page 64
WELDING RESEARCH
A
D
B
Fig. 9 — AZ91 Mg welded with AZ31 Mg welding wire: A — T­fS curves predicting crack
susceptibility; B — verification of predicted crack susceptibility. T­fS curves calculated
using Pandat (Ref. 38) and PanMagnesium (Ref. 39) of CompuTherm, LLC.
8C, D. The assumption was that the
strength and hence the crack resistance of a semisolid increases with increasing fraction solid, which has been
verified experimentally (Ref. 42). For
simplicity, it was also assumed that
the morphology of the semisolid, e.g.,
dendritic or granular, is not as important as the fraction solid. The same assumption also implies that the grain
size of the semisolid is not as important as the fraction solid. With a larger
grain size and hence a smaller grain
boundary area, the grain boundary liquid and tensile strains are concentrated in a smaller area to make cracks
open wider. A larger grain size can also
reduce the number of changes in direction per unit distance along the crack
path, thus making it easier for crack
propagation. However, the effect of
fraction solid is still expected to dominate. The prerequisites for liquation
cracking are significant liquation and
tension in the partially melted zone.
Without them a larger grain size alone
cannot cause cracking. In circular
welding, the prerequisites can be met
by welding with a sufficient heat input
to cause significant liquation in the
partially melted zone and tightly
clamping the workpiece to keep it
from contracting so that significant
strain can be induced normal to the
weld. Surprisingly, this criterion,
though very simple, works very well
for Al alloys as verified against numerous welding experiments reported in
the literature, including those conducted by The Welding Institute and
in the authors’ laboratory (Ref. 33).
Naturally, the susceptibility to liquation cracking is more significant
when the weld fS is significantly higher
than the workpiece fS. In general, a difference of 0.05 in fS is considered significant enough. According to Flemings (Ref. 41), significant strength de-
64-s WELDING JOURNAL / FEBRUARY 2016, VOL. 95
velops during solidification after fS
reaches about 0.3. Thus, Equation 3
applies after some strength has developed in the semisolid, say after fS
reaches 0.3.
The criterion is easy to use as the
curve of temperature T vs. fraction
solid fS of the weld and that of the
workpiece can be calculated based on
their compositions using commercial
software and databases. The composition of the weld depends not only on
the compositions of the workpiece and
the filler metal but also on the dilution, as shown in Fig. 3 previously.
Therefore, when a proper filler metal
is selected, a proper dilution level may
still be needed to obtain the desired
weld composition.
The T-fS curves of the weld and the
workpiece (which has the same composition as the partially melted zone) can
be calculated based on their compositions. The simple Scheil solidification
model is valid because cracking occurs
shortly after solidification starts (no
time for solid diffusion), e.g., 5 to 7 s as
measured in permanent-mold casting of
Mg alloys (Refs. 43–45) and even shorter in arc welding. According to Flemings
(Ref. 41), solid-state diffusion can be
neglected when 4kDStf/(d2) < 0.1, where
k is the equilibrium segregation ratio, DS
the coefficient of diffusion in solid, tf
the solidification time, and d the secondary dendrite arm spacing (DAS).
Based on the data of DAS of Al alloys vs.
the cooling rate (Ref. 41), with a secondary DAS of 1  10–5 m, the cooling rate
is about 100°C/s. With a typical freezing
temperature range of about 125°C for Al
alloys (such as the average of 2014,
2024, 2219, 6061, 7075 Al), the local
freezing time tf is about 1.25 s. DS is typically 1  10–12 m2/s and k around 0.15
(0.125 for Al-Si and 0.170 for Al-Cu).
Thus, 4kDStf/(d2) is around 0.0075 < 0.1,
which suggests negligible solid-state diffusion. Thus, the cooling rate is fast
enough to make diffusion negligible and
the fS calculation is not affected by diffusion, which is significant only at relatively slow cooling rates. The T-fS curves
were calculated using the thermodynamic software Pandat (Ref. 38) and the
Mg database PanMagnesium (Ref. 39)
of CompuTherm, LLC, in Madison, Wis.
The compositions of the welds calculated according to Fig. 3 are shown in Table
5, along with the dilution of the welding
wire by the workpiece measured based
Chai Supplement February 2016_Layout 1 1/14/16 1:48 PM Page 65
WELDING RESEARCH
A
B
Fig. 10 — AZ91 Mg welded with AZ92 Mg welding wire: A — T­fS curves predicting no
crack susceptibility; B — verification of predicted no crack susceptibility. T­fS curves calcu­
lated using Pandat (Ref. 38) and PanMagnesium (Ref. 39) of CompuTherm, LLC.
on the weld transverse cross section.
Since weld (S+L) and workpiece
(S+L) are in intimate contact with each
other along the weld edge, at any given
point along the weld edge the thermal
cycles (temperature T vs. time t) they
experience are identical. Thus, when a
different heat input or travel speed is
used, the thermal cycles both change
but are still identical to each other. In
fact, the thermal cycles of the two
semisolids are still identical even if the
cooling rate becomes slow enough to
cause significant solid-state diffusion,
which can be considered by using a solidification model that includes solidstate diffusion. Consequently, comparing the T-fS curves of the two semisolids is equivalent to comparing their
t-fS curves.
Figure 9 shows the weld made with
AZ91 Mg as the workpiece and the
AZ31 Mg as welding wire. The T-fS
curves in Fig. 9A show that the weld fS
is significantly greater than the workpiece fS up to about fS = 0.92 as indicated by the region encircled by the dotted oval. For instance, when the AZ91
Mg workpiece fS reaches 0.70 (at about
525°C), the weld fS is about 0.86, that
is, higher by about 0.16. Thus, the
curves suggest a significant susceptibility to cracking. Beyond 0.92, the
weld fS becomes slightly lower than the
workpiece fS, but this is too late to
keep cracking from occurring. The
crack susceptibility predicted by the
T-fS curves in Fig. 9A is verified by the
cracks visible in the transverse crosssection of the weld in Fig. 9B.
Figure 10 shows the weld made
with AZ91 Mg as the workpiece and
AZ92 Mg as the welding wire. The T-fS
curves in Fig. 10A show the weld fS is
less than the workpiece fS at any time
during solidification. Thus, the curves
suggest no significant susceptibility to
cracking. No crack can be observed on
the transverse cross section, as shown
in Fig. 10B. Thus, the crack susceptibility can be eliminated by selecting a
proper filler metal and dilution level,
and the simple criterion can be used as
a guide for the selection.
The weld made with AZ31 Mg as the
workpiece and AZ92 Mg as the welding
wire is shown in Fig. 11. The T-fS curves
in Fig. 11A show the weld fS is significantly less than the workpiece fS up to
about fS = 0.96, beyond which they become almost identical to each other.
The absence of crack susceptibility predicted by the T-fS curves is verified by
the absence of cracking in the transverse cross section in Fig. 11B.
Figure 12 shows the weld made
with AZ31 Mg as the workpiece and
AZ61 Mg as the welding wire. The T-fS
curves in Fig. 12A show that weld fS is
significantly less than the workpiece
fS, up to about fS = 0.96. So, no cracking is expected up to this point. Beyond fS = 0.96, the weld fS becomes
slightly higher than the workpiece fS,
by less than about 0.02. Thus, the
curves suggest no significant susceptibility to cracking. No cracks can be observed on the transverse cross section,
as shown in Fig. 12B.
Conclusions
1. The crack susceptibility of Mg
welds can be predicted by comparing
the T-fS curves of the weld and the
workpiece, that is, susceptible if weld
fS > workpiece fS. The susceptibility is
significant if the difference is significant, e.g., 0.05 and above.
2. This simple criterion can be used
as a useful guide to selecting a proper
Mg filler metal and dilution level to
avoid liquation cracking.
3. As-cast AZ91Mg can be highly
susceptible to liquation during welding
due to the presence of numerous
coarse  (Mg17Al12) particles in the Mgrich matrix , causing liquation by  +
  LE at a eutectic temperature as low
as 437°C. The large amount of liquid
formed separates grains and causes
cracking under the tension induced by
the solidifying and contracting weld
metal nearby. The use of a proper filler
metal and dilution is essential for
avoiding cracking in AZ91 Mg.
4. Wrought AZ31 Mg alloy is much
less susceptible to liquation due to its
solution heat treatment during producFEBRUARY 2016 / WELDING JOURNAL 65-s
Chai Supplement February 2016_Layout 1 1/14/16 1:48 PM Page 66
WELDING RESEARCH
A
B
Fig. 11 — AZ31 Mg welded with AZ92 Mg welding wire: A — T­fS curves predicting no
crack susceptibility; B — verification of predicted no crack susceptibility. T­fS curves cal­
culated using Pandat (Ref. 38) and PanMagnesium (Ref. 39) of CompuTherm, LLC.
tion, which dissolves  (Mg17Al12) and
raises the liquation temperature from
the eutectic temperature to the much
higher solidus temperature of 600°C.
This work was supported by the National Science Foundation under Grant
No. IIP-1034695, the American Welding
Society Foundation (the AWS Foundation Fellowship Program for Xiao Chai),
and the University of Wisconsin Foundation through the Industry/University
Collaborative Research Center (I/UCRC)
for Integrated Materials Joining Science
for Energy Applications.
Acknowledgments
The authors would like to thank: 1)
CompuTherm, LLC, in Madison, Wis.,
and Professor Rainer Schmid-Fetzer,
Clausthal University of Technology,
Clausthal-Zellerfeld, Germany, for the
software package Pandat and the database PanMagnesium; 2) Tom Kurilich
of US Magnesium, LLC, Salt Lake City,
Utah, for donating 45 kg (100 lb) of
pure Mg ingots; 3) Derek Landwehr
for machining numerous specimens
for the circular-patch test.
References
1. Watarai, H. 2006. Trend of research
and development for magnesium alloys —
Reducing the weight of structural materials in motor vehicles. Quarterly Review 18:
84–97.
2. “Magnesium Vision 2020: A North
American automotive strategic vision for
magnesium.” 2006. USAMP, United States
Automotive Materials Partnership — A
consortium of the United States Council
for Automotive Research, MG 2020, released 11, 1, pp. 1–34.
3. Kulekci, M. K. 2008. Magnesium and
66-s WELDING JOURNAL / FEBRUARY 2016, VOL. 95
its alloys applications in automotive industry. International Journal of Advanced Manufacturing Technology 39: 851–865.
4. Feng, J., Wang, Y., and Zhang, Z.
2005. Status and expectation of research
on welding of magnesium alloys. The Chinese Journal of Nonferrous Metals 15(2):
165–178 (in Chinese).
5. Deinzer, G. H., and Rethmeier, M.
2006. Magnesium Technology — metallurgy,
design data and applications, by H. E.
Friedrich and B. L. Mordike, SpringerVerlag, pp. 349–363.
6. Cao, X., Jahazi, M., Immarigeon, J. P.,
and Wallace W. 2006. A review of laser
welding techniques for magnesium alloys.
Journal of Materials Processing Technology
171: 188–204.
7. Liu, L. 2012. Welding and Joining of
Magnesium Alloys. Cambridge, UK: Woodhead Publishing.
8. Wagner, D. C., Yang, Y. K., and Kou,
S. 2013. Spatter and porosity in gas metal
arc welding of magnesium alloys: Mechanisms and elimination. Welding Journal
92(12): 347-s to 362-s.
9. CSC-MIG Weld Process Controller,
Jetline Engineering, Irvine, Calif.
http://pdf.directindustry.com/pdf/jet-line
-engineering/csc-mig-weld-processcontroller/27577-54392.html.
10. Chai, X., Yang, Y. K., Carlson, B. E.,
and Kou, S. 2015. Gas metal arc welding of
magnesium alloys: Oxide films, high
crowns, and fingers. Welding Journal 94(1):
16-s to 33-s.
11. Huang, C., and Kou, S. 2000. Partially
melted zone in aluminum welds — Liquation mechanism and directional solidification. Welding Journal 79(5): 113-s to 120-s.
12. Huang, C., and Kou, S. 2001. Partially melted zone in aluminum welds —
Planar and celluar solidification. Welding
Journal 80(2): 46-s to 53-s.
13. Huang, C., and Kou, S. 2001. Partially melted zone in aluminum welds —
Solute segregation and mechanical behavior. Welding Journal 80(1): 9-s to 17-s.
14. Kou, S., Firouzdor, V., and Haygood,
I. 2011. Hot Cracking in Welds of Aluminum
and Magnesium Alloys, Hot Cracking Phenomena in Welds III, edited by J. C. Lippold,
Th. Böllinghaus, and C. E. Cross, SpringerVerlag, pp. 3–23.
15. Kou, S. 2003. Welding Metallurgy,
2nd edition. Hoboken, N.J.: John Wiley &
Sons. pp. 122–141, 97–121, and 301–340.
16. Kou, S. 2012. Fluid flow and solidification in welding: Three decades of fundamental research at the University of
Wisconsin. Welding Journal 91(11): 287-s
to 302-s.
17. Zhou, W., Long, T. Z., and Mark, C.
K. 2007. Hot cracking in tungsten inert gas
welding of magnesium alloy AZ91D. Materials Science and Technology 23(11):
1294–1299.
Chai Supplement February 2016_Layout 1 1/14/16 1:49 PM Page 67
WELDING RESEARCH
A
B
Fig. 12 — AZ31 Mg welded with AZ61 Mg welding wire: A — T­fS curves predicting no
crack susceptibility; B — verification of predicted no crack susceptibility. T­fS curves calcu­
lated using Pandat (Ref. 38) and Pan Mgnesium (Ref. 39) of CompuTherm, LLC.
18. Sun, D. X., Sun, D. Q., Gui, X. Y., and
Xuan, Z. Z. 2009. Hot cracking of metal inert gas arc welded magnesium Alloy
AZ91D. ISIJ International 49: 270–274.
19. Lang, B., Sun, D. Q., Xuan, Z. Z., and
Qin, X. F. 2008. Hot cracking of resistance
spot welded magnesium alloy. ISIJ International 48: 77–82.
20. Sun, D. Q., Lang, B., Sun, D. X., and
Li, J. B. 2007. Microstructures and mechanical properties of resistance spot welded magnesium alloy joints. Materials Science and Engineering A 460–461: 494–498.
21. Yang, Y. K., Dong, H., Cao, H.,
Chang, Y. A., and Kou, S. 2008. Liquation
of Mg alloys in friction-stir spot welding.
Welding Journal 87(7): 167-s to 177-s.
22. Wagner, D. C., Chai, X., Tang, X.,
and Kou, S. 2015. Liquation cracking in arc
and friction-stir welding of Mg-Zn alloys.
Metallurgical and Materials Transactions A
46: 315–327.
23. Yuan, T., Chai, X., Luo, Z., and Kou,
S. 2015. Predicting susceptibility of magnesium alloys to weld-edge cracking. Acta
Materialia 90: 242–251.
24. Prasad, K. E., Li, B., Dixit, N., Shaffer, M., Mathaudhu, S. N., and Ramesh, K.
T. 2014. The dynamic flow and failure behavior of magnesium and magnesium alloys. JOM 66: 291–304.
25. Magnesium Alloys, Products Literature, Washington Alloys, Rancho Cucamonga, Calif., weldingwire.com/Images/
Interior/documentlibrary/magnesium%20
alloys.pdf.
26. Electron AZ31B Sheet, Plate & Coil,
Data Sheet 482, Magnesium Electron,
Manchester, England.
27. Houldcroft, R. T. 1954. Dilution and
uniformity in aluminum alloy weld beads.
British Welding Journal 1: 468–472.
28. Kou, S. 1996. Transport Phenomena
and Materials Processing. Hoboken, N.J.:
John Wiley and Sons, pp. 499–514.
29. Kou, S., and Sun, D. K. 1985. Fluid
flow and weld penetration in stationary arc
welds. Metallurgical Transactions A 16A:
203–213.
30. Kou, S., and Wang, Y. H. 1986.
Computer simulation of convection in
moving arc weld pools. Metallurgical Transactions A 17A: 2271–2277.
31. Tsai, M. C., and Kou, S. 1989.
Marangoni convection in weld pools with a
free surface. International Journal for Numerical Methods in Fluids 9: 1503–1516.
32. Huang, C., and Kou, S. 2004. Liquation cracking in full-penetration Al-Cu
welds. Welding Journal 83(2): 50-s to 58-s.
33. Huang, C., and Kou, S. 2004. Liquation cracking in full-penetration Al-Mg-Si
welds. Welding Journal 83(4): 111-s to 122-s.
34. Cao, G., and Kou, S. 2006. Predicting
and reducing liquation-cracking susceptibility based on temperature vs. fraction solid.
Welding Journal 85(1): 9-s to 18-s.
35. Binary Alloy Phase Diagrams,
American Society for Metals. 1986. Metals
Park, Ohio, p. 1565.
36. Pepe, J. J., and Savage, W. F. 1967.
Effects of constitutional liquation in 18-Ni
maraging steel weldments. Welding Journal
46(9): 411-s to 422-s.
37. Pepe, J. J., and Savage, W. F. 1970.
Weld heat-affected zone of the 18Ni
maraging steels. Welding Journal 49(12):
545-s to 553-s.
38. Pandat. 2013. Phase Diagram Calculation Software Package for Multicomponent Systems, CompuTherm, LLC, Madison,
Wis.
39. PanMagnesium. 2013. Thermodynamic Database for Magnesium Alloys,
CompuTherm, LLC, Madison, Wis.
40. Kou, S., and Le, Y. 1982. The effect
of quenching on the solidification structure and transformation behavior of stainless steel welds. Metallurgical Transactions
A 13A: 1141.
41. Flemings, M. C. 1974. Solidification
Processing. New York, N.Y.: McGraw-Hill,
p. 14.
42. Dahle, K., and Arnberg, L. 1997.
Development of strength in solidifying aluminium alloys. Acta Materialia 45:
547–559.
43. Cao, G., and Kou, S. 2007. Real-time
monitoring of hot tearing in AZ91E magnesium casting. Trans Amer Foun Soc. 115:
7–34.
44. Cao, G., and Kou, S. 2006. Hot tearing in ternary Mg-Al-Ca alloy castings.
Metallurgical and Materials Transactions A
37A: 3647–3663.
45. Cao, G., Haygood, I., and Kou, S.
2010. Onset of hot tearing in ternary MgAl-Sr alloy castings. Metallurgical and Materials Transactions A 41A: 2139–2150.
FEBRUARY 2016 / WELDING JOURNAL 67-s