Download as pdf or txt
Download as pdf or txt
You are on page 1of 79

INVERSE SEGREGATION

IN
ALUMINUM--GOPPER ALLOIS

William V. Youdelis

A Thesis,
Presented to the Faculty of Graduate
Studies and Research in partial fulfil-
ment of the requirements for the Degree
of Master of Engineering in Metallurgical
Engineering

August, 1956
ACKNOWLEDGEMENT

The author is grateful to Prof. J. s. Kirkaldy,


director of this research, for his original contributions
to the theory presented in this investigation and to
Prof. J. u. MacEwan, Chairman of the Department of Metal-
lurgical Engineering,for his assistance in the experiment-
al work. The financial assistance received fr•m Aluminium
Laberatories Ltd. in the form of the Aluminium Laborator-
ies Fellowship was also deeply appreciated.
CONTENTS

Acknewledgement
INTRODUCTION. • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 1
HISTORICAL
A. Theories Involving the Dependence of Inverse
Segregation •n Conditions Set Up in the Liquid
Alley Prior to the Beginning of Solidification •••••• S
(1) SmitHs Theory•••••••••••••••••••••••••••••••••• 5
(2) Benedicks' Tbeory •••••••••••••••••••••••••••••• 6
(3) Underc$oliDg TheGrY•••••••••••••••••••••••••••• 7
B. Theories Involving the Dependence $f Iaverse
Segregatioa on a Transfer of Liquid Metal
During the Course 0f Solidificatioa ••••• ~ ••••••••• ~~ 9
(1) Contraction Pressure Theery ....................... 9
(2) Crystallization Pressure Theory ••••••••••••••••
.
lO .

(3) Gas Evelution Theory ••••••••••••••••••••••••••• ll


(4) Watson's Theory •••••••••••••••••••••••••••••••• l2
(5) Imterdendritic Flew TheQry ••••••••••••••••••••• l3
C. Qualitative Th eGry of the Segregatieo Mechaaiam
in Directionally Solidified Ingets •••••••••••••••••• l4
D. Cemparisien of Theery With Experiment ••••••••••••
.
~•.l7 .

(1) Segregatien in Al-Cu Alloys-Adams •••••••••••••• l7


(2) Inverse Segregation in Centinuous Castiags-
H. Kastner•••••••••••••••••••••••••••••••••••••l9
THEORI
Mathematical F~rmulation ef Inverse Segregation
Based on the Volume Contraction Theory •••••••••••••• 22
(1) Model of the Solidification Process •••••••••••• 22
(2) The Differenti•l Equation •••••••••• •••••••••••• 24
(3) Calculati~n ef Maximum Segregatien ia
Al-Cu System••••••••••••••••••••••••••••••••••• 31
(4) Positiênal Variation of Solute Throughout
the Ingot••••••••••••••••••••••••••••••••••••·· ~ 35
EXPERIMENTAL
A. ~terials ••••••••••••••••••••••••••••••••••••••••••• 37
B. Apparatus ••••••••••••••••••••••••••••• ~ ••••••••••••• 37
c. Experimental Precedùr••••••••••••••••·····~··••••••• 39
(1) Meltdown ••••••••••••••••••••••••••••••••••••••• 39
(2) Stirring ••• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 40
(J) Degassing •• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 40
(4) Solidificatien ••••••••••••••••••••••••••••••••• 41
(5) Sampling •• •••.•••••••••••••••• • •• • •• • • • • • • • • • • • 42
(6) Aaalysis ••••••••••••••••••••••••••••••••••••••• 43

o.
(7) Estimated Accuracy •••••••••• ~················~~
Resulta and Discussion •••••••••••••••••••••••••••••• 45 "
(1) Maximum Segregation•••••••••••••••••••••••••••• 54
(2) Positienal Variatien •••••••••••••••••••.••••••• 55
(J) Failure of Theory at Ingot Top ••••••••••••••••• 55
CONCLUSION.. • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 57
SUJvll.iARI • •••••••••••••••••••••• e ••••••••••••••• ·• • • • • • • • • • • • • • 59
LIST OF REFERENCES•••••••••••••••••••••••••••••••••••••••••• 60
APPENDIX !•••••••••••••••••••••••••••••••••••••••••••••••••• 63
APPENDIX II. . • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 65
APPENDIX III.. • • • • • . . • • • • . • • • • • . . . • • • • • • • • • • • • . • • • • • • • • • • • • • 71
ILLUSTRATIONS

Figure 1
Representative Equilibrium Diagr~··••••••••••••••••• 16
Table I
Calculated and Experimental Values for Maximum Segreg-
ation in Al-Cu Ingots •••••••••••••••••••••••••••••••• 46
Figure 2
Representation et Crystal Growth in Unidirectionally
Solidified Ingots •••••••••••••••••••••••••••••••••••• 16
Table II
Copper Content at Various Positions in Ingot 2A •••••• ~g

Figure 3
Schematic Diagram of Solidifying Ingot and Concen-
tration Distribution of Interdendritic Liquid •••••••• 26
Table III
Copper Content at Various Positions in Inget 3A •••••• 48
Figure 4
Ck and Cs' vk and v8 , T Diagrams fGr Al-Cu ••••••••••• 33
Figure 5
Variation of a with Ca••••••••••••••••••••••••••••••• 34
Figure 6
Diagram ef Solidification Apparatus •••••••••••••••••• 38
Figure 7
Segregation Curve fer Al-Cu Alloys and Experimental
Vaiues ••••••••••••••••••••••••••••••••••••••••••••••• 47
Figure à
Composition Curves for Ingots 2A and 3B •••••••••••••• 49
Figure 9
Macrostructure of Ingots 2A and 3A Showing Long
Columnar Crystal Gr.wth •••••••••••••••••••••••••••••• 50
Figure 10
Chill Face Microstructure of Ingots 3B and 4A •••••••• 52
Figure 11
Chill Face Microstructure of Ingots 2A and 3A •••••••• 53
Figure 12
Diffusion System with Stationary Boundry and Fixed
Concentration C0 for all Time •••••••••••••••••••••••• 63
Figure 13
Dendrite MGrphology at the Chill Face •••••••••••••••• 65
Figure 14
Dendrite Morphology at Ingot Top ••••••••••••••••••••• 67
I NT R 0 DUCT I 0 N

Normal and Inverse Segregation

Segregation which occurs in slowly cooled castings and


in the direction indicated by the equilibrium diagram is called
"normal segregation". The first crystals to separate from the
molten alloy during solidification are rich in the higher
melting point alloy and occupy a position close to the mold face.
As the solidification proceeds the growing dendrites push before
them the residual liquid which is becoming continually enriched
in the lower melting point alloy (usually the alloy of high
solute concentration). This phenome~is due to the fact that
in nearly all alloys solidifying under slow but nevertheless non-
equilibrium conditions, solid state diffusion is slow enough
that the central core of each dendrite maintains a lower solute
concentration than is found throughout the remainder of the
ingot.
Segregation in the opposite direction is frequently
found in alloys that have been rapidly cooled from the molten
state. The lower melting point constituent is concentrated
towards the outside of the ingot along with the primary crystals
of the higher melting point alloy and its concentration
decreases as the centre of the ingot is approached. This type of

- 1 -
2

11
segregation was in earlier times referred to as liquation",
11
but is better known today as inverse segregation".
A phenomenon which may generally be seen to
accompany inverse segregation is the appearance of a so-
called "sweat" or enriched residual liquid on the surrace or
the cast body. This sweating or residual liquid is orten
rererred to by some workers as inverse segregation. Although
the sweating phenomena usually contributes to the magnitude or
inverse segregation, it is now known that the two mechanisms
are entirely dirrerent and can occur independently.
Experimental data has shown that the necessary condi-
tions for inverse segregation are a long solidification range
for the alloy and a fairly rapid cooling rate (as compared to
equilibrium cooling).

Effects of Segregation on Alloy Properties

The effect of chemical inhomogeneity on the physical


properties of metal alloys is well known. Chemical inhomoge -
nei ty· in structural sections ro lled from ingots is al ways a
result of segregation, normal or inverse , a nd is known to have
such deleterious effects as reducing the strength, toughness,
resistanc e to creep deformation and fatigu e r esistanc e of the
final section. Many failures of structural sections, particu-
larly those subjected to cyclic stresses over long periods or
time are a ttributed to chemical inhomogeneiti es in the metal.
The resistance of certain metals to corrosion is orten
decreased by the addition of alloying constituents, the latter
3

being required to increase the strength properties of the


base metal. This is especially true of the metal aluminum.
The amount of alloying constituent added is often a compromise
between the strength and corrosion requirements. It is there-
fore evident that segregation of an alloying component to the
outer regions of an ingot will result in poorer corrosion
resistance at the surface and subsequent forming operations
will not appreciably alter the distribution of the segregated
components.
Inverse segregation has become of great practical
concern with the development of continuous casting techniques.
The high degree of inverse segregation and sweating of
continuous cast ingots are attributed to the rapid cooling and
associated contraction stresses which are characteristic of the
technique.
The usual industrial practice to avoid the effects of
inverse segregation and sweating is to "scalp" the ingots. For
soft metals the enriched surface layer of the ingots can be
removed by passing the ingot through a metal planer. This
procedure becomes increasingly more difficult and expensive
for the hard alloys which, on account of their high alloy content,
show the most intense segregation.
While the empirical control of inverse segregation and
sweating would be satisfactory from the commercial point of
view, the best long term approach is to elucidate the fundamental
mechanisms of these phenomena. This can best be achieved by
studying the simple process of uni-dimensional solidification in
a binary alloy eliminating as many complicating variables as
possible. This investigation of inverse segregation in the
4

binary Al-Cu alloy system is directed along these lines and


the experimental results are compared with a mathematical
theory of inverse segregation which is developed in part for
the first time in this work.
- 4-A-

HISTORICAL

The phenomenon of inverse segregation was observed


in copper-silver alloys as long ago as 15SO, and references
were frequently made to it in succeeding centuries. The first
extensive investigation into its cause may be said to date
from 1S75. when Roberts-Austen(l)* began researches which ex-
tended Qver many years. The early work was performed chiefly
on precious metals, fer it was in attempts to obtain uniferm
"trial plates" against which to compare the fineness of gpld
and silver coinage that this phenomenon became of considerable
practical concern.

THEORIES OF INVERSE SEGREGATION

The theories which have been advanced to account for


the occurrence of inverse segregation may be divided inte two
classes:
A. The assumption is made that since the outside of
aa ing•t is richer in the low melting point alloy thaR the
centre, the crystals which separated first were al se_, for some
reason, abnormally enriched in that alloy. This is accounted by

* The numbers in parentheses refer to the references listed


on p. 60.
5

some variation in properties of the liquid alloy, leading to


such an adjustment of composition.
B. It is maintained that solidification begins quite
normally in a manner predictable from the equilibrium diagram,
but that at some la.ter stage prior to complete solidification,
the residual liquid is forced from the centre to the outer
regions of the casting. This theory in principal is most
widely accepted today. Several forces have been suggested to
account for this displacement action with the more favoured one
today being the capillary effect resulting from channel formation
by the contraction of dendrites on solidification.

A. THEORIES INVOLVING THE DEPENDENCE OF


INVERSE SEGREGATION ON CONDITIONS SET
UP IN THE LIQUID ALLOY PRIOR TO THE
BEGINNING OF SOLIDIFICATION.

(1) SMITH'S THEORY (1917)

From an extensive study of surface tension and cohesion


in metals and alloys, Smith(2) put forth the hypothesis that
the component which lowers the freezing point of any particular
alloy is the component which tends to segregate towards the
chilling surface.
Smith observed that binary alloys prone to inverse
segregation were composed of metals , one of which has a high
cohesive force (surface tension). He used this fact in
conjunction with the "Le Chatelier Principle" to explain the
mechanism of inverse segregation. The Le Chatelier Principle
6

states that when a state of a system is changed , the system


will alter so as to present a greater resistance to that
change. In a copper-silver alloy, for example, the alloy
resista solidification by having copper atoms migrate to the
mold wall and so forma liquid with a lower metting point. The
higher affinity of copper atoms for copper atoms acta as a
driving force for differentiai migration with the result that
the copper atoms are able to disentangle themselves from the
silver atoms and migrate to the mold wall, thus displaying an
equal amount of silver atoms and leading to inverse segregation.
This theory is not acceptable for the following reason.
Inverse segregation can be brought about in normal segregating
alloys by increasing the cooling rate. If Smith's theory was
correct inverse segregation would appear preferentially under
slow cooling conditions since more time would be allowed for
atoms with a high mutual affinity to segregate by diffusion.

(2) BENEDICKS' THEORY (1925)

It is well known that whenever a temperature gradient


exista in a solution a concentration gradient is also set up,
the concentration of a solute (or solvent) being greater in the
cold region than in the hot. This phenomenon known as the
Ludwig-Soret effect is operative in aqueous and non aqueous
solutions. It was thought that this could likewise occur in
molten alloys, though its existence had never been directly
observed at the time. On the basie of the Ludwig-Soret effect
7

Benedicks(3) accounted for the segregation of carbon and


phosphorous observed in steel and cast iron at that time.
Benedicks' theory has been criticized on the following
grounds:
i) The establishment of the concentration
gradient is known in any aqueous
solutions to be exceedingly slow.
ii) Experimenta by Ballay(4) on several
different alloys did not show the
existence of any appreciable
concentration gradient when subjected
to a temperature gradient.
iii) To obtain the observed difference in
concentration, it is necessary to assume
a temperature gradient far in excess of
anything possible in practice (Desch(5), discussion).

(3) THE UNDERCOOLING THEORY ( 1922)

It was suggested by Hanson(2)(discussion) that under-


cooling may be a cause of inverse segregation. The idea was
more extensively examined by Masing(6), who concluded that this
mechanism could well play a part in the phenomenon. Masing
notes that crystallization from a super-saturated salt solution
depends on two factors: the latent heat of crystallization and
the diffusion between the crystals and the surrounding liquid.
In Figure 1, when a molten alloy containing C0 percent solute
8

is super-cooled to a temperature t below the solidus and then


begins to crystallize the above two factors determine the
course of solidification. If the latent heat of crystalliza-
tion is the predominating factor, a large amount of heat is
evolved, the temperature will soon increase to ~and crystals
having a composition indicated by the point B will separate.
The result is then the same as if the alloy had slowly cooled.
On the other hand, when the latent heat is small and diffusion
is the determining factor, the temperature of the alloy remains
constant at t and it is possible for crystals having composi-
tion C to separate, which will be richer in component B than
the mother liquid. Masing's reasoning is strongly criticized
by Genders(7) who shows that if a solid of composition C is
crystallized then a liquid of composition E must be formed.
This is impossible since beth are more concentrated than the
original liquid. The alternative supposition of Masing's theory
that a liquid of composition F is formed is also found untenable
by Genders since it implies the production of a liquid represent-
ing a still greater degree of super-cooling.
This theory is furthermore untenable for the following
reasons:
i) In spite of the most rapid chilling rates,
coring in single dendrites could be
detected; i.e., normal segregation
proceeds locally.
ii) It has been established that inverse
segregation takes place in seme slowly

cooled ingots.
9

iii) Recent work on nucleation has indicated


that a large volume or liquid metal
cannet be supercooled more than a rew
degrees.

B. THEORIES INVOLVING THE DEPENDENCE OF


INVERSE SEGREGATION ON A TRANSFER OF
LIQUID METAL DURING THE COURSE OF
SOLIDIFICATION.

(l) CONTRACTION PRESSURE THEORY (1922)

Kuhnel(8) presented the rollowing argument to explain


the mechanism or inverse segregation. When solidirication
begins a crystalline envelope rorms surrounding the molten
metal. Cooling causes the envelope to shrink, thereby
subjeéting the liquid centre to a pressure and leading to a
withdrawal or the crystals rrom the mold wall ror a short
distance. When the hydrostatic pressure or the melt becomes
greater than the tensile strength or the envelope, rupture
takes place by intercrystalline cracking. The cracks thus rormed
are rilled with outrlowing metal which almost instantaneously
welds the envelope again into a tight container.
Observations made by Iokibe{9), on the contraction
taking place in various parts of a cast ingot cannet be
reconciled at all with this theory. Furthermore, Masing{l0)
performed a calculation in which he showed that the temperature
gradient necessary in an ingot for contraction pressure to be
exerted is much greater than is found to exist in practice.
- 10

This the ory has been re jected mainly· on the grounds


that most liquids contract on solidification to a sufficiently
high degree to eliminate the possibility of contraction
pressure being brought upon the still molten alloy.

(2) CRYSTALLIZATION PRESSURE THEORY (1925)

Masing end Haase(ll) thought that the various theories


on inverse segregation existing at that time could not wholly
account for the concentration gradients observed in chilled
castings. They suggested that an answer could be found in
crystallization pressure, which in slowly cooled ingots
manifests itself as an expansion of the solidified outer zone
by· mu tu al pushing a part of the grovring dendrites. This expansion
was shown by Iokibe{l2) to depend strongly on directional
crystal growth.
Masing and Haase based their theory on Iokibe's observa-
tions. They· assumed that since i t is impossible in a chill
casting to have an actual expansion such as is found in slowly
cooled ingots owing to the rapid formation of a solid shell, the
growing dendrites exert pressure on the residuel liquid in the
centre of the ingot and expel it towards the outside.
This theory is found unacceptable for the same reason
that the previous one has been rejected. Due to the large liquid
contraction on solidification no pressure build-up in the
residual melt is possible as a result of the solidification
mechanism.
11

(3) GAS EVOLUTION THEORY (1927)

As a result of his work on bronzes containing 5 percent


tint Genders (7) was led to believe that gas evolution during
the later stages of solidification is the primary· cause of all
inverse segregation~ Every molten metal contains some
dissolved gas which as solidification proceeds, becomes
concentrated in the reaidual liquid until the latter becomes
supersaturated at the prevailing temperature , whereupon the gas
is liberated. Because of the high viscosity of the liquid near
solidified temperatures, the escape of this gas is impeded by
constrictions and gas pressure is set up which immediately
forces the liquid metal along interdendritic channels towards
the outside of the i~got. This gives rise at the same time to
the central porosity which is usually associated with inverse
segregation.
While it is agreed that gas pressure may be responsible
in part for inverse segregatio~ it cannot be accepted as the
prime reason for it has been shown by Ga ylor(l3) that inverse
segregation occurs in vacuum melted metal in which the amount
of gas present is very sma1l if not entirely absent. Adams(l4)
has shown that ingots solidified from ga ssy melts show even lesa
inverse s e gregation tha n dega ss ed ingots, pre s umably due to the
negative effect of porosity on complete flow back of the
residual liquid. Northcott(l5) furthermore ha s shown by bleeding
partia lly solidified ingot~ tha t inve rse segregation begins a t
a very early stage in solidification, before the liquid is likely
to have become supers a tu r ated wi th gas.
12

(4) WATSON'S THEORY (1932)

From observations based on chilling copper-silver


alloys which had been maintained for some hours in the
liquidus-solidus region, Watson(l6) enunciated his theory
of the "migration of primary dendrites" as the cause of
inverse segregation.
By maintaining the temperature of the alloy in the
solidus-liquidus region, large primary dendrites were grown
which migrated under the influence of gravity either to the
top of the crucible (if copper rich) or to the bottom (if
silver-·rich). Severe chilling action was then applied at the
ingot end where the dendrites had collected and a microscopie
examina.tion of the ingot showed that the large primary
dendrites had receded from their former position by a short
distance, and that a region of chilled metal existed between
the primary crystals and the ingot surface leaving the latter
impoverished relative to the constituent which had accumulated
there previously. Watson concluded that all inverse segrega-
tion phenanena could be explained on the assumption that the
primary crystals which were formed on the mold surface when an
ingot begins to solidify, immediately migrate towards the
centre of the ingot. The particular force which occasions
the migration was not identified, but it was suggested that
there were many possible ones. Such a radically unorthodox
conception of crystallization did not go unchallenged, and it
was criticized by Genders(l6) (discussion) and Allen(l7), both
13

of whom thought the experimental observations had been mis-


interpreted.

(5) INTERDENDRITIC FLOW THEORY (1921)

As a result of their work on tin-base alloys! Bauer


and Arndt(l8) were led to formulate the interdendritic flow
theory, which in a modified form is generally accepted today.
Bauer and Arndt supposed that the volume of the dendrites
projecting ahead of the solid interface is smaller than that
of the spaces between them at all times, and that the latter,
due to solidification contraction, are subject to an influx of
liquid which has been enriched in the low melting point
constituent by its rejection from the growing dendrites. The
reason given for assuming that the volume of the spaces exceeds
that of the dendrites is to allow for considerable influx of
the residual liquid and thereby to account for the big change
in composition that takes place in the outer regions of the ingot
during the course of solidification. A critical examination of
the mechanism indicates that this assumption about the distribu-
tion of liquid and solid is unnecessary. The pronounced normal
segregation observed in bismuth rich alloys (15) which expand
on solidification is strong evidence in favor of this theory.
The present concept of the interdendritic flow theory will be
discussed in detail in the following sections.
14

C. QUALITATIVE THEORY OF THE


SEGREGATION MECHANISM IN
DIRECTIONALLY SOLIDIFIED INGOTS

The present conception of the segregation mechanism


has been elucidated by Adams(14) as follows: Assume columnar
dendrite solidification (Figure 2) of an alloy of composition
C0 whose constitution is given in Figure 1. Furthermore,
assume that solidification is rapid enough that no diffusion
occurs ahead of the solidification front (tips of the dendrites)
but that the diffusion rate between the dendrites is
sufficiently high to eliminate liquid coring in the mushy zone.
This assumption is justified for rapid and moderately rapid
cooling rates (see ftppendix I) in circumstances where residual
liquid is contained in intricate intercrystalline and inter-
dendritic channels. The feeding of the liquid between the
growing crystals to compensate for solidification is assumed
to be perfect.
As the crystals grow upwards from (a) in Figure 2 they
thicken to (b) and at (c) the chill face from which the growth
began has reached the solidus temperature. Meanwhile, residual
liquid feeds downward, to compensate for freezing shrinkage, and
as it traverses the solid plus liquid region its temperature is
lowered and it becomes enriched in the segregating element. Thus,
in {b), the liquid which reaches the chill face has already
become enriched, and the concentration of the segregating element
at the chill face increases progressively until the solidus
temperature is reached (c). Since diffusion across the liquidus
15

level is assumed to be zero, the liquid reeding downwards


into the solidirying zone at t~ level, to replace the
enriched liquid moving toward the solidus level : will have
the same composition as the original liquid. Hence the upper
part or the solidirying layer will be impoverished in the
segregating element, as shown by the composition curve in
Figure 2. As solidirication proceeds rrom (c) to (d), the
impoverished region receives in its turn an enriched reeding
liquid and a nother impoverished region rorms above it. This
process repeats itselr until the impoverished region occurs at
the top or the ingot. Complete reeding or the last zone is
impossible, and the composition will rall sharply to that or
the rirst deposited solid - the tips or the projecting
dendrites.
On this theory the ma.in reatures pr segregation will be
an increase in the amount or the segregating element in the
region or the chill race, the width or the enriched zone being
determined by the distance rrom the chill race to the !~uidus

isothermal at the moment when the chill race reaches the solidus
temperature. Together with this will be a rapid concentration
decrease near the top of the ingot. In any practical case
imperrect reeding or reeding with liquid not' enriched to its
maximum will cause a rounding of the singularities at the top and
bottom or the ingot. If the temperature gradient in the liquid
should become more graduai as solidification proceeds, the mushy
zone will continue to grow with the result that there will be
positive segregation through a considerable length of the ingot.
- 16 -

Co
1

LJ9VII>
C.s

C OMPOS/ T!ON %B
FIGURE/. REPRE5E/IITA T!VE EÇJVILIB!NUM .IJIAGRANI

(C) (d}

1'

(€) (f)

FJGUR~ 2 REPRES~NTATION OF CRYSTAL GROWTH IN


UNI.lJIF?ECTIONALLY SOLI:D/FIE.D JN60T
17

This is shown by the dotted composition curve in Figure 2(h).

D. COMPARISON OF THEORY WITH


EXPERIMENT

(l) Segregation in Al-Cu Alloys - D. E. Adams

Adams(l4) measured the segregation in a number of


ingots of Al-Cu alloys containing 3 and 7 per cent copper ,
solidified in such a way that turbulence was avoided and
solidification proceeded in one direction only. This was
accomplished by pouring the melt into a preheated carbon mold
resting on a steel plate as its base . The base of the mold was
chilled by means of a jet of water flowing at a controlled
rate. Some of the melts were degassed, some were solidified
in vacuo and others were chilled from the top to determine the
extent to which gravity is a force effecting the transport of
the residual liquid.
Adams observed three well defined zones in these ingots:
(i) On exuded layer (sweat) of eute ctic at the outer surfa ce
of the chilled face. (ii) A zone adjacent to the chilled face
slightly enriched in coppe r, the zone becoming narrower with
more rapid cooling. (iii) A wider zone at the remote end of
the ingot impoverished in copper.
Adams round that exudations were not eliminated by grain
refinement nor by solidif ying in a va cuum , nor we re they effected
by gas conte nt . He noted that if the contact between the chill
plate and ingot was broken during the early stages of solidifica•
18

tion, the chill face of the 1ngot became pssty as a result of


reheating from the body of the metal, and this allowed. the
passage of liquid out to the surface resulting in s sweat
sppearance. The rejection of the eotectic 11qu1d to the
outer surface was accompanied by a zone adjacent to the
exudation that was impoverished in ~tectic.

To determine the increase in copper concentration at


the chill face resulting from the inverse segregation mechanism
itself, Adams corrected for surface exudations in analysing
the chill face for samples. The magnitude of segregation at any
locality in the ingot is expressed as the difference in copper
concentration (weight per cent) at that locality from the
average copper concentration of the ingot. The experimental
value is compared to an estimated value calculated on the basie
of feeding the shrinkage spaces with residual liquid during
solidification. For the 7 per cent copper alloy contraction
was taken to be 6 per cent, with feeding regarded as taking
place in two stages - during the cooling to the eutectic
temperature and during the final solidification at the ~tectic

temperature. The contraction space formed during the first


stage is considered to be fed with melt of 12 to 14 per cent
copper. The last stage is fed with liquid of eutectic
composition. The calculated value for inverse segregation for
the 7 per cent copper alloy was 0.58 per cent as compared to
the experimental value of 0.50 per cent.
Ingots solidified in vacuo or chilled from the top
showed substanially the same amount of inverse segregation as
19

ingots solidiried under normal atmospheric conditions. Gas


evolution, creating gas pressure in the ingot, had a negative
erfect on inverse segregation. This was attributed to
imperfect feeding of interdendritic spaces as evidenced by
excessive porosity in ingots solidified from gassy melts. The
copper content in the ingots followed in general the composition
curve in (h), Figure 2.
Since the experimental values compare favourably with
calculated values of inverse segregation, Adams concludes that
the feeding of the solidification contraction voids by residual
liquid adequately explains the observed segregation values in
unidirectionally solidified ingots. The absence of any
significant variation of inverse segregation in ingots solidified
in vacuum is cited as evidence that the major force determining
the movement of residual liquid arises from the surface tension
of the liquid.

(2) Inverse Segregation in Continuous Casting - H. Kastner

Kastner(l9) set out to explain the mechanism of inverse


segregation and sweating in continuously cast ingots. In one
of his numerous experimenta Kastner cast a square bar of 6
per cent tin bronze, 60 x 60 m.m. in cross-section. The metal
was poured into a water-cooled copper mold, 900 m.m. long and
moved up and down to simulate actual casting practise. The
ingots were further chilled by water sprays as they emerged from
the mold. Ka.stner observed a sweat to form on the outside
20

surface of the bar. The corners of the bar were free of any
sweating. The tin content of the bar was highest in the out-
aide regions of the bar and decreased as the bar center was
approached. About one-third the distance from the bar surface
to its center the tin concentration remained constant for a
short distance into the bar center. It then continued to
decrease to a value below the average tin content of the melt
as the center of the ingot was approached.
The variation in tin content through the cross section
of the bar was explained as follows: The solidification
interval can be broken down into two stages. (1) The initial
chilling of the molten metal as it comas in contact with the
water cooled mold. During t~ stage the heat flow is through
the mold walls in a horizontal direction. (11) Further
solidification of the partially solidified ingot is by heat
withdrawal through its base as it comes in contact with the
cooling water.
The second stage of solidification resulta from
contraction of the ingot shell away from the mold wall. This
contraction creates an air space between the ingot surface and
cooling wall, thus restricting heat flow in this direction.
Heat flow will then be along the path of least resistance, the
longitudinal direction through the base of the ingot. Movement
of residual liquid is always in the direction of maximum heat
flow {opposite to direction of dendrite growth). During stage
(1) residual liquid moves to the outer regions of the ingot,
enriching these regions with the segregating component. During
21 -

stage (11) residual liquid movement is downward and


consequently the tin concentration will remain constant ror
a short interval across the ingot section. It will again
decrease as the center is approached when the section under
consideration comes into direct contact with the cooling
water and heat flow is again in the horizontal direction. In
both cases the movement of residual liquid is assumed to be
caused by the suction effect of the hollow spaces formed
between the primary crystals by solidification shrinkage.
The absence or sweat at the bar corners, which were
observed to be in contact with the mold surface at all times,
indicates that sweating is a result of the partially solidified
ingot drawing away from the mold surface as proposed by Adams.
- 22 -

THEORY

MATHEMATICAL FORMULATION OF INVERSE SEGREGATION


BASED ON THE VOLUME CONTRACTION THEORY

The experi~ental work of Adams and Kastnev leaves little


doubt that, qualitatively, the mechanism of inverse segrega-
tion in castings is as postulated by Adams (page 14). One of
the principal deterrents to the complete understanding of the
process has been the lack of an analytical description. Although
a number of workers including Masing (10,20,21) and Sauerwald(22)
have attempted quantitative predictions, Scheil(23) was the
first to derive an analytical expression which can be checked
by experiment. He developed an expression for the "maximum
p~le" segregation as a function of alloy composition and
reported good order of magnitude agreement with the experimenta.
In the present study the equations are developed in a more gen-
eral form so that they predict not only the maximunt segregation
but also the positional variation of the segregation in a uni-
dimensionally solidified ingot.

(1) Model of the Solidification Process

We will consider a mode! similar to that proposed by


Adams (see Figure 2). The model considered is not meant to rep-
resent the detailed dendrite structure but only the general
solid-liquid distribution in the mushy zone. Here the ingot is
chilled from the bottom so that the hypothetical, triangular
two-dimensional dendrites grow in a vertical direction only.
- 23 -

We choose this mode! because it is fairly realistic y~t still


allows the analytic solution of the equations. More detailed
dendritie modela can also be treated but will in general in-
volve numerical integration of the equations. The following
assumptions will be made:
1. No sweating occurs.
2. Solidification is rapid enough that no diffusion occurs
ahead of the dendrite tips and a concentration gradient
exista in the growth direction only in the solidus-
liquidus region.
3. Gas levels are low and flow-back due to solidification
contraction is unrestricted so that no shrinkage por-
osity occurs.
4. Very close to equilibrium conditions exist at the solid-
liquid interface (negligible supercooling).
5. The solid diffusion coefficient is relatively laa:Il
so that complete coring occurs in the solid.
6. The liquiçi diffusion coefficient is relatively large
se -lthà'ti ',equtllb:tidm exista between all the liquid and
adjacent solid ( no gradients in the liquid at right
angles to the growth direction).
Adams has suggested that the above assumptions are just-
ified for all moderately rapid cooling rates. It may be arg~ed

that assumptions 5 and 6 are not altogether valid when con-


sidering the finite time interval to be accounted for in assump-
tion 5 and the slow liquid diffusion rates in assumption 6.
A rough calculation has been p~rformed (see Appendix I } in-
- 24 -

dicating that assumptions 5 and 6 do in fact adequately des-


cribe conditions existing under moderately rapid cooling rates.
The dendrite thickness and spacing were determined microscop-
-- . -

ically and the time allowed for diffusion was obtained from the
known solidification rates. Liquid and solid diffusion coef-
ficients were obtained from published data. For the fast cool-
ing rates the average atom in the liquid will diffuse, in the
time allowed, a distance of 0.04 cm. whereas the distance av-
ailable for motion is only 0.003 cm. which is the interdend-
ritic spacing. Thus the liquid should not be cored appreciably
during the initial stages of solidification. The average atom
in the solid duffuses only 0.001 cm. in the time available
whereas the maximum distance available is again 0.003 cm. fhus
coring in the solid will be appreciable although not complete.
The conclusions are essentially the same for the slow-cooled,
9% copper ingot.
For the calculation we require a detailed knowledge
of the specifie volumes of the liquid and solid alloys.
Sauerwald(24) has collected such data for a number of systems
and his curves for the Al-Cu system are reproduced as Figure
4 for reference in the calculations of a later section.

(2) The Differentia! Equation

Figures 1 and 3 show respectively the constitution


diagram of a representative alloy and a detailed section of the
dendrite system in the solidifying ingot. The following symbols
- 25 -

are used {mainly after Scheil).


v--volume/unit length.
v8 -specific volume of liquid.
Yk-specific volume of solid.
ms -mass/unit
. length of liquid.
~-mass/unit length of solid.

m -ini~ial mass/unit length of liquid (included in volume


0
V) •
m3 E-mass/unit length of liquid when concentration CSE is
just attained.
Cs-equilibrium concentration in weight par cent of liquid.
Ck-equilibrium concentration in weight per cent of solid
( exists at interface only).
c3 E-concentration of liquid at eutectic temperature.
CkE-concentration of solid interface at eutectic temper-
ature.
crk-mean concentration of the cored solid {o(phase in the
Al-Cu system).
C--mean concentration in mass fraction.
C0 -intial liquid concentration in weight per cent.
40-seg~gation = e-co.
r--final length of the ingot.
L--locatiop of volume V relative to chill face.
L0 -maximum steady state length of dendrite.

The representative volume V of incrementa! width ( see


Figure 3a) is related to the constituent masses by
- 26 -

j~

(a.)

L~

(b)

FIGURe 3. 5CHEMATIC DIAGRAM OF SOLII>I/ZYIN6 INGOT


AND CONCENTRATION .1>/S TRIBUT/ON OF'
IN TERIJEN1JR.tTIC LI QUI])
- 27 -

v s:k
= vkdmk + vsms (1]

On the assumption of complete flow-back on solidification we


can write
(2]

where the vi.!·a are explicit functions of the concentrations.


Rearranging we obtain

dm
s
..
~ -admk (3]
where
a: ~ ... m8 dvs [4]
va: vsdmk

Next we consider the solute mass balance for the volume V. This
is

[5]

In this ralation we equate the negative change of sol-


ute mass in the liquid(a negative quantity) to the increase of
solute maas in the solid minus the solute in the liquid trans-
ported into volume V by contraction in V (two terms) plus the
solute transported out of V by contraction in all regions of
the interdendritic channel inside position L. In the latter
term dC 5 /dL is the concentration gradient in the direction of
dendrite growth at position L and dS is the inward flow past
point L due to contraction in inner regions.
Grouping terms in (5] and using[3] it becomes

-- [6)
- 28 -

We can proceed further without a specifie model of the liquid-


solid distribution in the mushy region but it is convenient to
have a picture such as Fig~r• 3a in miag. · Note however that for
the moment m8 need not be a linear function of L as sht~· It
should also be reiterated that the picture is really just a
graph of the solid-liquid distribution in the mushy zone and is
not meant to represent the detailed dendrite structure. Insofar
as the flow-back of residual liquid is complete, the mixture
of liquid and solid is qniform and the dendrite branches are of
uniform size across any section L, the analysis remains valid.
To evaluate the factor dS/dL we must set up the bal-
ance between the volume flowing inward past point L and the
volume contraction at inner points. This is

t dL\m"dv" (7]
JllS S

where the integrals are to be carried out along the solid-liq-


uid interface in Figure 3a from point L to the point of inter-
ception with the mold (as in the early stages of solidification)
or to the point of vanishing liquid. Since the maas distribut-
ion throughout the region inside L is a unique function of m
s
at any point L, we can introduce a weighting factor into each
integral relating the integrand at any point inside L, desig-
nated by the quantity m:, to a corresponding integrand at point
L when m6 reaches a value m':m".
s a .
We will designate this weight-
ing factor as W(m');
s accordingly,(7)becomes
.
- 29 -

msmin
S
+dL W(m')m'dv'
ms
li • s [8]

The limit m8 min' which in general is an explicit function of


m6 , is required when the schematic phase boundary in Figure 3a
corresponding to m5 intercepta the mold wall. T~is only occurs
in the initial stages of solidification. In all other cases the
limit is zero.
IT we combine DJ and [8] we obtain the required expression
for dS/dL,

dS (9J
en:
and the differentiai-integral equation becomes

[10]

Severa! interesting cases can be immediately deduced


from the general relation {10] • First, at the chill face where
we must obtain the maximum segregation, 111''::'"-1
smiJi - 8
=
m so that

-ade s
(11]

This is identical with the expression used by Scheil to obtain


the maximum segregation. For metals which show volume contract-
ion, a'l so that segregaÇdon will be po~itive.
- 30 -

Another simple case is obtained when steady state con-


ditions obtain during solidification. On the general grounds
that the concentration distribution in the mushy soae remains
constant in time so that there is no net accumulation of sol-
ute, the segregation is zero. This means that in general the
integrand in (10] is constant whereby

dms -- -dC s [12]


ms ~k

which describes a situation where there is effectively no vol-


ume contraction and therefore no segregation. This expression
is obtained explicitly from [10] in the event that a and vs are
constant over the whole range of m5 since under these conditions
and for steady state growth W:l.
It is clear that insofar as the multiplying factor in
parenthesis in[lO)is unity, less than unity or greater than
unity we get zero, positive or negative segregation. Renee
considering again a mode! with constant a and v 8 and in which
ms is a linear function of Las shown in Figure 3, we can qual-
itatively predict the segregation in other regions of the ingot.
Near the chill face, if W:l(dm 8 /dL is constant)• the segregation
will be positive since m~ i 70. As we proceed further in m8
om n min
is different from zero less and less pt tije time so that the
segregation diminishes until it vanishes if steady state growth
is obtained. If however the dendrites continue to elongate at-
ter their initial formation then W(l and positive segregation
persista across the ingot •. ~rom the point at whi~h the mushy
- 31 -

zone reaches the end of the ingot, W must become greater than
unity and the segregation change sign. W, and therefore the
negative segregation, increases continuously up to the end of
the ingot producing the deficit in concentration necessary to
balance the excess in the chilled regions of the ingot.

(3) Calculation of Maximum Segregation in Al-Cu System

The magnitude of segregation at any locality in the


ingot can beat be expressed as the concentration difference of
the segregating component in that locality to its mean concen-
tration in the ingot. Symbolically this is written as

c - (131

where the aE term corresponds to (1-~E) in Scheils calculations.


The numerator denotes the mass of solute in the volume under
consideration(sample volume) and the denominator the total mass
of the alloy contained in that volume. Since the last liquid
to freeze is of eutectic composition its contraction space is
filled with eutectic liquid. This amount must be corrected for
by applying the ~ppropriate shrinkage factor ~· There now
remains to calculate the terms mSE' mkE~kE'and mkE in the seg-
regation equation.
The expression required in the calculation is the dif-
ferentia! equation[lll Its evaluation reduces to a simple an-
- 32 -

alytic integration if the value of a is constant over the ran-


ge of integration. If the value of a is represented as a step-
~ ~ C (~+l)•
function of C8 for which a is constant for C81 C8 8
the solution of equation(ll]is the end result of a step-wise
integration over the intervals above as Cs varies from C0 to
CSE. To calculate the value of a we combine (3) and [11), thus

{141

Substitution [14] into [4] we get

[15]

The values for v8 ,vk, and Ck for different values of Cs for


the Al-Cu system are obtained from data collected by Sauerwald
and reproduced in the curves of Figure 4. In the Al-Cu system
C~-Ck can be approximated to a linear function of 9s;thus,
Cs-Ck:ACs where A is a constant. In Figure 5 the value of a
is plotted as a function of Cs• The stepped values of a indic-
ate over what range of concentration it was held constant.
Evaluatingrl~between the limita C81 4 C5 • Cs(J+l) we have

[16]

Solution of(3] gives

mk(~+l) = mkv+ msY-ms(y+l)


(17]
a,
.. 33 - ~1 Il

4-
1
1 •
1
1
~-
1
1 !

550 '
1 •
·TEM RATVfr
--+--~,- """'-j -.-, ·- - l f - - - -

1
1
C,._ D C.t.,
1
Vs ANbYit',
1
f

1
~l-
I

1
1
-1
'
- 35 -
The weight of solute in the cored crystals is evaluated from
the following relationship:

[18]

which has the solution

The maximum segregation for several values of copper


concentration for the Al-Cu system was calculated using the
values of a given in Fi·g ure 5 and was found to be in good agree-
ment with Scheil's calculations. The segregation curve for the
Al-Cu system calculated by Scheil is reproduced in Figure 7 in
a later section.

(4) Positional Variation of Solute Throughout the Ingot

Equation ~o1can be used to calculate the positional


variation of the segregating constituent,throughout the ingot
in the direction of solidification. The expressions for the mass
terms, as they appear in the segregation equation, have been
dertved for the chill face zone and top zone of the ingot where
positive and negative segregation exist (see Appendix 11).
Between these two zones, if we assume steady state conditions,
the segregation is zero.
Again to obtain the solution of equation (10] by a simple
- 36 -

analytic integration it was necessary to make severa! approx-


imatimns. The specifie volumes, vs,vk, and consequently a, are
assumed to be constant over the whole range of~'ntegration.

It is clear at this point that the above assumption can oily


lead to quantitative results if C0 , and therefore mgE, are
relatively small. The use of the same value of vk for both the
Cisolution and the entire eutectic solid is in general a quite
untenable proceedure since a new phase appears in the latter
case.
Near the chill face of the ingot and at the top, the
segregation was broken i&to two steps corresponding to steady
state conditions where msmin; 0 and to the case where msmin
is a varying function of m5 • The weighting factor W is a con-
stant (W:l) in all regions except in the last stages of solid-
ification near the top of the ingot. Here the slope of the den-
drite interface, dL/dms, goes to zero on final solidification.
The equations dertted in Appendix II have been applied
to the 4.70 per cent copper ingot to compute the positional
variation of copper. The steady state dimension of the mushy
zone, 1 0 , was taken as the distance from the chill face to
where segregation was substantially zero (determined experi-
mentally). The calculated curves are compared with the experi-
mental resulta ~n Figure 8.
- 37 -
EXPERIMENTAL

One of the assumptions made in formulating the theory


of inverse segregation was the complete flow-back of residual
liquid. In order that the experimental data could be compared
to theory with the least amount of uncertainty, it was
necessary to eleminate gas porosity as muchas possible. Uni-
directionally grown dendrites provide optimum conditions for
least restriction to flow-back of the residual melt. To obtain
this dendrite structure it was decided to solidify the ingots in
a manner similar to the one used by Adams.

A. MATERIALS

Super-purity aluminum, of 99.99 per cent Al, was


supplied for the base alloy by Aluminium Laboratories Ltd.,
Kingston, Ontario. The alloying constituent cooper was obtained
in the form of wire bar of commercial purity and of the follow-
ing analysis: 99.95 per cent Cu, 0.03 per cent 0, remainder
trace elements.

B. APPARATUS

Figure 6 is a schematic diagram of the solidification


apparatus. The mold material consisted of insulating brick,
H.W. 2600, bored to a cylindrical shape, with a 1.75 in. - dia.
base, and the walls tapered slightly so that an ingot of 2.25 in.
- 38 -

HOT BRICK TOP

v
,L.--/_/]/
1

F!Rf. .BRICK MOU) THERMOCOVPLE'S


4-" x 4-"x2 12 "~

~~
!NSERTE]) HERE

1
1
1 1
WATER 1
r , ~ - --- , j
1
,
])éFL CCTOR (
-- ---
\~~======;------Y
f8" STEEL BASE
'WATER PL RTE
NOZZLE.

\
BASE PLI/Tc SfJPPORT

' CONTROL VALVE

FIGURE: 6 .2>1A6RAM OF .SOL/])!FICATION APPARATUS


- 39 -

high had a diameter of 1.60 in. at the top. The object of this
design was to minimize any tendency of the ingot to hang up and
thus draw away from the base plate. A steel plate 0.12S in.
thick formed the base of the mold. Prior to the pour the mold
was preheated to a red heat in a temperature controlled electric
furnace. Rapid chilling was obtained by means of a jet of water
flowing at a controlled rate, and slow cooling by a stream of
nitrogen also flowing at a uniform and reproducible rate.
Temperatures in the ingot were measured during cooling
by placing thermocouples into the melt at positions 0.2Sin. and
2.00 in. up from the base as shown in Figure s. The thermocouple
wires were connected to Hoskin~ type A-H, chromel-alumel thermo-
couples with associated calibrated millivoltmeters. The tempera-
ture could be read to an accuracy of ~ soc.

C• EXPERIMENTAL PROCEDURE

(1) Me 1 tdown

After thoroughly cleaning the surface of the metal stock,


weighed amounts were placed in 30-gram fire-assay crucibles that
were coated on the inside with a lundum cement. The metals were
then melted down in a gas-fired fu~nace maintained at approxima-
tely l000°C. It was found necessary to coat the crucibles with
a lundum cement to pre vent the molte n a luminum from absor bing
silica from the crucible wall. This absorption was found to be
appreciable without the alundum coating.
- 40-

(2) Stirring

When the metals had liquified the melt was poured from
one crucible to another several times to ensure thorough
mixing of the components. Careful attention was required in
this step of the procedure for it was found that the melt
homogè.nized wi th difficu 1 ty.

(3) Degassing

Aluminum dissolves hydrogen very rapidly from the


moisture in the air. It is quite possible to have hydrogen up
to 17 per cent by volume evolve during solidification. About
90 per cent of the total gas is evolved before the solidus
temperature is reached, the remainder is released at the solidus
temperature. Any oxide film on the alloy melt provides a
protective film against gas absorption and should be kept intact
as much as possible.
To remove the absorbed hydrogen dry argon was passed
through the stirred melts. This method of degassing provides a
gas-metal interface not saturated with the impurity gas into
which the impurity gas is absorbed and carried to the surface.
A steel tube was placed into the melt through which the argom
flowed from a storage cylinder at a controlled rate. The degass-
ing was carried out at a fairly low melt temperature to prevent
any iron being dissolved by the molten alloy. The agitation by
the gas provided a further means of stirring.
- u -
A number of melts were degassed, solidified, and the
ingots examined to determine the procedure most effective in
degassing. It was found that by passing argon through the melt
until it completely solidified and then reheating to the desired
temperature in an electric furnace, where the melt was quiescent
and its protective oxide film remained undisturbed, gas absorption
was almost completely eliminated. The higher copper ingots showed
slight evidence of poros1ty indicating that higher copper alloys
absorb hydrogen more readily from the atmosphere during the final
pour.
Many ingots were cast from melts prepared by prolonged
stirring in carbon crucibles. These ingots, particularly the
higher copper ones, showed appreciable porosity. The fact that
this porosity was absent when using fire assay crucibles suggests
that possibly carbonaceous gases are absorbed by Al-Cu alloy
melts.

(4) Solidification

The melt, after degassing, was reheated to a temperature


of 150°C superheat in an electric furnace before pouring into the
mold. The superheat temperature was determined by inserting a
chromel-alumel thermocouple into the melt before the pour. The
fire-brick mold was also preheated to a red heat prior to the pour.
This was to ensure heat withdrawal through the base plate only
during solidification and thus assure unidirectional dendrite
growth. After the melt had reached the desired degree of superheat
- 42-
the brick mold was placed on the chill plate and the melt poured.
A preheated brick top was placed over the surface to prevent
any heat flow in this direction. Accurately known cooling rates
were not required. However, approximate cooling rates were
established in the 5 per cent copper and the 30 per cent copper
ingots by inserting thermocouples into the melt at 0.25 in. and
2.00 in. up from the base as shown in Figure 6. The thermo-
couples projected into the melt a distance just short of the
ingot center. The cooling rates for the other ingots were
estimated from ·the time required for complete solidification.
From mea~ured temperature gradients present in the liquid just
before chilling the presence of at least 40°0. superheat at the
chill face of the ingot was established. This melt superheat
at the chill face was necessary to ensure quiescent conditions
during initial nucleation and subsequent dendrite growth.

(5) Sampling

Considerable difficulty was encountered in obtaining a


representative sample of the alloy for the mean copper determina-
tion of the alloy. The main difficulty arose from the copper
segregation in the sample itself which, if not sufficiently
pulverized and thoroughly mixed, gave inconsistant resulta in
check analysis. This problem was finally circumvented by using
as the average ingot sample the swarf from a 0.5 in. thick
vertical section eut from the ingot for micro-analysis. The
thickness of the cutting blade was 0.05 in. Since the copper
- 43 -
segregation was in the vertical direction this method or sampling
was considered to give the most representative samples of the
ingot.
Samples of the chill face were lathe turnings from a
0.05 in. alice taken parallel to the base of the sample half of
the ingot. A minimum sample thickness of 0.05 in. was required
to provide sufficient material for analysis in the event more
than one check analysis would be required. Samples at various
levels of the ingot were obtained by collecting the swarf of
cuts parallel to the ingot base that included the complete half
circuler section.
One method of sampling, Which proved to be unrepresenta-
tive, demonstrated an interesting facet of the solidification
mechanism of liquid metals when in motion. In this trial
technique melt was drawn into a carbon tube 8 in. high with a
0.20 in.-dia. bore. The sample roda obtained contained a higher
copper content at the top than the bottom. The difference
averaged around 0.1 per cent copper. The molten alloy,
solidifying as it is drawn up into the cooler regions of the
tube, can be likened to melt drawn by capillary forces between
vertical dendriteesince the melt at the lower temperature also
becomes enriched in copper.

(6) Analysis

Several analytical methode for copper determination were


experimented with and it was found that the iodide method, ASTM
-~-

designation, E 34-42T, was the most accurate ror the conditions


or this investigation and was used except ror two minor changes.
The presence of only copper in the aluminum alloy allowed the
omission of the lengthy procedure designed to dissolve any
silicon in the sample. Furthermore the hydrogen sulphide wash
water was acidified with sulphuric rather than rormic acid.
Analysis on numerous samples of known copper content were
carried out until the laboratory technique was developed to give
the desired degree of accuracy. Two analyses on the sample at
the chill face and of C0 were carried out so as to provide a
check on the analytical work. In the very few cases that
analysis did not check within the estimated experimental erro~

the resulta were discarded and a new set of analyses completed


until agreement was obtained. Check analyses were not carried
out on samples used to determine the positional variation of
copper content since the final curves served as a check on
internal consistency.

(7) Estimated Accuracy

The two major sources or error are sample weighing and


titration. The titration comprises about 90 per cent of the
total error. The greater the amount of titrating solution used
the smaller the error becomes; however it was ~ound that if
the volume of thiosulphate solution used exceeded 25 mL the
color change at the end point became less and less definitive.
- 45 -
Accordingly~ samples were weighed so that the final solution
for titration required no more than 25 · m~ of thiosulphate
solution. At least two titrations for each sample were
obtained.
On the basis that the end point of the titration is
always either exact or exceeded only (the presence of the violet
color could be detected right up to the end point) by at most
one-half drop of the titrating solution, the error in an indivi-
dual copper determination would be ±0.022 percent. The probable
error in ~ calculated is approximately~0.031 per cent. Thus,
the total error in dC varies from ± 0.00 15 per cent for the
5 per cent copper ingot to ± 0.010 per cent for the 30 per cent
copper ingot. The calculation of errors are given in detail in
Appendix III.
It is not possible to estimate the sampling errors and errors
due to irreproducibility of casting conditions but they are probably
considerably less than the analytical errors.

D. RESULTS AND DISCUSSION

The experimental resulta are listed in Tables I, II, and


III. The segregation at the chill face is given in Table I and
the copper concentra tion a t va rious positions in ingots 2A and
3A are given in Tables II and III. These results are illustrated
and compared to theory in Figures 7 a nd 8.
The macros tructure of ingots 2A a nd 3A are shown i n
Figure 9. Long columnar grain growth is evident in both ingots
LOCATION COPPER ANALYSIS MAXIMUM SEGREGATION
distance ~ %
INGOT SAMPLE from chill SAMPLE EXPERIMENTAL THEORETICAL
face-in. 1 2 Ave. Actual Corr.

2A c. 4.70 4.70 4.70


Chill Face 0.025 5.12 5.12 5.12 0.42 0.44 0.43

3A CG 9.09 9.09 9.09


Chili Face 0.040 9.54 9.52 9.53 0.44 0.51 0.52

4A c0 . 14.70 14.70 14.70


1
Chi11 Face 0.025 15.20 15.18 15.19 0.49 0.51 0.47 ~
1

5A co 21.79 21.81 21.80


Chili Face 0.030 22.12 22.09 22.10 0.30 0.32 0.30

6A c. 29.05 29.05 29.05


Chi11 Face 0.035 29.13 29.13 29.13 o.os 0.10 0.11

3B Co 6;~ 61 8.63 8.62


(slow
cooled) Chill Face 0.025 9.13 9.13 9.13 0.51 0.53 0.52

TABLE I
CALCULATED AND EXPERIMENTAL VALU!O FOR MAXIMUM SEGREGATION IN Al - Cu INGOTS
(

- - ·~ -·-

LEGEND
SCH~tLS C URVE
0 EXPERIMENT-FASTCOOf-

"~
v
o .6 '"f • â ~XPe'RIM.éNT- SLOWCQOt..

Q
~
"'1!{

!:(
+ ----~· 0 :
1
~ ~ ~
j' :-.;;,.
i( ï
1 ~
, {!)
;
1
~ .1
·-_-,--~t- ~
' 1

, ~
- --· ----··- -----·-
!
'
-~·· · ---. --1--···--.
1 •
1
, (!)
1
t t·

____ ... _ _ _ ---1-- ·

.1 :0 : /S

.- 1 ·~ COPPêR PER CENT j

--+-·------r
1 • 1
-·--- - - __ L
F!Gl/RE 7
.·- ·-
SêGREGATION CVRV~ FOR AL-CU ALLOY AN.D ëXPERIMENTAL
1
VALUES
- 48-
TABLE II. Copper Content at Various Positions in Ingot 2A

LOCATION COPPER ANALYSIS


Distance from chi11
face in. %Cu

Chi11 face 5.14


0.025 5.12
0.065 4.77
0.35 4.71
0.62 4.76
0.90 4.75
1.22 4.71
1.57 4.72
1.90 4.67
2.25 4.35
Average 4.70

TABLE III. Copper Content at Various Positions in Ingot 3A

LOCATION COPPER ANALYSIS


Distance from chi11
face in. %Cu

Chi11 face 9.60


0.04 9.54
0.15 9.32
0.30 9.26
0.60 9.26
LOO 9-25
1.50 9.12
1.80 8.88
2.20 8.29
Average 9.10
- 49-

~CHILL FACE .'


TOP
'-.i
\

- \,~ cALCl/LATU>
rD<PER!MENTA.L
4 .7.
'
~ "" /r1EAIV CoPPéR

·- \
1
\
-

0
1
0·4-
,1
0-8
1
1·2 z..o 1

:DISTANCE FROM CHILL FACE INS


(a) IN60T 2A

~ CHILL FACE
.9.5 TOP----7-

~ 9-/
~0 +
u
1-..

v~ 8 ·7
ct
I.4J
0...

ô3
0 04- 0 ·8 1 ·z. 1·6 1·8

J:J/ST,4,NCE FROM CilILL FACE /NS


(b) INGor 3A
F'IG URE !!> COMPOSITION CiiRVES FOR INGOTS 2A A N.D 3A
- 50 -

(a) Ingot 2A

(b) Ingot JA

Figure 9. Macrostructure of Ingots


2A and JA Showing Long
Columnar Crystal Growth.
- 51 -
and this structure was characteristic throughout all ingots.
The chill face microstructure of ingots 4A and 3B are shown
in Figure 10. The long unidirectional dendrite pattern
observed in ingot 4A was present in all ingots. The direction
of dendrite growth became more random near the top of the
ingots where heat withdrawal became slow allowing large masses
of liquid to freeze at once. No exudation is present on the
chill face of ingot 4A. The eutectic on the chill face of ingot
3B is very thin and spotty and is continuous with the eutectic
in the interior of the ingot. No impoverished zone is apparent.
The chill face microstructures of ingots 2A and 3A are shown in
Figure 11. A thin film of eutectic exudation is evident on the
chill face of the ingots varying in thickness from 10 to 50
microns. A zone impoverished in eutectic is evident immediately
roove the exudation. Ingots 5A and 6A were free of exudations
at the chill face.
No correction for exudation was made on the analysis
of the chill face sample since the amount of exudation was
small relative to the sample thickness. Since the impoverished
zone is included in the sample thickness along with the
exudation, the analysis would represent the average copper
content of a point in the ingot a distance one-half the sample
thickness up from the ingot bottom. For most ingots the chill
face sample thickness was 0.05 in. Thus, the analyses would
give the average copper contents at positions 0.025 in. above
the ingot bottoms. The sample of ingot 3A was 0.08 in. thick.
.. 52 -

..... i
( ·~

(a) Ingot )B, Chill Face , x200.


Continuous eutectic at ohill
face . N i mpoverished zone.

(b) Ingot 4A , Chill Face , x200 ..


No euteotic present . Long
directional dendrite pattern.

Fi gure 10. Chill Face Microstructur e of Ingots 3B


and 4A .
.. 53 -

(a) Ingot 2A, Chill Face, x200.


Exudation and impoverished
zone present,

(b) Ingot JA, Chill Fac~ x340.


Exudation and impoverished
zone present .

Figure lJ,. . Chill Face Miorostructure of Ingots


2A and JA.
- 54 -
c
In T.able I, the segregation at 0.025 in. is corrected
to give the maximum segregation at the chill face. This was
done by rather arbitrarily adding 0.02 per cent to the
segregation at each point to correct the value to the chill
face. There is some doubt as to the magnitude of this
correction. It may be that in the last 0.01 to 0.02 in. out to
the chill face the copper content rise may be gradual or rapid
depending on the dendrite morphology right at the chill face
(distribution of cored crystals and entectic liquid). On the
model chosen for the theory, it must necessarily rise right to
the chill face. In any case the difference is not too significant.
For ingot 3A the composition curve was extrapolated to obtain
the maximum segregation at the chill face.

(l) Maximum Segregation

The maximum segregation values determined by experiment


compare favorably with the theory as illustrated by Figure 7.
Since Scheil's curve gives the maximum segregation possible due
to the contraction mechanism we would expect to find experimental
values on or below the curve. The values above the curve may be
attributed to analytical errors or lack of exudation corrections.
In any case the segregation exceeding the maximum as given by the
curve is small and within experimental errors. The slow cooled
9 per cent copper ingot shows no significant variation in segrega-
tion as a result of the cooling rate. This confirms Adams'
observations that maximum segregation is independant of cooling
rate provided the rate is fast enough. It is known that inverse
- 55 -
segregation in sorne alloys will revert to normal segregation
if the cooling rate is sufficiently decreased.

(2) Positional Variation

The positional variation across the ingot as calculated


agrees as well as might be expected with the experimental
curve of ingot 2A (Figure 8a) considering the much simplified
model used. The theoretical values were calculated using the
expressions derived in Appendix II. The length of the mushy
zone, Lo, for ingot 2A was taken as the distance from the chill
face to where the composition curve indicated substantially no
segregation. Referring to Figure 8a this length is 0.375 in.
The slight rise in copper content above the mean value at
distances past L9 from the chill face indicate that true steady
state conditions were not present during solidification. This is
more apparent in Figure 8b where the composition curve for the
9 per cent copper alloy indicate~ positive segregation across most
of the ingot. This is to be expected in the higher copper alloys
as it was shown by Ruddle (25) that by increasing the solute
content in an alloy, the width of the solidification zone increased
and continues to increase as solidification proceeds.

(3) Failure of Theory at Ingot Top

The theory breaks down in regions near the top of the


ingot as illustrated by the divergence of experimental and
- 56 -

theoretical segregation values. In the formulation of the


segregation theory an implicit assumption was made that the
liquid concentration Cs in the mushy zone varied from a maximum
of Cs·E at the solidus interface to C0 at the liquidus interface.
This assumption, while valid throughout most of the ingot, is
obviously incorrect near the top. Appreciable forward
diffusion of solute will occur out to the tips of the dendrites
as solidification of the mushy zone is being completed. Thus
the concentration at the top will be somewhat higher than that
of the first solid deposited from a liquid of Co concentration
as required by the theory. Furthermore, the theory breaks down
in this region as a result of an approximation made in the
evaluation of the integrais (Appendix II). This is the reason
for the failure of the mass balance evident in the theoretical
curve of Fig. 8a.
- 57 -

CONCLUSION

The close agreement of the experimental values for


maximum segregation in the Al-Cu ingots to those calculated
by Scheil on the basis of the contraction mechanism provides
conclusive evidence that the interdendritic flow theory is
sufficient in itself to explain the total observed segregation
values in Al-Cu ingots. Scheil's theory did not take into
consideration the dilution effect of the flow-back during the
solidification contraction and so was unable to explain the
positional variation of the solute throughout the ingot. The
extended theory of inverse segregation developed in conjunction
with this investigation satisfactorily explains the positional
variation of copper concentration observed in the low copper
ingots. Scheil's equation is a specia~ case of the general
theory developed. The close agreement of theory with experiment
is evidence that the dendrite model chosen realistically depicts
the general mass distribution in the solidification zone. The
approximation for a in the segregation calculations for the
chill face zone is valid provided the entectic mass is small.
For higher copper concentrations the resulta would be more
uncertain. The approximation used in the calculation of the
segregation curves near the top of the 1ngots (Appendix II),
where negative segregation exists is the least valid of those
used.in the calculations and this prevents an exact solute mass
balance over the ingot. However, these approximations do not
detract anything from the value of the theory since they arise
from analytical rather than conceptuel difficulties.
- 58 -

It appears that little more can be said about the


mechanism of inverse segregation itself. The question of
interest that still remain are '~t are the forces effecting
movement of the residual liquid", and "or what magnitude and
:elative importance are they". Fast efforts to identify this
force point to surface energy of the liquid as providing the
principal force, if not the only force required. Nevertheless
many more fruitful experimenta can be directed to this
particular end.
- 59 -

SUMMARY

The history of inverse segregation has been reviewed


with special attention to the more recent experimental work
ef Adams and Kastnsr. The present concept of the interdendri-
tic flow theory has been discussed in detail and a mathemati-
cal formulation of the mechanism for unidirectional solidif-
icatien has been developed in which the segregation is obtain-
as a functi~n of position in the ingot and of mean solute
concentration. It was shown that Scheil's derivations are a
special case of the more general theory in which the dilution
effect due to flow back on contraction is introduced.
The calculatiGns for maximum segregation far the Al-Cu
alloys and the positional variation of segregation in uni-
directionally solidified ingots compare well with the exper-
imental resulta in the Al-Cu alloy system.
It is concluded that a mechanism involving solidific-
atien contraction with flow back of enriched residual liquid
quantitatively predicts the observed segregation values in
Al-Cu ingots.
- 60 -

LIST OF REFERENCES

1. w. c. Roberts-Austen, Royal Mint Rep., Nos. 1,2,3,4,29,


30,31,32.
2. S. W. Smith, "Surface Tension and Cehesion in Metals and
Alloys", J. Inst. Metals, 1917, !:z, 65-103.
3. C. Benedicks, "Action of Hot Wall", Trans. Amer. Inst.
Mining Met. Eng., 1925, 11, 597~626.

4. M. Ballay, "Th~ Ludwig-Soret Etfect in Metallic Alloys",


Rev. Met., 1928, ~' 427-454, 509-520.
5. Iron and Steel Inst. Comm., "Repo~ on the Heterogeneity
ot Steel Ingots", J. Iron and Steel Inst., 1926, 113, 39-
176.
6. G. Masing, "The Explanation of Inverse Segregation", z.
Metal!., 1922, ~' 204-206.
7. R. Ge~ders, "The Mechanism of Inverse Segregation in Alloys",
J. Inst. Metals, 1927, lZ, 241-267.
8. R. Kuhnel,"Iaverse Segregation", z. Metall., 1922, .ll:,
462-464.
9. K, Iekibe, "On the Cause of Inverse Segregation", Sei.
Rep. Tohoku Imp. Univ., 1931, (i), lQ, 608-648.
10. G. Masing, "A Few Problems in Non-Ferrous Castings", Metals
and Alloys, 1933, ~, 99-104.
11. G. Masing and C. Haase, "On the Question of Inverse Seg-
regation", Wiss. Veroff. Siemens-Konzern, 1925, !, 113.
12. K. Iokibe, "On the Copper-Zinc Alloys which Expand on
Solidification", J. Inst. Metals, 1924, 1!, 225-256.
- 61 -

13. M. L. v. Gaylor, "A Study of the Relation between Macro-


and Micro-Structure ia Some Non-Ferrous Alloys", J. Inst.
of Metals, 1930, ~' 97-114.
14. D. E. Adams, "Segregation in A1uminum-Copper Alloys",
J. Inst. of Metals, 1948-49, 12, 809.
15. L. Nerthcott, "The Influence of Macrostructure Upon Seg-
regation", J. Inst. of Metals, 1946, 1&, 31-50.
16. J. H. Watson, "Liquation or Inverse Segregation in Silver-
Copper Alloys", J. Inst. of Metals, 1932, k2, 347-362.
17. N. P. Allen, "Further Observations on the Distribution
and Pososity in Aluminum and Copper Ingots, with Some
Netes on Inverse Segregation", J. Inst. of Metals, 1933,
,g, 193-220.
18. o. Bauer and H. Arndt, "Segregation Phenomena", z. Metal!.,
1921, !l, 497-506, 559-564.
19. H. Kastner, "Inverse Segregation in Continuous Casting",
z. Metall, 1950, ~' 193-205, 247-254. (Translation-British
Non-Ferrous Metals Research Association MP 431).
20. G. Masing and c. Haase, Wiss. Veroff. Siemens-Konzern,
1927' 2,, 211.
21. G. Masing andE. Scheuer, "Investigations on Segregation",
z. Metal!., 1933, ~' 173-179.
22. F. Sauerwald, Giesserei, 1942, 20, 25.
23. E. Scheil, z. Metal!., 1947, ~' 69, 1947, ~' 69.
24. F. Sauerwald, Metal!., 1943, ~' 543.
25. R. w. Ruddle, "A Preliminary Study of the Solidification
of Castings", J. Inst. of Metals, 1950, J:l, 1.
- 62 -

26. c. J. Smithells, Metals Reference Book, Vol. II,


Butterworth Scientific Publications, London 1955.
- 63 -

APPENDIX I

Calculation of Coring in the Solid and Liquid

In a dii'f'using
system having stationary
boundrj .aonditions (see
Figure 12) the distance
x1 , determined by the slope
at the origin of the con-
centration curve, bas the
Figure 12. Diffusion system following value x1:innt
with stationary boundry and where t is the time and D
fixed concentration C0 for the diffusion coeifficient
al1 time. ·( constant with time). From
Figure 12 it can be seen
that a quantity of this magnitude (we use x1ant ) can be
taken as an estimate of the mean distance an atom diffuses
intime t 1 •
The liquid and solid diffusion coefficients for cop-
per in aluminum have been listed by Smithells(26). To obtain
the liquid diffusion coefficients at the temperatures invol-
ved in the calcu1ation it was necessary to extrapolate the
data. At 600°C. the extrapolation gives a 1iquid diffusion
coefficient of 4.8 x 10-5 cm~/sec. The solid diffusion coef-
ficient for copper in a1uminum at 650°C. is about 3xlo- 8 cm~/sec.
- 64-

Ingot 3A--10% Cu, fast cooled

The time required for the mushy zone to pass a point


near the chill face is estimated from the cooling rate to be
35 seconds. Thus the mean penetration x1 : ~ :15 x lo-5 x 35
equals 0.04 cm. The average interdendritic spacing for the
ingot is 0.003 cm. The mean penetration of the cored solid
by the copper atoms is, ,calculating as above, · · 0.001 cm.
compared to an available penetration distance of 0.003 cm.

Ingot 3B--l~c-O,u, slow cooled

The time required for the mushy zone to pass a point:


near the chill face is estimated to be 100 seconds. Calculat-
ing x1 as above the mean penetration of the liquid is 0.07 cm.
compared to the interdendritic spacing of 0.006 cm. and the
solid penetration is 0.0017 cm. compared to an available pen-
etration distance of 0.006 cm.
From the above estimates it is apparent that the liq-
uid will not be::. appreciably cored during the initial •tages
of solidification. Coring in the solid will be apprstoiable
but not complete. The moving boundry conditions will modift
this crude diffusion picture somewhat but the qualitative
conclusions will not be altered.
- 65 -

APPENDIX II

Segregation Near the Chill Face of the Ingot

The evaluation
of~O) must be broken ia-

to two steps correspond-


(1)
ing to the case where msmin
is a varying function of
ms and a later stage where
msmin • o. This latter
stage involves steady state
growth since the dendrites
are fully formed. For
Figure 13. Dendrite morphology
case (1) subject to the
at the chill face.
assumption that v6 is con-
stant, equation [10) becomes, ·
since W:dm~/dm~ a 1

for m0)m 6)m0 .L where


ro
q • l - a
a

Solving we have

(la)
- 66 -

and

(2a)

Solving for c81 corresponding to m


81
= mo.L we get
~

::-: .. A
mo-tq -
mo.t+ q
]a (Ja)
t;

For stage (2) m8 min ; 0, the differentia! equation is

with the solution

(4a)

and

mkE : mkl + ms1 - mgE (5a)


a

To calculate the solute in the cored crystals we have

For region (1)


- 67 -

Hence
-a
• l]!(mo+q)(~) I dCs
Thus
("'Jc"lrk l1.. }:! (Dlo+q)c{(~:~At - 1] (6a)

For region ( 2)

Hence

with the solution

(7a)

The · integration
must be done in two stages
corresponding to cases
.
(1)
' (1) and (2) where respect-
ively ms takes the values
m0 )m8 ,m0 (L-L0 -r) and
.,
r:0
(2) m0 ( L-L0 -r) > ms ) 0. In
Lo
Figure 14. Dendrite morphology
at ingot top. region (1) the differen-
- 68 -

tial equation is that ~or steady. state growth, i.e.

-dC
~
The solution is

{lb)

The value of Cs correspon ding to m61 : m0 (L+L0 -r) is


t;

(2b)

also

(Jb)

In region {2) W(mé) : dmg/dm~ =(IDo-m~-ms) so that

which simpli~ies to

dm8 : -kdC 8 (4b)


AC 8

where

k = ~1 (5b)
- 69 -

In this calculation we will average the variable part


th us

ms
--
1 -
2m0 1- ~1
l 1 + 2mo
1 - ms msl
mo 1 -
mo

Solving equation (4b) we get for C8 :CsE

{6b)

(7b)

To evaluate the solute in the ·cored crystals we have

In section (1)

Hence

Thus

(Sb)

In section (2)
- 70 -

Thus

Solving

--·
- 71 -

APPENDIX III

Calculation of Analytical Errors

One ml. division on the burrette is equivalent to 50


drops of titrating solution. Thus, one-half drop equals 0.01
ml. Approximately 25 ml. of titrating solution was used for
each analysis and for the standardization runs. The titration
is assumed to be exact or exceeded by a probable one-half
drop which gives the probable error in the titration as
+ 0.04%.
All samples taken were weighed to an accuracy of ± o.Ol%.
Two grams of copper were weighed for preparation_of the cop-
per sulphate solution. with a weighing error of ~.0.0025% that
is negligible in comparison with the other errors.
No error resulta when measuring out the copper sul-
phate BQlutioa for standardization as the volume measured is
from division mark to division mark. The last fraction of a
drop was removed. from the burrette tip by bringing the tit-
rating flask in contact with it.
The total probable error in an individual determin-
ation, calculated as the root mean square of the sum or -the
individual errors, is approx.l mately f 0.041%. In calculating
âC the standardization errors cancel and it is possible to
replace the positive titration error of + 0.04% by % 0.02%
without altering the final result. The total probable error
inAC, calculated again as the root mean square of the sum of
- 72 -

the individu al errors, is ± 0.031%. For the 5% copper alloy


'
this would give a total probable error inAC off 0.0015%
and ± 0.01% in the 30% copper alloy.

You might also like